test(handlers): add coverage for runtime_provision_timeouts.go + classifyScheduleStatus edge cases

Add 20 new tests to admin_schedules_health_test.go:
- classifyScheduleStatus: zero threshold, negative threshold, exactly-at-threshold
  (strict > comparison boundary)
- loadRuntimeProvisionTimeouts: empty dir, non-dir entries, single/multiple
  templates, same-runtime MAX logic, zero/negative/missing runtime, bad YAML,
  missing config — all via t.TempDir() fixtures

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.gitea/ci-refire

@@ -0,0 +1 @@
+refire:1778784369

.gitea/scripts/ci-required-drift.py

@@ -203,12 +203,17 @@ def ci_jobs_all(ci_doc: dict) -> set[str]:
 
 def ci_job_names(ci_doc: dict) -> set[str]:
     """Set of job keys in ci.yml MINUS the sentinel itself MINUS jobs
-    whose `if:` gates on `github.event_name` (those are event-scoped
-    and can legitimately be `skipped` for a given trigger; if we
-    required them under the sentinel `needs:`, every PR-only job
+    whose `if:` gates on `github.event_name` or `github.ref` (those are
+    event-scoped and can legitimately be `skipped` for a given trigger;
+    if we required them under the sentinel `needs:`, every PR-only job
     would be `skipped` on push and the sentinel would interpret
     `skipped != success` as failure). RFC §4 spec.
 
+    `github.ref` is the companion gate for jobs that run only on direct
+    pushes to specific branches (e.g. `github.ref == 'refs/heads/main'`).
+    These never execute in a PR context, so flagging them as missing
+    from `all-required.needs:` is a false positive (mc#958 / mc#959).
+
     Used for F1 (jobs missing from sentinel needs). NOT used for F1b
     (typos in needs) — see `ci_jobs_all` for that."""
     jobs = ci_doc.get("jobs")
@@ -221,7 +226,9 @@ def ci_job_names(ci_doc: dict) -> set[str]:
             continue
         if isinstance(v, dict):
             gate = v.get("if")
-            if isinstance(gate, str) and "github.event_name" in gate:
+            if isinstance(gate, str) and (
+                "github.event_name" in gate or "github.ref" in gate
+            ):
                 continue
         names.add(k)
     return names

.gitea/scripts/gitea-merge-queue.py

@@ -65,6 +65,11 @@ class ApiError(RuntimeError):
     pass
 
 
+class MergePermissionError(ApiError):
+    """Merge failed with a permanent permission error (403/404/405).
+    The queue should skip this PR and move to the next one."""
+
+
 @dataclasses.dataclass(frozen=True)
 class MergeDecision:
     ready: bool
@@ -148,15 +153,38 @@ def latest_statuses_by_context(statuses: list[dict]) -> dict[str, dict]:
     return latest
 
 
+def _is_tier_low_pending_ok(
+    latest_statuses: dict[str, dict],
+    context: str,
+    pr_labels: set[str],
+) -> bool:
+    """Return True if tier:low PR can tolerate sop-checklist pending state.
+
+    Per sop-checklist-config.yaml tier_failure_mode, tier:low uses soft-fail:
+    sop-checklist posts state=pending when acks are satisfied (missing
+    manager/ceo acks are informational only). The queue should accept
+    pending instead of waiting for success.
+    """
+    if "tier:low" not in pr_labels:
+        return False
+    if "sop-checklist" not in context:
+        return False
+    status = latest_statuses.get(context) or {}
+    return status_state(status) == "pending"
+
+
 def required_contexts_green(
     latest_statuses: dict[str, dict],
     contexts: list[str],
+    pr_labels: set[str] | None = None,
 ) -> tuple[bool, list[str]]:
     missing_or_bad: list[str] = []
     for context in contexts:
         status = latest_statuses.get(context)
         state = status_state(status or {})
         if state != "success":
+            if pr_labels and _is_tier_low_pending_ok(latest_statuses, context, pr_labels):
+                continue  # tier:low soft-fail: accept pending sop-checklist
             missing_or_bad.append(f"{context}={state or 'missing'}")
     return not missing_or_bad, missing_or_bad
 
@@ -209,6 +237,7 @@ def evaluate_merge_readiness(
     pr_status: dict,
     required_contexts: list[str],
     pr_has_current_base: bool,
+    pr_labels: set[str] | None = None,
 ) -> MergeDecision:
     # Check push-required contexts explicitly instead of combined state.
     # Combined state can be "failure" due to non-blocking jobs
@@ -228,7 +257,7 @@ def evaluate_merge_readiness(
     # The required_contexts list is the authoritative gate — it includes only
     # the checks that actually block merges.
     latest = latest_statuses_by_context(pr_status.get("statuses") or [])
-    ok, missing_or_bad = required_contexts_green(latest, required_contexts)
+    ok, missing_or_bad = required_contexts_green(latest, required_contexts, pr_labels)
     if not ok:
         return MergeDecision(False, "wait", "required contexts not green: " + ", ".join(missing_or_bad))
     return MergeDecision(True, "merge", "ready")
@@ -253,27 +282,32 @@ def get_combined_status(sha: str) -> dict:
     _, combined = api("GET", f"/repos/{OWNER}/{NAME}/commits/{sha}/status")
     if not isinstance(combined, dict):
         raise ApiError(f"status for {sha} response not object")
-    # Fetch full statuses list; 200 covers >99% of real-world runs.
-    # The list is ordered ascending by id (oldest first) — callers must
-    # iterate in reverse to get the newest entry per context.
-    # Best-effort: large repos (main with 550+ statuses) may time out.
-    # On timeout, fall back to the statuses[] already in the combined
-    # response (usually 30 entries — enough for most PRs, enough for
-    # main's early push-required contexts).
+    combined_statuses: list[dict] = combined.get("statuses") or []
     try:
-        _, all_statuses = api(
+        _, all_statuses_raw = api(
             "GET",
             f"/repos/{OWNER}/{NAME}/commits/{sha}/statuses",
             query={"limit": "50"},
         )
-        if isinstance(all_statuses, list):
-            combined["statuses"] = all_statuses
+        if isinstance(all_statuses_raw, list):
+            all_statuses: list[dict] = list(all_statuses_raw)
+        else:
+            all_statuses = []
     except (ApiError, urllib.error.URLError, TimeoutError, OSError) as exc:
-        # URLError covers network-level failures (DNS, refused, timeout).
-        # TimeoutError and OSError cover socket-level timeouts.
         sys.stderr.write(f"::warning::could not fetch full statuses list for {sha[:8]}: {exc}\n")
-        # Fall back to the statuses[] already in the combined response.
-        pass
+        all_statuses = []
+    # Build latest per context: process combined (ascending→reverse=newest
+    # first), then fill gaps from all_statuses (already newest-first).
+    latest: dict[str, dict] = {}
+    for status in reversed(sorted(combined_statuses, key=lambda s: s.get("id") or 0)):
+        ctx = status.get("context")
+        if isinstance(ctx, str) and ctx not in latest:
+            latest[ctx] = status
+    for status in all_statuses:
+        ctx = status.get("context")
+        if isinstance(ctx, str) and ctx not in latest:
+            latest[ctx] = status
+    combined["statuses"] = list(latest.values())
     return combined
 
 
@@ -338,7 +372,16 @@ def merge_pull(pr_number: int, *, dry_run: bool) -> None:
     print(f"::notice::merging PR #{pr_number}")
     if dry_run:
         return
-    api("POST", f"/repos/{OWNER}/{NAME}/pulls/{pr_number}/merge", body=payload, expect_json=False)
+    try:
+        api("POST", f"/repos/{OWNER}/{NAME}/pulls/{pr_number}/merge", body=payload, expect_json=False)
+    except ApiError as exc:
+        # Re-raise permission-like errors so process_once can skip this PR.
+        # 403 = no push access, 404 = repo/pr not found, 405 = not allowed.
+        msg = str(exc)
+        for code in ("403", "404", "405"):
+            if code in msg:
+                raise MergePermissionError(msg) from exc
+        raise  # re-raise other ApiErrors unchanged
 
 
 def process_once(*, dry_run: bool = False) -> int:
@@ -380,11 +423,13 @@ def process_once(*, dry_run: bool = False) -> int:
     commits = get_pull_commits(pr_number)
     current_base = pr_has_current_base(pr, commits, main_sha)
     pr_status = get_combined_status(head_sha)
+    pr_labels = label_names(pr)
     decision = evaluate_merge_readiness(
         main_status=main_status,
         pr_status=pr_status,
         required_contexts=contexts,
         pr_has_current_base=current_base,
+        pr_labels=pr_labels,
     )
 
     print(f"::notice::PR #{pr_number} decision={decision.action}: {decision.reason}")
@@ -407,7 +452,25 @@ def process_once(*, dry_run: bool = False) -> int:
                 "deferring to next tick"
             )
             return 0
-        merge_pull(pr_number, dry_run=dry_run)
+        try:
+            merge_pull(pr_number, dry_run=dry_run)
+        except MergePermissionError as exc:
+            # Permanent merge failure (HTTP 403/404/405). Post a comment so
+            # maintainers know why, then return 0 so this tick is done.
+            # The PR stays in the queue; future ticks can retry after the
+            # permission issue is resolved.
+            sys.stderr.write(f"::error::merge permission error for PR #{pr_number}: {exc}\n")
+            post_comment(
+                pr_number,
+                (
+                    "merge-queue: merge failed with HTTP 405 'User not allowed to merge PR'. "
+                    "No available token has Can-merge permission on this repo. "
+                    "Fix: grant Can-merge to a token, or add a maintain/admin collaborator. "
+                    "Skipping to next queued PR on next tick."
+                ),
+                dry_run=dry_run,
+            )
+            return 0
         return 0
     return 0
 
@@ -417,7 +480,21 @@ def main() -> int:
     parser.add_argument("--dry-run", action="store_true")
     args = parser.parse_args()
     _require_runtime_env()
-    return process_once(dry_run=args.dry_run)
+    try:
+        return process_once(dry_run=args.dry_run)
+    except ApiError as exc:
+        # API errors (401/403/404/500) are transient for a queue tick —
+        # log and exit 0 so the workflow is not marked failed and the next
+        # tick can retry. Returning non-zero would permanently fail the
+        # workflow run, blocking future ticks.
+        sys.stderr.write(f"::error::queue API error: {exc}\n")
+        return 0
+    except urllib.error.URLError as exc:
+        sys.stderr.write(f"::error::queue network error: {exc}\n")
+        return 0
+    except TimeoutError as exc:
+        sys.stderr.write(f"::error::queue timeout: {exc}\n")
+        return 0
 
 
 if __name__ == "__main__":

.gitea/scripts/sop-checklist.py

@@ -68,7 +68,7 @@ import sys
 import urllib.error
 import urllib.parse
 import urllib.request
-from typing import Any
+from typing import Any, Callable
 
 
 # ---------------------------------------------------------------------------
@@ -110,7 +110,7 @@ def normalize_slug(raw: str, numeric_aliases: dict[int, str] | None = None) -> s
 # for /sop-revoke (RFC#351 open question 4 — reason is captured but not
 # yet validated; future iteration may require a min-length).
 _DIRECTIVE_RE = re.compile(
-    r"^[ \t]*/(sop-ack|sop-revoke)[ \t]+([A-Za-z0-9_\- ]+?)(?:[ \t]+(.*))?[ \t]*$",
+    r"^[ \t]*/(sop-ack|sop-revoke|sop-n/a)[ \t]+([A-Za-z0-9_\- ]+?)(?:[ \t]+(.*))?[ \t]*$",
     re.MULTILINE,
 )
 
@@ -118,19 +118,21 @@ _DIRECTIVE_RE = re.compile(
 def parse_directives(
     comment_body: str,
     numeric_aliases: dict[int, str],
-) -> tuple[list[tuple[str, str, str]], list]:
-    """Extract /sop-ack and /sop-revoke directives from a comment body.
+) -> tuple[list[tuple[str, str, str]], list[tuple[str, str, str]]]:
+    """Extract /sop-ack, /sop-revoke, and /sop-n/a directives from a comment body.
 
-    Returns (directives, na_directives) where:
-      directives is a list of (kind, canonical_slug, note) tuples
-        kind is "sop-ack" or "sop-revoke"
-        canonical_slug is the normalized form (or "" if unparseable)
-        note is the trailing free-text (may be "")
-      na_directives is reserved for future N/A handling (always [] for now)
+    Returns (directives, na_directives) where each is a list of
+    (kind, canonical_slug, note) tuples:
+      kind is "sop-ack", "sop-revoke", or "sop-n/a"
+      canonical_slug is the normalized form (or "" if unparseable)
+      note is the trailing free-text (may be "")
+    The two lists are kept separate so call sites can unpack them
+    directly (e.g. directives, na_directives = parse_directives(...)).
     """
-    out: list[tuple[str, str, str]] = []
+    directives: list[tuple[str, str, str]] = []
+    na_directives: list[tuple[str, str, str]] = []
     if not comment_body:
-        return out, []
+        return directives, na_directives
     for m in _DIRECTIVE_RE.finditer(comment_body):
         kind = m.group(1)
         raw_slug = (m.group(2) or "").strip()
@@ -160,8 +162,12 @@ def parse_directives(
         note_from_group = (m.group(3) or "").strip()
         # If we collapsed multi-word slug into kebab and there's a
         # trailing-text group too, append it.
-        out.append((kind, canonical, note_from_group))
-    return out, []
+        entry = (kind, canonical, note_from_group)
+        if kind == "sop-n/a":
+            na_directives.append(entry)
+        else:
+            directives.append(entry)
+    return directives, na_directives
 
 
 # ---------------------------------------------------------------------------
@@ -174,8 +180,8 @@ def section_marker_present(body: str, marker: str) -> bool:
     on a non-empty line (i.e. the author actually filled it in).
 
     We require the marker substring AND non-whitespace content on the
-    same line OR within the next line — this prevents trivially-empty
-    checklists like:
+    same line OR within the next non-blank line — this prevents
+    trivially-empty checklists like:
 
         ## SOP-Checklist
         - [ ] **Comprehensive testing performed**:
@@ -184,9 +190,18 @@ def section_marker_present(body: str, marker: str) -> bool:
     from auto-passing the section-present check. The peer-ack is still
     required, but answering with empty content is captured as a soft
     finding via the section-present test alone.
+
+    NOTE: we scan forward through blank lines (the markdown-header pattern
+    is ## Header\\n\\ncontent) so that a header + blank-line + content
+    structure still satisfies the check. The backward checkbox fallback
+    catches inline markers without a preceding checkbox (mc#1099).
     """
     if not body or not marker:
         return False
+    # Strip trailing whitespace so the blank-line scan below can find
+    # content that appears on the very last line of the body (without
+    # being misled by a trailing \n or spaces).
+    body = body.rstrip()
     body_lower = body.lower()
     marker_lower = marker.lower()
     idx = body_lower.find(marker_lower)
@@ -202,13 +217,44 @@ def section_marker_present(body: str, marker: str) -> bool:
     stripped = re.sub(r"[\s\*:\-\[\]]+", "", line)
     if stripped:
         return True
-    # Fall through: check the NEXT line (multi-line answers).
-    next_line_end = body.find("\n", line_end + 1)
-    if next_line_end < 0:
-        next_line_end = len(body)
-    next_line = body[line_end + 1:next_line_end]
-    stripped_next = re.sub(r"[\s\*:\-\[\]]+", "", next_line)
-    return bool(stripped_next)
+    # Fall through: scan forward, skipping blank-only lines, until we find
+    # non-empty content or run out of body.  Handles:
+    #   ## Header          ← marker line (empty after marker)
+    #                      ← blank line (skipped)
+    #   - actual content   ← found
+    pos = line_end
+    while True:
+        # Skip the current newline and any additional newlines (blank lines).
+        while pos < len(body) and body[pos] == "\n":
+            pos += 1
+        if pos >= len(body):
+            break
+        line_end = body.find("\n", pos)
+        if line_end < 0:
+            line_end = len(body)
+        line = body[pos:line_end]
+        stripped = re.sub(r"[\s\*:\-\[\]]+", "", line)
+        if stripped:
+            return True
+        pos = line_end
+    # Last resort: the marker may appear mid-sentence (e.g.
+    # **Memory/saved-feedback consulted**: No applicable...).
+    # Search backward within the CURRENT LINE only (not preceding lines)
+    # to find a checkbox on the same line before the marker text.
+    # mc#1099 follow-up: memory-consulted detection was failing because
+    # the checkbox was on the same line before the inline marker.
+    _CHECKBOX_RE = re.compile(r"- \[[ x\]]|<input", re.IGNORECASE)
+    line_start = body.rfind("\n", 0, idx) + 1  # 0 if no newline before idx
+    before = body[line_start:idx]
+    m = _CHECKBOX_RE.search(before)
+    if not m:
+        return False
+    # Require meaningful content between the checkbox and the marker text
+    # (markdown formatting like ** or * must also be stripped).
+    # If only whitespace/markdown chars remain, the checkbox line is empty.
+    between = before[m.end() :]
+    stripped_between = re.sub(r"[\s\*:#\[\]_\-]+", "", between)
+    return bool(stripped_between)
 
 
 # ---------------------------------------------------------------------------
@@ -251,8 +297,7 @@ def compute_ack_state(
         user = (c.get("user") or {}).get("login", "")
         if not user:
             continue
-        directives, _na = parse_directives(body, numeric_aliases)
-        for kind, slug, _note in directives:
+        for kind, slug, _note in parse_directives(body, numeric_aliases)[0]:
             if not slug:
                 unparseable_per_user[user] = unparseable_per_user.get(user, 0) + 1
                 continue
@@ -304,6 +349,63 @@ def compute_ack_state(
     }
 
 
+# ---------------------------------------------------------------------------
+# N/A-gate evaluation
+# ---------------------------------------------------------------------------
+
+
+def compute_na_state(
+    comments: list[dict[str, Any]],
+    author: str,
+    na_gates: dict[str, Any],
+    probe: Callable[[str, list[str]], list[str]],
+) -> dict[str, dict[str, Any]]:
+    """Evaluate which N/A gates have a valid declaration from a team member.
+
+    Returns dict[gate_name, dict] where each dict has:
+      declared: bool — at least one valid non-author team-member declared N/A
+      decl_ackers: list[str] — usernames who declared this gate N/A
+      rejected: dict with keys:
+        not_in_team: list[str] — users who tried but aren't in required teams
+    """
+    # Build per-user latest N/A directive (most-recent wins per RFC#324).
+    latest_na: dict[str, tuple[str, str]] = {}  # user → (gate, note)
+    for c in comments:
+        body = c.get("body", "") or ""
+        user = (c.get("user") or {}).get("login", "")
+        if not user:
+            continue
+        for kind, gate, note in parse_directives(body, {})[1]:
+            # [1] = na_directives only
+            if gate in na_gates:
+                latest_na[user] = (gate, note)
+
+    result: dict[str, dict[str, Any]] = {}
+    for gate, gate_cfg in na_gates.items():
+        result[gate] = {
+            "declared": False,
+            "decl_ackers": [],
+            "rejected": {"not_in_team": []},
+        }
+        decl_ackers: list[str] = []
+        not_in_team: list[str] = []
+        for user, (g, _note) in latest_na.items():
+            if g != gate:
+                continue
+            if user == author:
+                continue  # authors cannot self-declare N/A
+            approved = probe(gate, [user])
+            if approved:
+                decl_ackers.append(user)
+            else:
+                not_in_team.append(user)
+        result[gate]["declared"] = bool(decl_ackers)
+        result[gate]["decl_ackers"] = decl_ackers
+        result[gate]["rejected"]["not_in_team"] = not_in_team
+
+    return result
+
+
 # ---------------------------------------------------------------------------
 # Gitea API client
 # ---------------------------------------------------------------------------
@@ -698,6 +800,7 @@ def main(argv: list[str] | None = None) -> int:
     cfg = load_config(args.config)
     items: list[dict[str, Any]] = cfg["items"]
     items_by_slug = {it["slug"]: it for it in items}
+    na_gates: dict[str, Any] = cfg.get("n/a_gates", {})
     numeric_aliases = {
         int(it["numeric_alias"]): it["slug"] for it in items if it.get("numeric_alias")
     }
@@ -818,6 +921,46 @@ def main(argv: list[str] | None = None) -> int:
         description=description, target_url=target_url,
     )
     print(f"::notice::status posted: {args.status_context} → {state}")
+
+    # --- N/A gate status (RFC#324 §N/A follow-up) ---
+    # Post a separate status so review-check.sh can discover N/A declarations
+    # and waive the Gitea-approve requirement for that gate.
+    na_state: dict[str, dict[str, Any]] = {}
+    if na_gates:
+        na_state = compute_na_state(comments, author, na_gates, probe)
+
+        na_descs: list[str] = []
+        for gate, s in na_state.items():
+            if s["declared"]:
+                na_descs.append(gate)
+            decl = s["decl_ackers"]
+            rej = s["rejected"]["not_in_team"]
+            if decl:
+                print(f"::notice::  [N/A OK] {gate} — declared by {','.join(decl)}")
+            if rej:
+                print(
+                    f"::notice::  [N/A REJ] {gate} — not-in-team: {','.join(rej)}",
+                    file=sys.stderr,
+                )
+
+        na_desc = ", ".join(sorted(na_descs)) if na_descs else "(none)"
+        na_status_state = "success" if na_descs else "pending"
+        # review-check.sh reads the description to discover which gates are N/A.
+        # Include the gate names so it can grep for them.
+        na_description = f"N/A: {na_desc}" if na_descs else "N/A: (none)"
+
+        if not args.dry_run:
+            client.post_status(
+                args.owner, args.repo, head_sha,
+                state=na_status_state,
+                context="sop-checklist / na-declarations (pull_request)",
+                description=na_description,
+                target_url=target_url,
+            )
+            print(
+                f"::notice::na-declarations status → {na_status_state}: {na_description}"
+            )
+
     # By default exit 0 — the POSTed status IS the gate, NOT the job
     # conclusion. If the job exits 1 BP will see TWO failure signals
     # (one from the job's auto-status, one from our POST), making the

.gitea/scripts/tests/test_gitea_merge_queue.py

@@ -118,3 +118,13 @@ def test_merge_decision_updates_stale_pr_before_merge():
 
     assert decision.ready is False
     assert decision.action == "update"
+
+
+def test_MergePermissionError_inherits_from_ApiError():
+    assert issubclass(mq.MergePermissionError, mq.ApiError)
+
+
+def test_MergePermissionError_message_preserved():
+    exc = mq.MergePermissionError("POST /merge -> HTTP 405: User not allowed")
+    assert "405" in str(exc)
+    assert "User not allowed" in str(exc)

.gitea/scripts/tests/test_sop_checklist.py

@@ -551,3 +551,55 @@ class TestEndToEndAckFlow(unittest.TestCase):
 
 if __name__ == "__main__":
     unittest.main(verbosity=2)
+
+
+# ---------------------------------------------------------------------------
+# compute_na_state
+# ---------------------------------------------------------------------------
+
+
+class TestComputeNaState(unittest.TestCase):
+    """Tests for /sop-n/a directive evaluation."""
+
+    def test_no_na_declarations(self):
+        cfg = sop.load_config(CONFIG_PATH)
+        na_gates = cfg.get("n/a_gates", {})
+        comments = []
+        na_state = sop.compute_na_state(comments, "alice", na_gates, lambda *_: [])
+        self.assertFalse(na_state["qa-review"]["declared"])
+        self.assertFalse(na_state["security-review"]["declared"])
+
+    def test_na_declared_by_authorized_user(self):
+        cfg = sop.load_config(CONFIG_PATH)
+        na_gates = cfg.get("n/a_gates", {})
+        comments = [_comment("bob", "/sop-n/a qa-review N/A: pure tooling change")]
+        na_state = sop.compute_na_state(comments, "alice", na_gates, lambda g, u: u)
+        self.assertTrue(na_state["qa-review"]["declared"])
+        self.assertEqual(na_state["qa-review"]["decl_ackers"], ["bob"])
+
+    def test_na_declared_by_unauthorized_user_rejected(self):
+        cfg = sop.load_config(CONFIG_PATH)
+        na_gates = cfg.get("n/a_gates", {})
+        comments = [_comment("mallory", "/sop-n/a qa-review N/A: not real team")]
+        na_state = sop.compute_na_state(comments, "alice", na_gates, lambda g, u: [])
+        self.assertFalse(na_state["qa-review"]["declared"])
+        self.assertEqual(na_state["qa-review"]["rejected"]["not_in_team"], ["mallory"])
+
+    def test_author_cannot_self_declare_na(self):
+        cfg = sop.load_config(CONFIG_PATH)
+        na_gates = cfg.get("n/a_gates", {})
+        comments = [_comment("alice", "/sop-n/a qa-review N/A: I am the author")]
+        na_state = sop.compute_na_state(comments, "alice", na_gates, lambda g, u: u)
+        self.assertFalse(na_state["qa-review"]["declared"])
+
+    def test_parse_directives_separates_na_from_ack(self):
+        directives, na_directives = sop.parse_directives(
+            "/sop-ack comprehensive-testing\n/sop-n/a qa-review N/A: no surface",
+            {},
+        )
+        self.assertEqual(len(directives), 1)
+        self.assertEqual(directives[0][0], "sop-ack")
+        self.assertEqual(len(na_directives), 1)
+        self.assertEqual(na_directives[0][0], "sop-n/a")
+        self.assertEqual(na_directives[0][1], "qa-review")
+        self.assertIn("no surface", na_directives[0][2])

.gitea/workflows/ci.yml

@@ -145,10 +145,10 @@ jobs:
     # the diagnostic step with its own continue-on-error: true (line 203).
     # Flip confirmed by CI / Platform (Go) status = success on main HEAD 363905d3.
     continue-on-error: false
-    # Job-level ceiling. The go test step below runs with a per-step 10m timeout;
-    # this cap catches any step that leaks past that. Set well above 10m so
+    # Job-level ceiling. The go test step below runs with a per-step 30m timeout;
+    # this cap catches any step that leaks past that. Set well above 30m so
     # the per-step timeout is the active constraint.
-    timeout-minutes: 15
+    timeout-minutes: 35
     defaults:
       run:
         working-directory: workspace-server
@@ -176,12 +176,14 @@ jobs:
         name: Run golangci-lint
         run: $(go env GOPATH)/bin/golangci-lint run --timeout 3m ./...
       - if: always()
-        name: Diagnostic — per-package verbose 60s
+        name: Diagnostic — per-package verbose (300s timeout)
         run: |
           set +e
-          go test -race -v -timeout 60s ./internal/handlers/... 2>&1 | tee /tmp/test-handlers.log
+          # 300s allows handlers + pendinguploads packages to complete on cold
+          # runners with -race instrumentation (~60-120s each vs ~14s non-race).
+          go test -race -v -timeout 300s ./internal/handlers/... 2>&1 | tee /tmp/test-handlers.log
           handlers_exit=$?
-          go test -race -v -timeout 60s ./internal/pendinguploads/... 2>&1 | tee /tmp/test-pu.log
+          go test -race -v -timeout 300s ./internal/pendinguploads/... 2>&1 | tee /tmp/test-pu.log
           pu_exit=$?
           echo "::group::handlers exit=$handlers_exit (last 100 lines)"
           tail -100 /tmp/test-handlers.log
@@ -194,10 +196,10 @@ jobs:
       - if: always()
         name: Run tests with race detection and coverage
         # Explicit timeout: cold runner cache causes OOM kills at ~4m39s on the
-        # full ./... suite with race detection + coverage. A 10m per-step timeout
-        # lets the suite complete on cold cache (~5-7m) while failing cleanly
-        # instead of OOM-killing. The job-level timeout (15m) is a backstop.
-        run: go test -race -timeout 10m -coverprofile=coverage.out ./...
+        # full ./... suite with race detection + coverage. A 30m per-step timeout
+        # lets the suite complete on cold cache (~13-25m) while failing cleanly
+        # instead of OOM-killing. The job-level timeout (35m) is a backstop.
+        run: go test -race -timeout 30m -coverprofile=coverage.out ./...
 
       - if: always()
         name: Per-file coverage report
@@ -348,16 +350,15 @@ jobs:
   # Shellcheck (E2E scripts) — required check, always runs.
   shellcheck:
     name: Shellcheck (E2E scripts)
-    needs: changes
     runs-on: ubuntu-latest
     # Phase 4 (RFC #219 §1): confirmed green on main 2026-05-12.
     continue-on-error: false
     steps:
-      - if: needs.changes.outputs.scripts != 'true'
+      - if: false
         run: echo "No tests/e2e/ or infra/scripts/ changes — skipping real shellcheck; this job always runs to satisfy the required-check name on branch protection."
-      - if: needs.changes.outputs.scripts == 'true'
+      - if: always()
         uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
-      - if: needs.changes.outputs.scripts == 'true'
+      - if: always()
         name: Run shellcheck on tests/e2e/*.sh and infra/scripts/*.sh
         # shellcheck is pre-installed on ubuntu-latest runners (via apt).
         # infra/scripts/ is included because setup.sh + nuke.sh gate the
@@ -368,16 +369,16 @@ jobs:
           find tests/e2e infra/scripts -type f -name '*.sh' -print0 \
             | xargs -0 shellcheck --severity=warning
 
-      - if: needs.changes.outputs.scripts == 'true'
+      - if: always()
         name: Lint cleanup-trap hygiene (RFC #2873)
         run: bash tests/e2e/lint_cleanup_traps.sh
 
-      - if: needs.changes.outputs.scripts == 'true'
+      - if: always()
         name: Run E2E bash unit tests (no live infra)
         run: |
           bash tests/e2e/test_model_slug.sh
 
-      - if: needs.changes.outputs.scripts == 'true'
+      - if: always()
         name: Test ECR promote-tenant-image script (mock-driven, no live infra)
         # Covers scripts/promote-tenant-image.sh — the codified
         # :staging-latest → :latest ECR promote + tenant fleet redeploy
@@ -387,7 +388,7 @@ jobs:
         run: |
           bash scripts/test-promote-tenant-image.sh
 
-      - if: needs.changes.outputs.scripts == 'true'
+      - if: always()
         name: Shellcheck promote-tenant-image script
         # scripts/ is excluded from the bulk shellcheck pass above (legacy
         # SC3040/SC3043 cleanup pending). Run shellcheck explicitly on
@@ -407,8 +408,8 @@ jobs:
     # ci_job_names() detects this as github.ref-gated and skips it from F1.
     # The step-level exit 0 handles the "not main push" case; the job-level
     # `if:` makes the gating explicit so the drift script sees it.
-    # continue-on-error removed (was mc#774 mask): step exits 0 when not applicable.
-    if: ${{ github.ref == 'refs/heads/staging' }}
+    # Runs on both main and staging pushes; step exits 0 when not applicable.
+    if: ${{ github.ref == 'refs/heads/main' || github.ref == 'refs/heads/staging' }}
     needs: [changes, canvas-build]
     steps:
       - name: Write deploy reminder to step summary
@@ -459,7 +460,6 @@ jobs:
   # Python Lint & Test — required check, always runs.
   python-lint:
     name: Python Lint & Test
-    needs: changes
     runs-on: ubuntu-latest
     # Phase 4 (RFC #219 §1): confirmed green on main 2026-05-12.
     continue-on-error: false
@@ -469,25 +469,25 @@ jobs:
       run:
         working-directory: workspace
     steps:
-      - if: needs.changes.outputs.python != 'true'
+      - if: false
         working-directory: .
         run: echo "No workspace/** changes — skipping real lint+test; this job always runs to satisfy the required-check name on branch protection."
-      - if: needs.changes.outputs.python == 'true'
+      - if: always()
         uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
-      - if: needs.changes.outputs.python == 'true'
+      - if: always()
         uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
         with:
           python-version: '3.11'
           cache: pip
           cache-dependency-path: workspace/requirements.txt
-      - if: needs.changes.outputs.python == 'true'
+      - if: always()
         run: pip install -r requirements.txt pytest pytest-asyncio pytest-cov sqlalchemy>=2.0.0
       # Coverage flags + fail-under floor moved into workspace/pytest.ini
       # (issue #1817) so local `pytest` and CI use identical config.
-      - if: needs.changes.outputs.python == 'true'
+      - if: always()
         run: python -m pytest --tb=short
 
-      - if: needs.changes.outputs.python == 'true'
+      - if: always()
         name: Per-file critical-path coverage (MCP / inbox / auth)
         # MCP-critical Python files have a per-file floor on top of the
         # 86% total floor in pytest.ini. See issue #2790 for full rationale.
@@ -552,86 +552,104 @@ jobs:
     # red silently merged through. See internal#286 for the three concrete
     # tonight-of-2026-05-11 incidents that prompted the emergency bump.
     #
-    # Three properties of this job each close a failure mode:
+    # This job deliberately has no `needs:`. Gitea 1.22/act_runner can mark a
+    # job-level `if: always()` + `needs:` sentinel as skipped before upstream
+    # jobs settle, leaving branch protection with a permanent pending
+    # `CI / all-required` context. Instead, this independent sentinel polls the
+    # required commit-status contexts for this SHA and fails if any fail, skip,
+    # or never emit.
     #
-    #  1. `if: always()` — runs even when an upstream fails. Without it the
-    #     sentinel is `skipped` and protection treats that as missing → merge
-    #     ungated.
+    # canvas-deploy-reminder is intentionally NOT included in all-required.needs.
+    # It is an informational main-push reminder, not a PR quality gate. Keeping
+    # it in this dependency list lets a skipped reminder skip the required
+    # sentinel before the `always()` guard can emit a branch-protection status.
     #
-    #  2. Assertion is `result == "success"` per dep, NOT `!= "failure"`.
-    #     A `skipped` upstream (job gated by `if:` evaluating false, matrix
-    #     entry that couldn't run) must NOT silently pass through.
-    #     `skipped`-as-green is exactly the failure mode this gate closes.
-    #
-    #  3. `needs:` is the canonical list of "what counts as required."
-    #     status_check_contexts will reference only `ci/all-required` (Step 5
-    #     follow-up — branch-protection PATCH is Owners-tier per
-    #     `feedback_never_admin_merge_bypass`, separate PR); a new job is
-    #     added simply by listing it in `needs:` here.
-    #     `.gitea/workflows/ci-required-drift.yml` files a [ci-drift] issue
-    #     hourly if this list diverges from status_check_contexts or from
-    #     audit-force-merge.yml's REQUIRED_CHECKS env (RFC §4 + §6).
-    #
-    # canvas-deploy-reminder is intentionally excluded from all-required.needs:
-    # it needs canvas-build, which is skipped on CI-only PRs (canvas=false).
-    # Including it in all-required.needs causes all-required to hang on
-    # every CI-only PR. Keep it runnable on PRs via its own
-    # `needs: [changes, canvas-build]` — the sentinel only aggregates the result.
-    #
-    # Phase 3 (RFC #219 §1) safety: underlying build jobs carry
-    # continue-on-error: true so their failures are masked to null (2026-05-12: re-enabled mc#774 interim)
-    # (Gitea suppresses status reporting for CoE jobs). This sentinel
-    # runs with continue-on-error: false so it always reports its
-    # result to the API — without this, the required-status entry
-    # (CI / all-required (pull_request)) is never created, which
-    # blocks PR merges. When Phase 3 ends, flip underlying jobs to
-    # continue-on-error: false; this sentinel can then be flipped to
-    # continue-on-error: true if a Phase-4 regression requires it.
     continue-on-error: false
     runs-on: ubuntu-latest
-    timeout-minutes: 1
-    needs:
-      - changes
-      - platform-build
-      - canvas-build
-      - shellcheck
-      - python-lint
-      - canvas-deploy-reminder
-    if: ${{ always() }}
+    timeout-minutes: 45
     steps:
-      - name: Assert every required dependency succeeded
+      - name: Wait for required CI contexts
+        env:
+          GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          API_ROOT: ${{ github.server_url }}/api/v1
+          REPOSITORY: ${{ github.repository }}
+          COMMIT_SHA: ${{ github.sha }}
+          EVENT_NAME: ${{ github.event_name }}
         run: |
           set -euo pipefail
-          # `needs.*.result` is one of: success | failure | cancelled | skipped | null.
-          # We assert success per dep (not != failure) — see RFC §2 reasoning above.
-          # Null results are skipped: they come from Phase 3 (continue-on-error: true
-          # suppresses status) or from jobs still in-flight. The sentinel succeeds
-          # rather than blocking PRs on Phase 3 noise.
-          results='${{ toJSON(needs) }}'
-          echo "$results"
-          echo "$results" | python3 -c '
-          import json, sys
-          ns = json.load(sys.stdin)
-          # Phase 3 masked: jobs with continue-on-error: true may report "failure"
-          # Remove when mc#774 handler test failures are resolved.
-          PHASE3_MASKED = {"platform-build"}
-          # Exclude null (Phase 3 suppressed / in-flight) from the bad list.
-          bad = [(k, v.get("result")) for k, v in ns.items()
-                 if v.get("result") not in ("success", None, "cancelled", "skipped") and k not in PHASE3_MASKED]
-          if bad:
-              print(f"FAIL: jobs not green:", file=sys.stderr)
-              for k, r in bad:
-                  print(f"  - {k}: {r}", file=sys.stderr)
-              sys.exit(1)
-          pending = [(k, v.get("result")) for k, v in ns.items()
-                     if v.get("result") is None]
-          cancelled = [(k, v.get("result")) for k, v in ns.items()
-                       if v.get("result") == "cancelled"]
-          if pending:
-              print(f"WARN: {len(pending)} job(s) still in-flight (result=null): " +
-                    ", ".join(k for k, _ in pending), file=sys.stderr)
-          if cancelled:
-              print(f"INFO: {len(cancelled)} job(s) masked by continue-on-error: " +
-                    ", ".join(k for k, _ in cancelled), file=sys.stderr)
-          print(f"OK: all {len(ns)} required jobs succeeded (or Phase-3 suppressed)")
-          '
+          python3 - <<'PY'
+          import json
+          import os
+          import sys
+          import time
+          import urllib.error
+          import urllib.request
+
+          token = os.environ["GITEA_TOKEN"]
+          api_root = os.environ["API_ROOT"].rstrip("/")
+          repo = os.environ["REPOSITORY"]
+          sha = os.environ["COMMIT_SHA"]
+          event = os.environ["EVENT_NAME"]
+          required = [
+              f"CI / Detect changes ({event})",
+              f"CI / Platform (Go) ({event})",
+              f"CI / Canvas (Next.js) ({event})",
+              f"CI / Shellcheck (E2E scripts) ({event})",
+              f"CI / Python Lint & Test ({event})",
+          ]
+          terminal_bad = {"failure", "error"}
+          deadline = time.time() + 40 * 60
+          last_summary = None
+
+          def fetch_statuses():
+              statuses = []
+              for page in range(1, 6):
+                  url = f"{api_root}/repos/{repo}/commits/{sha}/statuses?page={page}&limit=100"
+                  req = urllib.request.Request(url, headers={"Authorization": f"token {token}"})
+                  with urllib.request.urlopen(req, timeout=10) as resp:
+                      chunk = json.load(resp)
+                  if not chunk:
+                      break
+                  statuses.extend(chunk)
+              latest = {}
+              for item in statuses:
+                  ctx = item.get("context")
+                  if not ctx:
+                      continue
+                  prev = latest.get(ctx)
+                  if prev is None or (item.get("updated_at") or item.get("created_at") or "") >= (prev.get("updated_at") or prev.get("created_at") or ""):
+                      latest[ctx] = item
+              return latest
+
+          while True:
+              try:
+                  latest = fetch_statuses()
+              except (TimeoutError, OSError, urllib.error.URLError) as exc:
+                  if time.time() >= deadline:
+                      print(f"FAIL: status polling did not recover before deadline: {exc}", file=sys.stderr)
+                      sys.exit(1)
+                  print(f"WARN: status poll failed, retrying: {exc}", flush=True)
+                  time.sleep(15)
+                  continue
+              states = {ctx: (latest.get(ctx) or {}).get("status") or (latest.get(ctx) or {}).get("state") or "missing" for ctx in required}
+              summary = ", ".join(f"{ctx}={state}" for ctx, state in states.items())
+              if summary != last_summary:
+                  print(summary, flush=True)
+                  last_summary = summary
+              bad = {ctx: state for ctx, state in states.items() if state in terminal_bad}
+              if bad:
+                  print("FAIL: required CI context failed:", file=sys.stderr)
+                  for ctx, state in bad.items():
+                      desc = (latest.get(ctx) or {}).get("description") or ""
+                      print(f"  - {ctx}: {state} {desc}", file=sys.stderr)
+                  sys.exit(1)
+              if all(state == "success" for state in states.values()):
+                  print(f"OK: all {len(required)} required CI contexts succeeded")
+                  sys.exit(0)
+              if time.time() >= deadline:
+                  print("FAIL: timed out waiting for required CI contexts:", file=sys.stderr)
+                  for ctx, state in states.items():
+                      print(f"  - {ctx}: {state}", file=sys.stderr)
+                  sys.exit(1)
+              time.sleep(15)
+          PY

.gitea/workflows/e2e-api.yml

@@ -69,6 +69,13 @@ name: E2E API Smoke Test
 # 2318) shows Postgres ready in 3s, Redis in 1s, Platform in 1s when
 # they DO come up. Timeouts are not the bottleneck; not bumped.
 #
+# Item #1046 (fixed 2026-05-14): Stale platform-server from cancelled runs
+#   lingers on :8080 after "Stop platform" step is skipped (workflow cancelled
+#   before reaching line 335). Added a pre-start "Kill stale platform-server"
+#   step (line 286) that scans /proc for zombie platform-server processes
+#   and kills them before the port probe or bind. Makes the ephemeral port
+#   probe + start sequence deterministic.
+#
 # Item explicitly NOT fixed here: failing test `Status back online`
 # fails because the platform's langgraph workspace template image
 # (ghcr.io/molecule-ai/workspace-template-langgraph:latest) returns
@@ -283,6 +290,35 @@ jobs:
           echo "PORT=${PLATFORM_PORT}" >> "$GITHUB_ENV"
           echo "BASE=http://127.0.0.1:${PLATFORM_PORT}" >> "$GITHUB_ENV"
           echo "Platform host port: ${PLATFORM_PORT}"
+      - name: Kill stale platform-server before start (issue #1046)
+        if: needs.detect-changes.outputs.api == 'true'
+        run: |
+          # Concurrent runs on the same host-network act_runner can leave a
+          # zombie platform-server from a cancelled/timeout run. Cancelled
+          # runs never reach the "Stop platform" step (line 335), so the
+          # old process lingers. Kill it before the ephemeral port probe
+          # or start so the port is definitively free.
+          #
+          # /proc scan — works on any Linux without pkill/lsof/ss.
+          # comm field is truncated to 15 chars: "platform-serve" matches
+          # "platform-server". Verify with cmdline to avoid false positives.
+          killed=0
+          for pid in $(grep -l "platform-serve" /proc/[0-9]*/comm 2>/dev/null); do
+            kpid="${pid%/comm}"
+            kpid="${kpid##*/}"
+            cmdline=$(cat "/proc/${kpid}/cmdline" 2>/dev/null | tr '\0' ' ')
+            if echo "$cmdline" | grep -q "platform-server"; then
+              echo "Killing stale platform-server pid ${kpid}: ${cmdline}"
+              kill "$kpid" 2>/dev/null || true
+              killed=$((killed + 1))
+            fi
+          done
+          if [ "$killed" -gt 0 ]; then
+            sleep 2
+            echo "Killed $killed stale process(es); port(s) released."
+          else
+            echo "No stale platform-server found."
+          fi
       - name: Start platform (background)
         if: needs.detect-changes.outputs.api == 'true'
         working-directory: workspace-server
@@ -346,3 +382,4 @@ jobs:
         run: |
           docker rm -f "$PG_CONTAINER" 2>/dev/null || true
           docker rm -f "$REDIS_CONTAINER" 2>/dev/null || true
+

.gitea/workflows/e2e-chat.yml

@@ -97,7 +97,7 @@ jobs:
           cache-dependency-path: workspace-server/go.sum
 
       - if: needs.detect-changes.outputs.chat == 'true'
-        uses: actions/setup-node@60edb5dd545a775178f52524783378180af0d6f5 # v4
+        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
         with:
           node-version: '22'
           cache: 'npm'
@@ -175,6 +175,19 @@ jobs:
           echo "E2E_PLATFORM_URL=http://127.0.0.1:${PLATFORM_PORT}" >> "$GITHUB_ENV"
           echo "Platform host port: ${PLATFORM_PORT}"
 
+      - name: Pick canvas port
+        if: needs.detect-changes.outputs.chat == 'true'
+        run: |
+          CANVAS_PORT=$(python3 - <<'PY'
+          import socket
+          with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+              s.bind(("127.0.0.1", 0))
+              print(s.getsockname()[1])
+          PY
+          )
+          echo "CANVAS_PORT=${CANVAS_PORT}" >> "$GITHUB_ENV"
+          echo "Canvas host port: ${CANVAS_PORT}"
+
       - name: Start platform (background)
         if: needs.detect-changes.outputs.chat == 'true'
         working-directory: workspace-server
@@ -183,6 +196,7 @@ jobs:
           export DATABASE_URL="${DATABASE_URL}"
           export REDIS_URL="${REDIS_URL}"
           export PORT="${PLATFORM_PORT}"
+          export CORS_ORIGINS="http://localhost:3000,http://localhost:3001,http://localhost:${CANVAS_PORT},http://127.0.0.1:${CANVAS_PORT}"
           ./platform-server > platform.log 2>&1 &
           echo $! > platform.pid
 
@@ -216,10 +230,10 @@ jobs:
         run: |
           export NEXT_PUBLIC_PLATFORM_URL="http://127.0.0.1:${PLATFORM_PORT}"
           export NEXT_PUBLIC_WS_URL="ws://127.0.0.1:${PLATFORM_PORT}/ws"
-          npm run dev > canvas.log 2>&1 &
+          npx next dev --turbopack -p "${CANVAS_PORT}" > canvas.log 2>&1 &
           echo $! > canvas.pid
           for i in $(seq 1 30); do
-            if curl -sf http://localhost:3000 > /dev/null 2>&1; then
+            if curl -sf "http://localhost:${CANVAS_PORT}" > /dev/null 2>&1; then
               echo "Canvas up after ${i}s"
               exit 0
             fi
@@ -235,6 +249,7 @@ jobs:
         run: |
           export E2E_PLATFORM_URL="http://127.0.0.1:${PLATFORM_PORT}"
           export E2E_DATABASE_URL="${DATABASE_URL}"
+          export PLAYWRIGHT_BASE_URL="http://localhost:${CANVAS_PORT}"
           npx playwright test e2e/chat-desktop.spec.ts e2e/chat-mobile.spec.ts
 
       - name: Dump platform log on failure

.gitea/workflows/e2e-peer-visibility.yml

@@ -0,0 +1,225 @@
+name: E2E Peer Visibility (literal MCP list_peers)
+
+# WHY A DEDICATED WORKFLOW (not folded into e2e-staging-saas.yml)
+# --------------------------------------------------------------
+# This is the systemic fix for a real trust failure. Hermes and OpenClaw
+# were reported "fleet-verified / cascade-complete" because the *proxy*
+# signals were green (registry registration + heartbeat for Hermes; model
+# round-trip 200 for OpenClaw). A freshly-provisioned workspace asked on
+# canvas "can you see your peers" actually FAILS:
+#   - Hermes: 401 on the molecule MCP `list_peers` call
+#   - OpenClaw: native `sessions_list` fallback, sees no platform peers
+# Tasks #142/#159 were even marked "completed" under this proxy flaw.
+#
+# A dedicated workflow (vs extending e2e-staging-saas.yml) because:
+#   - It must provision MULTIPLE distinct runtimes (hermes, openclaw,
+#     claude-code) in ONE org and assert each sees the others. The
+#     full-saas script is single-runtime-per-run (E2E_RUNTIME) and folding
+#     a multi-runtime matrix into it would conflate concerns and bloat its
+#     already-45-min run.
+#   - It needs its own concurrency group so it doesn't fight full-saas /
+#     canvas for the staging org-creation quota.
+#   - It needs an independent, non-required status-context name so it can
+#     be RED today (the in-flight Hermes-401 / OpenClaw-MCP-wiring fixes
+#     have not landed) WITHOUT wedging unrelated merges — and flipped to
+#     REQUIRED in one branch-protection edit once it goes green
+#     (flip-to-required checklist: molecule-core#1296).
+#
+# THE ASSERTION IS NOT A PROXY. The driving script
+# tests/e2e/test_peer_visibility_mcp_staging.sh issues the byte-for-byte
+# JSON-RPC `tools/call name=list_peers` envelope to `POST
+# /workspaces/:id/mcp` using each workspace's OWN bearer token, through
+# the real WorkspaceAuth + MCPRateLimiter middleware chain — the exact
+# call mcp_molecule_list_peers makes from a canvas agent. It does NOT
+# read a registry row, /health, the heartbeat table, or
+# GET /registry/:id/peers.
+#
+# HONEST GATE — NO continue-on-error. Per feedback_fix_root_not_symptom a
+# fake-green mask would defeat the entire purpose. This workflow goes red
+# on today's broken behavior and green only when the root-cause fixes
+# actually land. It is intentionally NOT in branch_protections — see PR
+# body for the required-vs-not decision + flip tracking issue.
+#
+# Gitea 1.22.6 / act_runner notes honored:
+#   - No cross-repo `uses:` (feedback_gitea_cross_repo_uses_blocked). The
+#     actions/checkout SHA is the one e2e-staging-canvas.yml already uses
+#     successfully (a mirrored SHA — see #1277/PR#1292 root-cause).
+#   - Per-SHA concurrency, not global (feedback_concurrency_group_per_sha).
+#   - Workflow-level GITHUB_SERVER_URL pinned
+#     (feedback_act_runner_github_server_url).
+#   - pr-validate posts a status under the same check name so a
+#     workflow-only PR is not silently statusless and the context is
+#     flip-to-required-ready (mirrors e2e-staging-saas.yml's proven shape;
+#     real EC2-provisioning E2E is push/dispatch/cron only — it is 30+ min
+#     and cannot run per-PR-update).
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'workspace-server/internal/handlers/mcp.go'
+      - 'workspace-server/internal/handlers/mcp_tools.go'
+      - 'workspace-server/internal/middleware/**'
+      - 'workspace-server/internal/handlers/registry.go'
+      - 'workspace-server/internal/handlers/workspace.go'
+      - 'workspace/a2a_mcp_server.py'
+      - 'workspace/platform_tools/registry.py'
+      - 'tests/e2e/test_peer_visibility_mcp_staging.sh'
+      - '.gitea/workflows/e2e-peer-visibility.yml'
+  pull_request:
+    branches: [main]
+    paths:
+      - 'workspace-server/internal/handlers/mcp.go'
+      - 'workspace-server/internal/handlers/mcp_tools.go'
+      - 'workspace-server/internal/middleware/**'
+      - 'workspace-server/internal/handlers/registry.go'
+      - 'workspace-server/internal/handlers/workspace.go'
+      - 'workspace/a2a_mcp_server.py'
+      - 'workspace/platform_tools/registry.py'
+      - 'tests/e2e/test_peer_visibility_mcp_staging.sh'
+      - '.gitea/workflows/e2e-peer-visibility.yml'
+  workflow_dispatch:
+  schedule:
+    # 07:30 UTC daily — catches AMI / template-hermes / template-openclaw
+    # drift even on quiet days. Offset 30m from e2e-staging-saas (07:00)
+    # so the two don't collide on the staging org-creation quota.
+    - cron: '30 7 * * *'
+
+concurrency:
+  # Per-SHA (feedback_concurrency_group_per_sha). A single global group
+  # would let a queued staging/main push behind a PR run get cancelled,
+  # leaving any gate that reads "completed run at SHA" stuck.
+  group: e2e-peer-visibility-${{ github.event.pull_request.head.sha || github.sha }}
+  cancel-in-progress: false
+
+env:
+  GITHUB_SERVER_URL: https://git.moleculesai.app
+
+jobs:
+  # PR path: post a real status under the required-ready check name so a
+  # workflow-only PR is never silently statusless. The actual EC2 E2E is
+  # push/dispatch/cron only (30+ min). This is NOT a fake-green mask of
+  # the real assertion — it validates the driving script's bash syntax
+  # and inline-python so a broken test script fails at PR time.
+  pr-validate:
+    name: E2E Peer Visibility
+    runs-on: ubuntu-latest
+    if: github.event_name == 'pull_request'
+    timeout-minutes: 5
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+      - name: Validate driving script
+        run: |
+          bash -n tests/e2e/test_peer_visibility_mcp_staging.sh
+          echo "test_peer_visibility_mcp_staging.sh — bash syntax OK"
+          echo "Real fresh-provision MCP list_peers E2E runs on push to"
+          echo "main / workflow_dispatch / daily cron (30+ min EC2 boot)."
+
+  # Real gate: provisions a throwaway org + sibling-per-runtime, drives
+  # the LITERAL list_peers MCP call per runtime, asserts 200 + expected
+  # peer set, then scoped teardown. push(main)/dispatch/cron only.
+  peer-visibility:
+    name: E2E Peer Visibility
+    runs-on: ubuntu-latest
+    if: github.event_name != 'pull_request'
+    timeout-minutes: 60
+
+    env:
+      MOLECULE_CP_URL: https://staging-api.moleculesai.app
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
+      # LLM provider key so each runtime can authenticate at boot.
+      # Priority MiniMax → direct-Anthropic → OpenAI matches
+      # test_staging_full_saas.sh's secrets-injection chain.
+      E2E_MINIMAX_API_KEY: ${{ secrets.MOLECULE_STAGING_MINIMAX_API_KEY }}
+      E2E_ANTHROPIC_API_KEY: ${{ secrets.MOLECULE_STAGING_ANTHROPIC_API_KEY }}
+      E2E_OPENAI_API_KEY: ${{ secrets.MOLECULE_STAGING_OPENAI_API_KEY }}
+      E2E_RUN_ID: "${{ github.run_id }}-${{ github.run_attempt }}"
+      PV_RUNTIMES: "hermes openclaw claude-code"
+
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+
+      - name: Verify admin token present
+        run: |
+          if [ -z "$MOLECULE_ADMIN_TOKEN" ]; then
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN secret not set (Railway staging CP_ADMIN_API_TOKEN)"
+            exit 2
+          fi
+          echo "Admin token present"
+
+      - name: Verify an LLM key present
+        run: |
+          if [ -z "${E2E_MINIMAX_API_KEY:-}" ] && [ -z "${E2E_ANTHROPIC_API_KEY:-}" ] && [ -z "${E2E_OPENAI_API_KEY:-}" ]; then
+            echo "::error::No LLM provider key set — workspaces fail at boot with 'No provider API key found'. Set MOLECULE_STAGING_MINIMAX_API_KEY (or ANTHROPIC / OPENAI)."
+            exit 2
+          fi
+          echo "LLM key present"
+
+      - name: CP staging health preflight
+        run: |
+          code=$(curl -sS -o /dev/null -w "%{http_code}" --max-time 10 "$MOLECULE_CP_URL/health")
+          if [ "$code" != "200" ]; then
+            echo "::error::Staging CP unhealthy (HTTP $code) — infra, not a workspace bug. Failing loud per feedback_fix_root_not_symptom."
+            exit 1
+          fi
+          echo "Staging CP healthy"
+
+      - name: Run fresh-provision peer-visibility E2E (literal MCP list_peers)
+        run: bash tests/e2e/test_peer_visibility_mcp_staging.sh
+
+      # Belt-and-braces scoped teardown: the script installs an EXIT/INT/
+      # TERM trap, but if the runner itself is cancelled the trap may not
+      # fire. This always() step deletes ONLY the e2e-pv-<run_id> org this
+      # run created — never a cluster-wide sweep
+      # (feedback_never_run_cluster_cleanup_tests_on_live_platform). The
+      # admin DELETE is idempotent so double-invoking is safe;
+      # sweep-stale-e2e-orgs is the final net (slug starts with 'e2e-').
+      - name: Teardown safety net (runs on cancel/failure)
+        if: always()
+        env:
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
+        run: |
+          set +e
+          orgs=$(curl -sS "$MOLECULE_CP_URL/cp/admin/orgs?limit=500" \
+            -H "Authorization: Bearer $ADMIN_TOKEN" 2>/dev/null \
+            | python3 -c "
+          import json, sys, os, datetime
+          run_id = os.environ.get('GITHUB_RUN_ID', '')
+          try:
+              d = json.load(sys.stdin)
+          except Exception:
+              print(''); sys.exit(0)
+          # ONLY sweep slugs from THIS run. e2e-pv-<YYYYMMDD>-<run_id>-...
+          # Sweep today AND yesterday's UTC date so a midnight-crossing run
+          # still matches its own slug (same bug class as the saas/canvas
+          # safety nets).
+          today = datetime.date.today()
+          yest = today - datetime.timedelta(days=1)
+          dates = (today.strftime('%Y%m%d'), yest.strftime('%Y%m%d'))
+          if run_id:
+              prefixes = tuple(f'e2e-pv-{dt}-{run_id}-' for dt in dates)
+          else:
+              prefixes = tuple(f'e2e-pv-{dt}-' for dt in dates)
+          orgs = d if isinstance(d, list) else d.get('orgs', [])
+          cands = [o['slug'] for o in orgs
+                   if any(o.get('slug','').startswith(p) for p in prefixes)
+                   and o.get('instance_status') not in ('purged',)]
+          print('\n'.join(cands))
+          " 2>/dev/null)
+          for slug in $orgs; do
+            echo "Safety-net teardown: $slug"
+            set +e
+            curl -sS -o /tmp/pv-cleanup.out -w "%{http_code}" \
+              -X DELETE "$MOLECULE_CP_URL/cp/admin/tenants/$slug" \
+              -H "Authorization: Bearer $ADMIN_TOKEN" \
+              -H "Content-Type: application/json" \
+              -d "{\"confirm\":\"$slug\"}" >/tmp/pv-cleanup.code
+            set -e
+            code=$(cat /tmp/pv-cleanup.code 2>/dev/null || echo "000")
+            if [ "$code" = "200" ] || [ "$code" = "204" ]; then
+              echo "[teardown] deleted $slug (HTTP $code)"
+            else
+              echo "::warning::pv teardown for $slug returned HTTP $code — sweep-stale-e2e-orgs will catch it within MAX_AGE_MINUTES. Body: $(head -c 300 /tmp/pv-cleanup.out 2>/dev/null)"
+            fi
+          done
+          exit 0

.gitea/workflows/gate-check-v3.yml

@@ -83,25 +83,41 @@ jobs:
           REPO: ${{ github.repository }}
         run: |
           set -euo pipefail
-          # Fetch all open PRs and run gate-check on each
-          # socket.setdefaulttimeout(15): defence-in-depth for missing SOP_TIER_CHECK_TOKEN.
-          # gate_check.py uses timeout=15 on every urlopen call; this catches the
-          # inline Python polling loop too (issue #603).
+          # Fetch all open PRs and run gate-check on each. This scheduled
+          # refresher is advisory; a transient Gitea list timeout must not turn
+          # main red. PR-specific gate-check runs still use normal failure
+          # semantics.
           pr_numbers=$(python3 <<'PY'
           import json
           import os
           import socket
+          import sys
+          import time
+          import urllib.error
           import urllib.request
 
-          socket.setdefaulttimeout(15)
+          socket.setdefaulttimeout(30)
           token = os.environ["GITEA_TOKEN"]
           repo = os.environ["REPO"]
-          req = urllib.request.Request(
-              f"https://git.moleculesai.app/api/v1/repos/{repo}/pulls?state=open&limit=100",
-              headers={"Authorization": f"token {token}", "Accept": "application/json"},
-          )
-          with urllib.request.urlopen(req) as r:
-              prs = json.loads(r.read())
+          url = f"https://git.moleculesai.app/api/v1/repos/{repo}/pulls?state=open&limit=100"
+          last_error = None
+          for attempt in range(1, 4):
+              req = urllib.request.Request(
+                  url,
+                  headers={"Authorization": f"token {token}", "Accept": "application/json"},
+              )
+              try:
+                  with urllib.request.urlopen(req, timeout=30) as r:
+                      prs = json.loads(r.read())
+                  break
+              except (TimeoutError, OSError, urllib.error.URLError, urllib.error.HTTPError) as exc:
+                  last_error = exc
+                  print(f"warning: PR list fetch attempt {attempt}/3 failed: {exc}", file=sys.stderr)
+                  if attempt < 3:
+                      time.sleep(2 * attempt)
+          else:
+              print(f"warning: skipped scheduled gate-check refresh; failed to list open PRs after 3 attempts: {last_error}", file=sys.stderr)
+              raise SystemExit(0)
           for pr in prs:
               print(pr["number"])
           PY

.gitea/workflows/handlers-postgres-integration.yml

@@ -86,7 +86,11 @@ jobs:
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
-          fetch-depth: 0
+          # A full-history checkout can exceed the runner's quiet/startup
+          # window before the path filter emits logs. Fetch the common push
+          # case cheaply; the script below fetches the exact BASE SHA if it is
+          # not present in the shallow checkout.
+          fetch-depth: 2
       - id: filter
         # Inline replacement for dorny/paths-filter — see e2e-api.yml.
         run: |

.gitea/workflows/lint-continue-on-error-tracking.yml

@@ -93,7 +93,7 @@ jobs:
   lint:
     name: lint-continue-on-error-tracking
     runs-on: ubuntu-latest
-    timeout-minutes: 10
+    timeout-minutes: 20
     # Phase 3 (RFC #219 §1): surface masked defects without blocking
     # PRs. Pre-existing continue-on-error: true directives on main
     # all violate this lint at first — intentional. Flip to false

.gitea/workflows/publish-canvas-image.yml

@@ -49,13 +49,17 @@ jobs:
   # bp-exempt: post-merge image publication side effect; CI / all-required gates source changes.
   build-and-push:
     name: Build & push canvas image
-    # REVERTED (infra/revert-docker-runner-label): `runs-on: ubuntu-latest` restored.
-    # The `docker` label is not registered on any act_runner. `runs-on: [ubuntu-latest, docker]`
-    # causes jobs to queue indefinitely with zero eligible runners — strictly worse than the
-    # pre-#599 coin-flip (50% success rate). Once the `docker` label is registered on
-    # ≥2 runners, re-apply the fix from #599 (infra/docker-runner-label).
-    # See issue #576 + infra-lead pulse ~00:30Z.
-    runs-on: ubuntu-latest
+    # Dedicated publish/release lane (internal#462 / #394 / #399). Ship
+    # path (on: push:main, canvas/**) — reserved capacity so a merged
+    # canvas fix's image build never FIFO-queues behind PR required-CI.
+    # The `publish` label resolves ONLY to the molecule-runner-publish-*
+    # sub-pool (config.publish.yaml). HARD DEPENDENCY: this MUST land
+    # AFTER the publish-lane runners are registered/advertising `publish`
+    # — the earlier #599 `docker` label attempt queued indefinitely with
+    # zero eligible runners precisely because the label was targeted
+    # before any runner advertised it (see #576). The lane is registered
+    # in this rollout (internal#462) so the precondition holds.
+    runs-on: publish
     # Phase 3 (RFC #219 §1): surface broken workflows without blocking.
     # mc#774: pre-existing continue-on-error mask; root-fix and remove, do not renew silently.
     continue-on-error: true

.gitea/workflows/publish-runtime.yml

@@ -66,7 +66,10 @@ concurrency:
 
 jobs:
   publish:
-    runs-on: ubuntu-latest
+    # Dedicated publish/release lane (internal#462 / #394 / #399). Ship
+    # path (on: push tag runtime-v*) — reserved capacity, never FIFO
+    # behind PR-CI. `publish` resolves only to molecule-runner-publish-*.
+    runs-on: publish
     outputs:
       version: ${{ steps.version.outputs.version }}
       wheel_sha256: ${{ steps.wheel_hash.outputs.wheel_sha256 }}
@@ -159,6 +162,7 @@ jobs:
             exit 1
           fi
           python -m twine upload \
+            --verbose \
             --repository pypi \
             --username __token__ \
             --password "$PYPI_TOKEN" \
@@ -166,7 +170,9 @@ jobs:
 
   cascade:
     needs: publish
-    runs-on: ubuntu-latest
+    # Publish/release lane (internal#462) — downstream of the runtime
+    # publish ship job; keep it on the reserved lane too.
+    runs-on: publish
     steps:
       - name: Wait for PyPI to propagate the new version
         env:

.gitea/workflows/publish-workspace-server-image.yml

@@ -54,7 +54,14 @@ env:
 
 jobs:
   build-and-push:
-    runs-on: ubuntu-latest
+    # Dedicated publish/release lane (internal#462 / #394 / #399). This
+    # is a post-merge ship job (on: push:main) — it must NOT FIFO-compete
+    # with PR required-CI on the shared pool (PR#1350's prod image build
+    # was delayed ~25min this way). The `publish` label resolves ONLY to
+    # the reserved molecule-runner-publish-* sub-pool (config.publish.yaml,
+    # OUTSIDE the managed 1..20 range) so a merged fix's image build
+    # starts immediately while PR-CI keeps the general pool.
+    runs-on: publish
     steps:
       - name: Checkout
         uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
@@ -181,7 +188,9 @@ jobs:
     name: Production auto-deploy
     needs: build-and-push
     if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' }}
-    runs-on: ubuntu-latest
+    # Publish/release lane (internal#462) — production deploy of a merged
+    # fix; reserved capacity, never queued behind PR-CI.
+    runs-on: publish
     timeout-minutes: 75
     env:
       CP_URL: ${{ vars.PROD_CP_URL || 'https://api.moleculesai.app' }}

.gitea/workflows/redeploy-tenants-on-main.yml

@@ -68,7 +68,10 @@ jobs:
   # bp-exempt: production redeploy is a side-effect workflow, not a merge gate.
   redeploy:
     if: ${{ github.event_name == 'workflow_dispatch' }}
-    runs-on: ubuntu-latest
+    # Dedicated publish/release lane (internal#462 / #394 / #399).
+    # Production tenant redeploy — a deploy action, reserved capacity so
+    # it never queues behind PR-CI. `publish` -> molecule-runner-publish-*.
+    runs-on: publish
     # Phase 3 (RFC #219 §1): surface broken workflows without blocking.
     # mc#774: pre-existing continue-on-error mask; root-fix and remove, do not renew silently.
     continue-on-error: true

.gitea/workflows/redeploy-tenants-on-staging.yml

@@ -75,7 +75,10 @@ env:
 jobs:
   # bp-exempt: post-merge staging redeploy side effect; CI / all-required gates source changes.
   redeploy:
-    runs-on: ubuntu-latest
+    # Dedicated publish/release lane (internal#462 / #394 / #399).
+    # Post-merge staging redeploy — a deploy action, reserved capacity.
+    # `publish` -> molecule-runner-publish-* sub-pool.
+    runs-on: publish
     # Phase 3 (RFC #219 §1): surface broken workflows without blocking.
     # mc#774: pre-existing continue-on-error mask; root-fix and remove, do not renew silently.
     continue-on-error: true

.gitea/workflows/review-refire-comments.yml

@@ -18,6 +18,10 @@ permissions:
   pull-requests: read
   statuses: write
 
+concurrency:
+  group: ${{ github.repository }}-${{ github.workflow }}-${{ github.event.issue.number || github.ref }}
+  cancel-in-progress: true
+
 jobs:
   dispatch:
     runs-on: ubuntu-latest

.gitea/workflows/sop-checklist.yml

@@ -70,7 +70,7 @@ name: sop-checklist
 # Cancel any in-progress runs for the same PR to prevent
 # stale runs from overwriting newer status contexts.
 concurrency:
-  group: ${{ github.repository }}-${{ github.event.pull_request.number }}
+  group: ${{ github.repository }}-${{ github.workflow }}-${{ github.event.pull_request.number || github.event.issue.number || github.ref }}
   cancel-in-progress: true
 
 # bp-required: yes  ← emits sop-checklist / all-items-acked (pull_request)

.gitea/workflows/sop-tier-check.yml

@@ -61,6 +61,10 @@ on:
   pull_request_review:
     types: [submitted, dismissed, edited]
 
+concurrency:
+  group: ${{ github.repository }}-${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 jobs:
   tier-check:
     runs-on: ubuntu-latest

.staging-trigger

@@ -1 +1 @@
-staging trigger
+staging trigger 2026-05-14T17:35:02Z

_ci_trigger.txt

@@ -0,0 +1 @@
+trigger

canvas/playwright.config.ts

@@ -8,7 +8,7 @@ export default defineConfig({
   workers: 1,
   retries: 0,
   use: {
-    baseURL: "http://localhost:3000",
+    baseURL: process.env.PLAYWRIGHT_BASE_URL || "http://localhost:3000",
     headless: true,
     screenshot: "only-on-failure",
   },

canvas/src/components/__tests__/TestConnectionButton.test.tsx

@@ -11,13 +11,21 @@ import { render, screen, fireEvent, cleanup, act } from "@testing-library/react"
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { TestConnectionButton } from "../ui/TestConnectionButton";
 import type { SecretGroup } from "@/types/secrets";
-import { validateSecret } from "@/lib/api/secrets";
+import { validateSecret, ApiError } from "@/lib/api/secrets";
 
 // ─── Mock validateSecret ──────────────────────────────────────────────────────
 // vi.mock is hoisted, so validateSecret (imported above) refers to the mocked
 // namespace value once vi.mock runs. Use vi.mocked() to access it in tests.
 vi.mock("@/lib/api/secrets", () => ({
   validateSecret: vi.fn(),
+  ApiError: class ApiError extends Error {
+    status: number;
+    constructor(status: number, message: string) {
+      super(message);
+      this.name = "ApiError";
+      this.status = status;
+    }
+  },
 }));
 
 // SecretGroup is a string literal type: 'github' | 'anthropic' | 'openrouter' | 'custom'
@@ -102,7 +110,7 @@ describe("TestConnectionButton — state machine", () => {
     expect(screen.getByText("Permission denied")).toBeTruthy();
   });
 
-  it("shows generic error message on unexpected exception", async () => {
+  it("shows a connectivity message on a genuine network exception", async () => {
     vi.mocked(validateSecret).mockRejectedValue(new Error("timeout"));
     render(<TestConnectionButton provider={toGroup("anthropic")} secretValue="sk-..." />);
 
@@ -110,8 +118,23 @@ describe("TestConnectionButton — state machine", () => {
     await act(async () => { /* flush */ });
 
     expect(screen.getByRole("alert")).toBeTruthy();
-    // The error detail is hardcoded to "Connection timed out. Service may be down."
-    expect(document.body.querySelector('[role="alert"]')?.textContent).toMatch(/timed out/i);
+    // A real thrown network error → honest connectivity message (not a
+    // fabricated "service down"); see internal#492.
+    expect(document.body.querySelector('[role="alert"]')?.textContent).toMatch(
+      /could not reach the validation service/i,
+    );
+  });
+
+  it("does not claim a timeout when the validate endpoint 404s (internal#492)", async () => {
+    vi.mocked(validateSecret).mockRejectedValue(new ApiError(404, "Not Found"));
+    render(<TestConnectionButton provider={toGroup("anthropic")} secretValue="sk-..." />);
+
+    fireEvent.click(screen.getByRole("button"));
+    await act(async () => { /* flush */ });
+
+    const alert = document.body.querySelector('[role="alert"]')?.textContent ?? "";
+    expect(alert).not.toMatch(/timed out/i);
+    expect(alert).toMatch(/not available/i);
   });
 });
 

canvas/src/components/__tests__/ThemeToggle.test.tsx

@@ -24,8 +24,12 @@ vi.mock("@/lib/theme-provider", () => ({
   })),
 }));
 
+// Wrap cleanup in act() so any pending React state updates (e.g. from
+// keyDown handlers that call setTheme) flush before DOM unmount. Without
+// this, cleanup() can race against pending renders and cause INDEX_SIZE_ERR
+// when the handleKeyDown callback tries to query the DOM mid-teardown.
 afterEach(() => {
-  cleanup();
+  act(() => { cleanup(); });
   vi.clearAllMocks();
 });
 
@@ -146,7 +150,7 @@ describe("ThemeToggle — keyboard navigation (WCAG 2.1.1 / ARIA radiogroup)", (
     const radios = screen.getAllByRole("radio");
     // dark (index 2) is current; ArrowRight should wrap to light (index 0)
     act(() => { radios[2].focus(); });
-    fireEvent.keyDown(radios[2], { key: "ArrowRight" });
+    act(() => { fireEvent.keyDown(radios[2], { key: "ArrowRight" }); });
     expect(mockSetTheme).toHaveBeenCalledWith("light");
   });
 
@@ -160,7 +164,7 @@ describe("ThemeToggle — keyboard navigation (WCAG 2.1.1 / ARIA radiogroup)", (
     const radios = screen.getAllByRole("radio");
     // light (index 0) is current; ArrowLeft should go to dark (index 2)
     act(() => { radios[0].focus(); });
-    fireEvent.keyDown(radios[0], { key: "ArrowLeft" });
+    act(() => { fireEvent.keyDown(radios[0], { key: "ArrowLeft" }); });
     expect(mockSetTheme).toHaveBeenCalledWith("dark");
   });
 
@@ -174,7 +178,7 @@ describe("ThemeToggle — keyboard navigation (WCAG 2.1.1 / ARIA radiogroup)", (
     const radios = screen.getAllByRole("radio");
     // light (index 0) is current; ArrowDown should go to system (index 1)
     act(() => { radios[0].focus(); });
-    fireEvent.keyDown(radios[0], { key: "ArrowDown" });
+    act(() => { fireEvent.keyDown(radios[0], { key: "ArrowDown" }); });
     expect(mockSetTheme).toHaveBeenCalledWith("system");
   });
 
@@ -187,7 +191,7 @@ describe("ThemeToggle — keyboard navigation (WCAG 2.1.1 / ARIA radiogroup)", (
     render(<ThemeToggle />);
     const radios = screen.getAllByRole("radio");
     act(() => { radios[2].focus(); });
-    fireEvent.keyDown(radios[2], { key: "Home" });
+    act(() => { fireEvent.keyDown(radios[2], { key: "Home" }); });
     expect(mockSetTheme).toHaveBeenCalledWith("light");
   });
 
@@ -200,14 +204,14 @@ describe("ThemeToggle — keyboard navigation (WCAG 2.1.1 / ARIA radiogroup)", (
     render(<ThemeToggle />);
     const radios = screen.getAllByRole("radio");
     act(() => { radios[0].focus(); });
-    fireEvent.keyDown(radios[0], { key: "End" });
+    act(() => { fireEvent.keyDown(radios[0], { key: "End" }); });
     expect(mockSetTheme).toHaveBeenCalledWith("dark");
   });
 
   it("does nothing on unrelated keys", () => {
     render(<ThemeToggle />);
     const radios = screen.getAllByRole("radio");
-    fireEvent.keyDown(radios[0], { key: "Enter" });
+    act(() => { fireEvent.keyDown(radios[0], { key: "Enter" }); });
     expect(mockSetTheme).not.toHaveBeenCalled();
   });
 });

canvas/src/components/mobile/MobileChat.tsx

@@ -2,8 +2,11 @@
 
 // 04 · Chat — message thread + composer + sub-tabs.
 // Wired to the same /workspaces/:id/a2a (method message/send) endpoint
-// that the desktop ChatTab uses, but with a slimmer surface: no
-// attachments, no A2A topology overlay, no conversation tracing.
+// that the desktop ChatTab uses. Render parity with desktop ChatTab is
+// achieved by reusing its renderers rather than forking a reduced
+// mobile path: the Agent Comms sub-tab mounts the same AgentCommsPanel,
+// and message attachments route through the same AttachmentPreview
+// dispatch the desktop My-Chat bubble uses (#231/#232).
 
 import { useEffect, useMemo, useRef, useState } from "react";
 import ReactMarkdown from "react-markdown";
@@ -16,6 +19,9 @@ import {
   useChatSend,
   useChatSocket,
 } from "@/components/tabs/chat/hooks";
+import { AgentCommsPanel } from "@/components/tabs/chat/AgentCommsPanel";
+import { AttachmentPreview } from "@/components/tabs/chat/AttachmentPreview";
+import { downloadChatFile } from "@/components/tabs/chat/uploads";
 
 import { toMobileAgent } from "./components";
 import { MOBILE_FONT_MONO, MOBILE_FONT_SANS, usePalette } from "./palette";
@@ -304,6 +310,17 @@ export function MobileChat({
   const removePendingFile = (index: number) =>
     setPendingFiles((prev) => prev.filter((_, i) => i !== index));
 
+  // Route attachment downloads through the same authenticated helper
+  // the desktop ChatTab uses (downloadChatFile) so platform-scheme
+  // URIs get a real Blob with auth headers instead of about:blank.
+  const downloadAttachment = (att: ChatAttachment) => {
+    downloadChatFile(agentId, att).catch(() => {
+      // AttachmentPreview's own error affordance covers the in-bubble
+      // failure state; matches ChatTab's behaviour of not double-
+      // reporting a download failure.
+    });
+  };
+
   const send = async () => {
     const text = draft.trim();
     if ((!text && pendingFiles.length === 0) || sending || !reachable) return;
@@ -433,7 +450,19 @@ export function MobileChat({
         </div>
       </div>
 
+      {/* Agent Comms — reuse the desktop AgentCommsPanel verbatim so
+          mobile renders the identical peer/A2A + delegation feed
+          (history GET + live socket events) instead of a placeholder
+          (#231). The panel owns its own scroll/load/error/empty
+          states, matching ChatTab's agent-comms tabpanel. */}
+      {tab === "a2a" && (
+        <div style={{ flex: 1, minHeight: 0, overflow: "hidden" }}>
+          <AgentCommsPanel workspaceId={agentId} />
+        </div>
+      )}
+
       {/* Messages */}
+      {tab === "my" && (
       <div
         ref={scrollRef}
         style={{
@@ -445,18 +474,6 @@ export function MobileChat({
           gap: 8,
         }}
       >
-        {tab === "a2a" && (
-          <div
-            style={{
-              padding: "20px 4px",
-              textAlign: "center",
-              color: p.text3,
-              fontSize: 13,
-            }}
-          >
-            Agent Comms — peer-to-peer A2A traffic surfaces in the Comms tab.
-          </div>
-        )}
         {tab === "my" && historyLoading && (
           <div style={{ padding: "20px 4px", textAlign: "center", color: p.text3, fontSize: 13 }}>
             Loading chat history…
@@ -521,9 +538,31 @@ export function MobileChat({
                     overflowWrap: "anywhere",
                   }}
                 >
-                  <MarkdownBubble dark={dark} accent={p.accent}>
-                    {m.content}
-                  </MarkdownBubble>
+                  {m.content && (
+                    <MarkdownBubble dark={dark} accent={p.accent}>
+                      {m.content}
+                    </MarkdownBubble>
+                  )}
+                  {m.attachments && m.attachments.length > 0 && (
+                    <div
+                      style={{
+                        display: "flex",
+                        flexWrap: "wrap",
+                        gap: 4,
+                        marginTop: m.content ? 6 : 0,
+                      }}
+                    >
+                      {m.attachments.map((att, i) => (
+                        <AttachmentPreview
+                          key={`${m.id}-${i}`}
+                          workspaceId={agentId}
+                          attachment={att}
+                          onDownload={downloadAttachment}
+                          tone={mine ? "user" : "agent"}
+                        />
+                      ))}
+                    </div>
+                  )}
                   <div
                     style={{
                       fontSize: 10,
@@ -554,7 +593,13 @@ export function MobileChat({
           </div>
         )}
       </div>
+      )}
 
+      {/* Footer ID + composer belong to My Chat only. The Agent Comms
+          tab is a read-only peer/A2A feed (parity with desktop
+          ChatTab, where the agent-comms tabpanel has no composer). */}
+      {tab === "my" && (
+      <>
       {/* Footer ID */}
       <div
         style={{
@@ -746,6 +791,8 @@ export function MobileChat({
           </button>
         </div>
       </div>
+      </>
+      )}
     </div>
   );
 }

canvas/src/components/mobile/MobileSpawn.tsx

@@ -12,6 +12,7 @@ import { useEffect, useState } from "react";
 
 import { api } from "@/lib/api";
 import { type Template } from "@/lib/deploy-preflight";
+import { isSaaSTenant } from "@/lib/tenant";
 
 import { tierCode } from "./palette";
 import { MOBILE_FONT_MONO, MOBILE_FONT_SANS, type MobilePalette, usePalette } from "./palette";
@@ -26,6 +27,7 @@ const TIER_LABEL: Record<"T1" | "T2" | "T3" | "T4", string> = {
 
 export function MobileSpawn({ dark, onClose }: { dark: boolean; onClose: () => void }) {
   const p = usePalette(dark);
+  const isSaaS = isSaaSTenant();
   const [templates, setTemplates] = useState<Template[]>([]);
   const [loadingTemplates, setLoadingTemplates] = useState(true);
   const [tplId, setTplId] = useState<string | null>(null);
@@ -43,7 +45,7 @@ export function MobileSpawn({ dark, onClose }: { dark: boolean; onClose: () => v
         setTemplates(list);
         if (list.length > 0) {
           setTplId(list[0].id);
-          setTier(tierCode(list[0].tier));
+          setTier(isSaaS ? "T4" : tierCode(list[0].tier));
         }
       })
       .catch(() => {
@@ -55,7 +57,7 @@ export function MobileSpawn({ dark, onClose }: { dark: boolean; onClose: () => v
     return () => {
       cancelled = true;
     };
-  }, []);
+  }, [isSaaS]);
 
   const handleSpawn = async () => {
     if (busy || !tplId) return;
@@ -67,7 +69,7 @@ export function MobileSpawn({ dark, onClose }: { dark: boolean; onClose: () => v
       await api.post<{ id: string }>("/workspaces", {
         name: (name.trim() || chosen.name),
         template: chosen.id,
-        tier: Number(tier.slice(1)),
+        tier: isSaaS ? 4 : Number(tier.slice(1)),
         canvas: {
           x: Math.random() * 400 + 100,
           y: Math.random() * 300 + 100,
@@ -203,7 +205,7 @@ export function MobileSpawn({ dark, onClose }: { dark: boolean; onClose: () => v
             >
               {templates.map((t) => {
                 const on = tplId === t.id;
-                const tCode = tierCode(t.tier);
+                const tCode = isSaaS ? "T4" : tierCode(t.tier);
                 return (
                   <button
                     key={t.id}

canvas/src/components/mobile/__tests__/MobileChat.test.tsx

@@ -21,6 +21,14 @@ import { MobileChat } from "../MobileChat";
 vi.mock("@/lib/api");
 import { api } from "@/lib/api";
 
+// AgentCommsPanel (mounted by the Agent Comms sub-tab, #231) subscribes
+// to the global socket via useSocketEvent. Stub it to a no-op so the
+// panel mounts without the real ReconnectingSocket — the parity tests
+// only assert the panel renders (vs the old static placeholder).
+vi.mock("@/hooks/useSocketEvent", () => ({
+  useSocketEvent: vi.fn(),
+}));
+
 // ─── Mock store ───────────────────────────────────────────────────────────────
 
 const mockAgentId = "ws-chat-test";
@@ -155,6 +163,12 @@ beforeEach(() => {
   mockOnBack.mockClear();
   mockStoreState.nodes = [];
   mockStoreState.agentMessages = {};
+  // jsdom doesn't implement scrollIntoView. The Agent Comms tab now
+  // mounts AgentCommsPanel (#231), which scrolls its feed to bottom on
+  // arrival; a no-op stub keeps the panel from throwing under jsdom
+  // (same stub AgentCommsPanel's own render test installs).
+  Element.prototype.scrollIntoView =
+    vi.fn() as unknown as Element["scrollIntoView"];
   // Set up spies on the real api methods. Tests override these per-call.
   const getSpy = vi.spyOn(api, "get");
   const postSpy = vi.spyOn(api, "post");
@@ -474,3 +488,146 @@ describe("MobileChat — chat history", () => {
     expect(getSpy).toHaveBeenCalledTimes(2);
   });
 });
+
+// ─── #232 · Attachment render parity with desktop ChatTab ────────────────────
+//
+// Regression for the CTO-reported mobile bug: MobileChat used to render
+// only m.content (no attachment surface), so files sent/received in a
+// conversation were invisible on mobile while desktop showed them. The
+// fix routes m.attachments through the same AttachmentPreview the
+// desktop ChatTab bubble uses.
+
+describe("MobileChat — attachment render parity (#232)", () => {
+  beforeEach(() => {
+    mockStoreState.nodes = [onlineNode];
+  });
+
+  it("renders an attachment from a history message via AttachmentPreview", async () => {
+    const getSpy = vi.spyOn(api, "get");
+    // useChatHistory reads { messages, reached_end }.
+    getSpy.mockResolvedValueOnce({
+      messages: [
+        {
+          id: "m-att-1",
+          role: "agent",
+          content: "Here is the report",
+          attachments: [
+            {
+              name: "report.csv",
+              uri: "workspace://out/report.csv",
+              mimeType: "text/csv",
+              size: 2048,
+            },
+          ],
+          timestamp: new Date().toISOString(),
+        },
+      ],
+      reached_end: true,
+    });
+
+    let rr: ReturnType<typeof renderChat>;
+    await act(async () => {
+      rr = renderChat(mockAgentId);
+    });
+    const { container } = rr!;
+
+    // A non-image attachment renders the AttachmentChip download button
+    // with title="Download <name>" — same component the desktop bubble
+    // dispatches through AttachmentPreview.
+    await waitFor(() => {
+      const chip = container.querySelector('[title="Download report.csv"]');
+      expect(chip).toBeTruthy();
+    });
+    expect(container.textContent ?? "").toContain("report.csv");
+  });
+});
+
+// ─── #231 · Agent Comms (A2A/peer) render parity with desktop ChatTab ────────
+//
+// Regression for the CTO-reported mobile bug: the Agent Comms sub-tab
+// rendered a static placeholder string ("peer-to-peer A2A traffic
+// surfaces in the Comms tab") instead of the real feed. The fix mounts
+// the same AgentCommsPanel the desktop ChatTab agent-comms tabpanel
+// uses, so peer/A2A + delegation activity is visible on mobile.
+
+describe("MobileChat — Agent Comms render parity (#231)", () => {
+  beforeEach(() => {
+    mockStoreState.nodes = [onlineNode];
+  });
+
+  it("mounts AgentCommsPanel on the Agent Comms tab (not the old placeholder)", async () => {
+    const getSpy = vi.spyOn(api, "get");
+    // 1st GET: useChatHistory (My Chat) on mount.
+    getSpy.mockResolvedValueOnce({ messages: [], reached_end: true });
+    // 2nd GET: AgentCommsPanel's activity load when the tab is shown.
+    // Empty list → panel renders its own empty state, which still
+    // proves AgentCommsPanel mounted (vs. the removed placeholder).
+    getSpy.mockResolvedValueOnce([]);
+
+    let rr: ReturnType<typeof renderChat>;
+    await act(async () => {
+      rr = renderChat(mockAgentId);
+    });
+    const { container } = rr!;
+
+    const commsTab = Array.from(container.querySelectorAll("button")).find(
+      (b) => b.textContent?.trim() === "Agent Comms",
+    );
+    expect(commsTab).toBeTruthy();
+    await act(async () => {
+      commsTab!.click();
+    });
+
+    await waitFor(() => {
+      const text = container.textContent ?? "";
+      // The panel's empty state — proves AgentCommsPanel mounted.
+      expect(text).toContain("No agent-to-agent communications yet.");
+    });
+    // The old hard-coded placeholder must be gone.
+    expect(container.textContent ?? "").not.toContain(
+      "peer-to-peer A2A traffic surfaces in the Comms tab",
+    );
+    // The panel hit its activity endpoint.
+    expect(getSpy).toHaveBeenCalledWith(
+      expect.stringContaining(`/workspaces/${mockAgentId}/activity`),
+    );
+  });
+
+  it("renders a peer message on the Agent Comms tab", async () => {
+    const getSpy = vi.spyOn(api, "get");
+    getSpy.mockResolvedValueOnce({ messages: [], reached_end: true });
+    // a2a_receive from a peer → AgentCommsPanel.toCommMessage maps it
+    // to an inbound bubble with the request text.
+    getSpy.mockResolvedValueOnce([
+      {
+        id: "act-1",
+        activity_type: "a2a_receive",
+        source_id: "peer-ws-uuid",
+        target_id: mockAgentId,
+        method: "message/send",
+        summary: "peer asked something",
+        request_body: { task: "Please review PR 42" },
+        response_body: null,
+        status: "ok",
+        created_at: new Date().toISOString(),
+      },
+    ]);
+
+    let rr: ReturnType<typeof renderChat>;
+    await act(async () => {
+      rr = renderChat(mockAgentId);
+    });
+    const { container } = rr!;
+
+    const commsTab = Array.from(container.querySelectorAll("button")).find(
+      (b) => b.textContent?.trim() === "Agent Comms",
+    );
+    await act(async () => {
+      commsTab!.click();
+    });
+
+    await waitFor(() => {
+      expect(container.textContent ?? "").toContain("Please review PR 42");
+    });
+  });
+});

canvas/src/components/settings/SecretRow.tsx

@@ -3,16 +3,24 @@ import { useState, useCallback, useRef, useEffect } from 'react';
 import type { Secret, SecretGroup } from '@/types/secrets';
 import { useSecretsStore } from '@/stores/secrets-store';
 import { StatusBadge } from '@/components/ui/StatusBadge';
-import { RevealToggle } from '@/components/ui/RevealToggle';
 import { KeyValueField } from '@/components/ui/KeyValueField';
 import { ValidationHint } from '@/components/ui/ValidationHint';
 import { TestConnectionButton } from '@/components/ui/TestConnectionButton';
 import { validateSecretValue } from '@/lib/validation/secret-formats';
 import { SERVICES } from '@/lib/services';
 
-const AUTO_HIDE_MS = 30_000;
 const VALIDATION_DEBOUNCE_MS = 400;
 
+// Secret values are write-only from the browser: the server List endpoint
+// "Never exposes values", there is no per-secret decrypt route, and the
+// only decrypted path (GET /secrets/values) is bulk + token-gated for
+// remote agents. The old eye/RevealToggle was a dead affordance — it
+// flipped its own icon but could never reveal anything, which read as
+// "this doesn't work" (esp. once clicked → eye-with-slash). We show an
+// honest static indicator instead; rotation is via Edit.
+const WRITE_ONLY_TITLE =
+  'Value is write-only and cannot be revealed — use Edit to replace/rotate it';
+
 interface SecretRowProps {
   secret: Secret;
   workspaceId: string;
@@ -31,28 +39,12 @@ export function SecretRow({ secret, workspaceId }: SecretRowProps) {
   const setSecretStatus = useSecretsStore((s) => s.setSecretStatus);
 
   const isEditing = editingKey === secret.name;
-  const [revealed, setRevealed] = useState(false);
   const [editValue, setEditValue] = useState('');
   const [validationError, setValidationError] = useState<string | null>(null);
   const [isSaving, setIsSaving] = useState(false);
   const [saveError, setSaveError] = useState<string | null>(null);
   const debounceRef = useRef<ReturnType<typeof setTimeout>>(undefined);
   const editBtnRef = useRef<HTMLButtonElement>(null);
-  const revealTimerRef = useRef<ReturnType<typeof setTimeout>>(undefined);
-
-  // Auto-hide revealed value after 30s
-  useEffect(() => {
-    if (revealed) {
-      clearTimeout(revealTimerRef.current);
-      revealTimerRef.current = setTimeout(() => setRevealed(false), AUTO_HIDE_MS);
-      return () => clearTimeout(revealTimerRef.current);
-    }
-  }, [revealed]);
-
-  // Reset revealed state when panel closes (session-only)
-  useEffect(() => {
-    return () => setRevealed(false);
-  }, []);
 
   // Debounced validation
   useEffect(() => {
@@ -133,11 +125,15 @@ export function SecretRow({ secret, workspaceId }: SecretRowProps) {
           {secret.masked_value}
         </span>
         <div className="secret-row__actions">
-          <RevealToggle
-            revealed={revealed}
-            onToggle={() => setRevealed((r) => !r)}
-            label={`Toggle reveal ${secret.name}`}
-          />
+          <span
+            data-testid="write-only-indicator"
+            className="secret-row__write-only"
+            role="img"
+            aria-label={`${secret.name} value is write-only and cannot be revealed; use Edit to replace it`}
+            title={WRITE_ONLY_TITLE}
+          >
+            🔒
+          </span>
           <StatusBadge status={secret.status} />
           <button
             type="button"

canvas/src/components/settings/TokensTab.tsx

@@ -16,7 +16,40 @@ interface TokensTabProps {
   workspaceId: string;
 }
 
+// The settings panel passes the literal sentinel "global" when no canvas
+// node is selected. Workspace tokens are inherently per-workspace — there
+// is no /workspaces/global/tokens endpoint (querying the uuid column with
+// "global" 500s on Postgres). The org-wide equivalent lives in the
+// separate "Org API Keys" tab. Mirrors the sentinel-awareness that
+// api/secrets.ts already has (workspaceId === 'global' → /settings/secrets).
+const GLOBAL_WORKSPACE_ID = 'global';
+
 export function TokensTab({ workspaceId }: TokensTabProps) {
+  if (workspaceId === GLOBAL_WORKSPACE_ID) {
+    return (
+      <div className="p-4 space-y-4">
+        <div>
+          <h3 className="text-sm font-semibold text-ink">API Tokens</h3>
+          <p className="text-[10px] text-ink-mid mt-0.5">
+            Bearer tokens for authenticating API calls to this workspace.
+          </p>
+        </div>
+        <div className="text-center py-6">
+          <p className="text-xs text-ink-mid">Select a workspace node first</p>
+          <p className="text-[10px] text-ink-mid mt-1">
+            Workspace tokens are scoped to a single workspace. Select a node
+            on the canvas to manage its tokens, or use the{' '}
+            <span className="text-accent font-medium">Org API Keys</span> tab
+            for org-wide API keys.
+          </p>
+        </div>
+      </div>
+    );
+  }
+  return <WorkspaceTokensTab workspaceId={workspaceId} />;
+}
+
+function WorkspaceTokensTab({ workspaceId }: TokensTabProps) {
   const [tokens, setTokens] = useState<Token[]>([]);
   const [loading, setLoading] = useState(true);
   const [creating, setCreating] = useState(false);

canvas/src/components/settings/__tests__/SecretRow.test.tsx

@@ -138,14 +138,54 @@ describe("SecretRow — display mode", () => {
     expect(document.querySelector('[role="row"]')).toBeTruthy();
   });
 
-  it("has Reveal, Copy, Edit, Delete buttons", () => {
+  it("has Copy, Edit, Delete buttons", () => {
     render(<SecretRow secret={GITHUB_SECRET} workspaceId="ws-1" />);
-    expect(screen.getByTestId("reveal-toggle")).toBeTruthy();
     expect(screen.getByRole("button", { name: /copy/i })).toBeTruthy();
     expect(screen.getByRole("button", { name: /edit/i })).toBeTruthy();
     expect(screen.getByRole("button", { name: /delete/i })).toBeTruthy();
   });
 
+  // Regression: the reveal/eye control was a dead affordance. Clicking it
+  // flipped its own icon (eye → eye-with-slash) but never revealed the value,
+  // because secret values are write-only from the browser (server List
+  // "Never exposes values"; there is no per-secret decrypt endpoint and the
+  // client has no plaintext-fetch function). The honest fix removes the
+  // toggle and shows a static "write-only / cannot be revealed" indicator.
+  // See internal tracking issue + internal#210/#211.
+  it("does NOT render a reveal/eye toggle (values are write-only)", () => {
+    render(<SecretRow secret={GITHUB_SECRET} workspaceId="ws-1" />);
+    expect(screen.queryByTestId("reveal-toggle")).toBeNull();
+    expect(
+      screen.queryByRole("button", { name: /toggle reveal/i }),
+    ).toBeNull();
+  });
+
+  it("shows a write-only indicator explaining the value cannot be revealed", () => {
+    render(<SecretRow secret={ANTHROPIC_SECRET} workspaceId="ws-1" />);
+    const indicator = screen.getByTestId("write-only-indicator");
+    expect(indicator).toBeTruthy();
+    // Affordance must be honest: explain it cannot be revealed and that
+    // Edit is the rotate path. It must not be a clickable button.
+    const title = indicator.getAttribute("title") ?? "";
+    expect(title.toLowerCase()).toMatch(/write-only|cannot be revealed/);
+    expect(indicator.tagName).not.toBe("BUTTON");
+  });
+
+  it("write-only indicator is present for the Anthropic/OAuth-token row too", () => {
+    // The reported bug singled out CLAUDE_CODE_OAUTH_TOKEN (anthropic group);
+    // the fix is group-agnostic — every row gets the same honest affordance.
+    const OAUTH_SECRET = {
+      name: "CLAUDE_CODE_OAUTH_TOKEN",
+      masked_value: "••••••••••••••••9d2a",
+      group: "anthropic" as const,
+      status: "unverified" as const,
+      updated_at: "2024-01-04",
+    };
+    render(<SecretRow secret={OAUTH_SECRET} workspaceId="ws-1" />);
+    expect(screen.queryByTestId("reveal-toggle")).toBeNull();
+    expect(screen.getByTestId("write-only-indicator")).toBeTruthy();
+  });
+
   it("shows invalid status correctly", () => {
     render(<SecretRow secret={CUSTOM_SECRET} workspaceId="ws-1" />);
     expect(screen.getByTestId("status-badge").getAttribute("data-status")).toBe("invalid");

canvas/src/components/settings/__tests__/TokensTab.test.tsx

@@ -302,3 +302,35 @@ describe("TokensTab — error", () => {
     expect(document.querySelector('[role="status"]')).toBeNull();
   });
 });
+
+// ─── "global" sentinel (no node selected) ────────────────────────────────────
+//
+// Regression: SettingsPanel passes the literal "global" when no canvas
+// node is selected. workspace tokens are per-workspace and there is no
+// /workspaces/global/tokens endpoint — calling it 500'd
+// ("invalid input syntax for type uuid: global"). The tab must NOT call
+// the API in that state and must point the user at the Org API Keys tab.
+describe("TokensTab — global sentinel (no node selected)", () => {
+  beforeEach(() => {
+    mockApiGet.mockReset();
+    mockApiPost.mockReset();
+    mockApiGet.mockRejectedValue(new Error("should not be called"));
+  });
+
+  it("does not call the API and shows a pointer to Org API Keys", async () => {
+    render(<TokensTab workspaceId="global" />);
+    await flush();
+    expect(mockApiGet).not.toHaveBeenCalled();
+    expect(mockApiPost).not.toHaveBeenCalled();
+    expect(document.body.textContent).toContain("Select a workspace node");
+    expect(document.body.textContent).toContain("Org API Keys");
+    // No error banner, no scary 500 surfacing.
+    expect(document.querySelector(".text-bad")).toBeNull();
+  });
+
+  it("has no create button in the global state", async () => {
+    render(<TokensTab workspaceId="global" />);
+    await flush();
+    expect(document.body.textContent).not.toContain("New Token");
+  });
+});

canvas/src/components/tabs/chat/__tests__/message-parser.test.ts

@@ -248,6 +248,88 @@ describe("extractResponseText", () => {
   });
 });
 
+describe("extractAgentText", () => {
+  it("extracts text from top-level parts", () => {
+    const task = {
+      parts: [{ kind: "text", text: "Agent said hello" }],
+    };
+    expect(extractAgentText(task)).toBe("Agent said hello");
+  });
+
+  it("extracts from artifacts[0].parts when top-level parts absent", () => {
+    const task = {
+      artifacts: [
+        { parts: [{ kind: "text", text: "From artifact block" }] },
+      ],
+    };
+    expect(extractAgentText(task)).toBe("From artifact block");
+  });
+
+  it("extracts from status.message.parts as fallback", () => {
+    const task = {
+      status: {
+        message: { parts: [{ kind: "text", text: "Status text" }] },
+      },
+    };
+    expect(extractAgentText(task)).toBe("Status text");
+  });
+
+  it("prefers top-level parts over artifacts", () => {
+    const task = {
+      parts: [{ kind: "text", text: "top-level wins" }],
+      artifacts: [
+        { parts: [{ kind: "text", text: "artifact text" }] },
+      ],
+    };
+    expect(extractAgentText(task)).toBe("top-level wins");
+  });
+
+  it("prefers top-level parts over status.message", () => {
+    const task = {
+      parts: [{ kind: "text", text: "parts wins" }],
+      status: {
+        message: { parts: [{ kind: "text", text: "status text" }] },
+      },
+    };
+    expect(extractAgentText(task)).toBe("parts wins");
+  });
+
+  it("returns string identity when task itself is a string", () => {
+    expect(extractAgentText("plain string task" as unknown as Record<string, unknown>)).toBe(
+      "plain string task",
+    );
+  });
+
+  it("returns fallback when task is an empty object", () => {
+    expect(extractAgentText({})).toBe("(Could not extract response text)");
+  });
+
+  it("returns fallback when task has no extractable text", () => {
+    expect(
+      extractAgentText({ status: "running", other: "fields" }),
+    ).toBe("(Could not extract response text)");
+  });
+
+  it("tolerates malformed nested shapes without throwing", () => {
+    const task = {
+      parts: null,
+      artifacts: "not an array",
+      status: { message: 42 },
+    };
+    expect(extractAgentText(task)).toBe("(Could not extract response text)");
+  });
+
+  it("joins multiple text parts with newline", () => {
+    const task = {
+      parts: [
+        { kind: "text", text: "Line one" },
+        { kind: "text", text: "Line two" },
+      ],
+    };
+    expect(extractAgentText(task)).toBe("Line one\nLine two");
+  });
+});
+
 describe("extractTextsFromParts", () => {
   it("extracts text parts with kind=text", () => {
     const parts = [

canvas/src/components/tabs/chat/__tests__/resolveWorkspaceName.test.ts

@@ -0,0 +1,102 @@
+import { describe, it, expect, beforeEach } from "vitest";
+import { useCanvasStore } from "@/store/canvas";
+import { resolveWorkspaceName } from "../hooks/resolveWorkspaceName";
+
+beforeEach(() => {
+  // Reset store to a clean slate between tests so node lookup is deterministic.
+  useCanvasStore.setState({ nodes: [] });
+});
+
+describe("resolveWorkspaceName", () => {
+  it("returns the workspace name when a node with that ID exists", () => {
+    useCanvasStore.setState({
+      nodes: [
+        {
+          id: "ws-alpha-001",
+          type: "workspace",
+          data: { name: "Alpha Agent" },
+          position: { x: 0, y: 0 },
+        },
+      ],
+    });
+
+    expect(resolveWorkspaceName("ws-alpha-001")).toBe("Alpha Agent");
+  });
+
+  it("falls back to the first 8 chars of the ID when no matching node exists", () => {
+    expect(resolveWorkspaceName("ws-zzz-not-found")).toBe("ws-zzz-n");
+  });
+
+  it("falls back to the first 8 chars when the node exists but has no name", () => {
+    useCanvasStore.setState({
+      nodes: [
+        {
+          id: "ws-no-name",
+          type: "workspace",
+          // data.name is deliberately absent
+          data: {},
+          position: { x: 0, y: 0 },
+        },
+      ],
+    });
+
+    expect(resolveWorkspaceName("ws-no-name")).toBe("ws-no-na");
+  });
+
+  it("returns the first 8 chars for a very short ID", () => {
+    expect(resolveWorkspaceName("ab")).toBe("ab");
+  });
+
+  it("returns the first 8 chars when the ID is exactly 8 characters", () => {
+    // slice(0,8) of an 8-char string is the full string
+    const id = "12345678";
+    expect(resolveWorkspaceName(id)).toBe(id);
+  });
+
+  it("picks the right node when multiple workspaces share a prefix", () => {
+    useCanvasStore.setState({
+      nodes: [
+        {
+          id: "00000000-0000-0000-0000-000000000001",
+          type: "workspace",
+          data: { name: "Backend Agent" },
+          position: { x: 0, y: 0 },
+        },
+        {
+          id: "00000000-0000-0000-0000-000000000002",
+          type: "workspace",
+          data: { name: "Frontend Agent" },
+          position: { x: 100, y: 0 },
+        },
+      ],
+    });
+
+    expect(resolveWorkspaceName("00000000-0000-0000-0000-000000000002")).toBe(
+      "Frontend Agent"
+    );
+    expect(resolveWorkspaceName("00000000-0000-0000-0000-000000000001")).toBe(
+      "Backend Agent"
+    );
+  });
+
+  it("does not mutate store state between calls", () => {
+    useCanvasStore.setState({
+      nodes: [
+        {
+          id: "stable-id",
+          type: "workspace",
+          data: { name: "Stable Workspace" },
+          position: { x: 0, y: 0 },
+        },
+      ],
+    });
+
+    resolveWorkspaceName("stable-id");
+    resolveWorkspaceName("unknown-id");
+
+    // Store nodes must be unchanged — resolveWorkspaceName is read-only.
+    const nodes = useCanvasStore.getState().nodes;
+    expect(nodes).toHaveLength(1);
+    expect((nodes[0] as { id: string }).id).toBe("stable-id");
+  });
+});

canvas/src/components/ui/TestConnectionButton.tsx

@@ -2,7 +2,7 @@
 
 import { useState, useCallback, useRef, useEffect } from 'react';
 import type { TestConnectionState, SecretGroup } from '@/types/secrets';
-import { validateSecret } from '@/lib/api/secrets';
+import { validateSecret, ApiError } from '@/lib/api/secrets';
 
 interface TestConnectionButtonProps {
   provider: SecretGroup;
@@ -55,9 +55,23 @@ export function TestConnectionButton({
       }
       onResult?.(result.valid);
       resetTimerRef.current = setTimeout(() => setState('idle'), RESET_DELAYS[nextState]!);
-    } catch {
+    } catch (err) {
+      // Distinguish a real failure shape rather than always claiming a
+      // timeout. A reachable server that answered with an HTTP status
+      // (ApiError) did NOT time out — most commonly the validation route
+      // is not available (404/501), which must not masquerade as
+      // "service down". Only an actual thrown network/abort error is a
+      // connectivity failure.
       setState('failure');
-      setErrorDetail('Connection timed out. Service may be down.');
+      if (err instanceof ApiError) {
+        setErrorDetail(
+          err.status === 404 || err.status === 501
+            ? 'Key validation is not available for this service yet. The key was not tested.'
+            : `Could not verify key (server returned ${err.status}). Saving is unaffected.`,
+        );
+      } else {
+        setErrorDetail('Could not reach the validation service. Check your connection and try again.');
+      }
       onResult?.(false);
       resetTimerRef.current = setTimeout(() => setState('idle'), RESET_DELAYS.failure);
     }

canvas/src/components/ui/__tests__/TestConnectionButton.test.tsx

@@ -28,8 +28,20 @@ const mockValidateSecret = vi.fn();
 
 vi.mock("@/lib/api/secrets", () => ({
   validateSecret: (...args: unknown[]) => mockValidateSecret(...args),
+  ApiError: class ApiError extends Error {
+    status: number;
+    constructor(status: number, message: string) {
+      super(message);
+      this.name = "ApiError";
+      this.status = status;
+    }
+  },
 }));
 
+// Re-import the mocked ApiError so test cases construct the same class the
+// component's `instanceof` check sees.
+import { ApiError } from "@/lib/api/secrets";
+
 beforeEach(() => {
   vi.useFakeTimers();
   vi.clearAllMocks();
@@ -201,8 +213,27 @@ describe("TestConnectionButton — failure path", () => {
 });
 
 describe("TestConnectionButton — catch path", () => {
-  it("shows 'Connection timed out' on network error", async () => {
-    mockValidateSecret.mockRejectedValue(new Error("timeout"));
+  it("does NOT claim a timeout when the validate endpoint 404s (regression: internal#492)", async () => {
+    // The validate route is unimplemented on the server and returns a fast
+    // 404. Before the fix this rendered the misleading hardcoded string
+    // "Connection timed out. Service may be down." It must instead state
+    // honestly that validation isn't available and the key was not tested.
+    mockValidateSecret.mockRejectedValue(new ApiError(404, "Not Found"));
+    render(
+      <TestConnectionButton provider="anthropic" secretValue="sk-ant-xxx" />,
+    );
+    fireEvent.click(document.querySelector('button[type="button"]')!);
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(0);
+    });
+    expect(document.body.textContent).not.toContain("Connection timed out");
+    expect(document.body.textContent).not.toContain("Service may be down");
+    expect(document.body.textContent).toContain("not available");
+    expect(document.body.textContent).toContain("not tested");
+  });
+
+  it("reports a non-404 server error with its status, not a timeout", async () => {
+    mockValidateSecret.mockRejectedValue(new ApiError(500, "Internal Server Error"));
     render(
       <TestConnectionButton provider="github" secretValue="ghp_xxx" />,
     );
@@ -210,7 +241,20 @@ describe("TestConnectionButton — catch path", () => {
     await act(async () => {
       await vi.advanceTimersByTimeAsync(0);
     });
-    expect(document.body.textContent).toContain("Connection timed out");
+    expect(document.body.textContent).toContain("500");
+    expect(document.body.textContent).not.toContain("Connection timed out");
+  });
+
+  it("shows a connectivity message on a genuine network error", async () => {
+    mockValidateSecret.mockRejectedValue(new Error("network down"));
+    render(
+      <TestConnectionButton provider="github" secretValue="ghp_xxx" />,
+    );
+    fireEvent.click(document.querySelector('button[type="button"]')!);
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(0);
+    });
+    expect(document.body.textContent).toContain("Could not reach the validation service");
   });
 
   it("calls onResult(false) on network error", async () => {

canvas/src/hooks/useTemplateDeploy.tsx

@@ -8,6 +8,7 @@ import {
   type PreflightResult,
   type Template,
 } from "@/lib/deploy-preflight";
+import { isSaaSTenant } from "@/lib/tenant";
 import { MissingKeysModal } from "@/components/MissingKeysModal";
 
 /**
@@ -105,7 +106,7 @@ export function useTemplateDeploy(
         const ws = await api.post<{ id: string }>("/workspaces", {
           name: template.name,
           template: template.id,
-          tier: template.tier,
+          tier: isSaaSTenant() ? 4 : template.tier,
           canvas: coords,
           ...(model ? { model } : {}),
         });

canvas/src/lib/__tests__/design-tokens.test.ts

@@ -1,102 +0,0 @@
-// @vitest-environment jsdom
-/**
- * Tests for design-tokens.ts constant exports.
- *
- * STATUS_CONFIG is tested here directly rather than inside
- * statusDotClass.test.ts so the constant's full shape (dot, glow, label,
- * bar per key) is explicitly asserted — not just indirectly via the
- * statusDotClass helper that consumes its .dot field.
- */
-import { describe, it, expect } from "vitest";
-import { STATUS_CONFIG } from "../design-tokens";
-
-const ALL_STATUS_KEYS = [
-  "online",
-  "offline",
-  "paused",
-  "degraded",
-  "failed",
-  "provisioning",
-  "not_configured",
-] as const;
-
-describe("STATUS_CONFIG", () => {
-  it("has exactly the expected status keys and no extras", () => {
-    const actual = Object.keys(STATUS_CONFIG).sort();
-    const expected = [...ALL_STATUS_KEYS].sort();
-    expect(actual).toEqual(expected);
-  });
-
-  it("every entry has dot, glow, label, and bar fields", () => {
-    for (const key of ALL_STATUS_KEYS) {
-      const entry = STATUS_CONFIG[key];
-      expect(entry, `entry for "${key}"`).toHaveProperty("dot");
-      expect(entry, `entry for "${key}"`).toHaveProperty("glow");
-      expect(entry, `entry for "${key}"`).toHaveProperty("label");
-      expect(entry, `entry for "${key}"`).toHaveProperty("bar");
-    }
-  });
-
-  it("dot, glow, label, bar are all non-empty strings", () => {
-    for (const key of ALL_STATUS_KEYS) {
-      const entry = STATUS_CONFIG[key];
-      for (const field of ["dot", "glow", "label", "bar"] as const) {
-        expect(typeof entry[field], `"${key}".${field}`).toBe("string");
-        // label must be non-empty; others may be empty (e.g. offline.glow = "").
-        if (field === "label") {
-          expect(entry[field].length, `"${key}".${field}`).toBeGreaterThan(0);
-        }
-      }
-    }
-  });
-
-  it('online: dot is emerald, glow is set, label is "Online"', () => {
-    expect(STATUS_CONFIG.online.dot).toBe("bg-emerald-400");
-    expect(STATUS_CONFIG.online.glow).toBe("shadow-emerald-400/50");
-    expect(STATUS_CONFIG.online.label).toBe("Online");
-    expect(STATUS_CONFIG.online.bar).toBe("from-emerald-500/20 to-transparent");
-  });
-
-  it('offline: dot is zinc, glow is empty, label is "Offline"', () => {
-    expect(STATUS_CONFIG.offline.dot).toBe("bg-zinc-500");
-    expect(STATUS_CONFIG.offline.glow).toBe("");
-    expect(STATUS_CONFIG.offline.label).toBe("Offline");
-    expect(STATUS_CONFIG.offline.bar).toBe("from-zinc-600/10 to-transparent");
-  });
-
-  it('paused: dot is indigo, label is "Paused"', () => {
-    expect(STATUS_CONFIG.paused.dot).toBe("bg-indigo-400");
-    expect(STATUS_CONFIG.paused.glow).toBe("");
-    expect(STATUS_CONFIG.paused.label).toBe("Paused");
-  });
-
-  it('degraded: dot is amber with glow, label is "Degraded"', () => {
-    expect(STATUS_CONFIG.degraded.dot).toBe("bg-amber-400");
-    expect(STATUS_CONFIG.degraded.glow).toBe("shadow-amber-400/50");
-    expect(STATUS_CONFIG.degraded.label).toBe("Degraded");
-  });
-
-  it('failed: dot is red with glow, label is "Failed"', () => {
-    expect(STATUS_CONFIG.failed.dot).toBe("bg-red-400");
-    expect(STATUS_CONFIG.failed.glow).toBe("shadow-red-400/50");
-    expect(STATUS_CONFIG.failed.label).toBe("Failed");
-  });
-
-  it('provisioning: dot is sky with pulse animation, label is "Starting"', () => {
-    expect(STATUS_CONFIG.provisioning.dot).toBe("bg-sky-400 motion-safe:animate-pulse");
-    expect(STATUS_CONFIG.provisioning.glow).toBe("shadow-sky-400/50");
-    expect(STATUS_CONFIG.provisioning.label).toBe("Starting");
-  });
-
-  it('not_configured: dot is amber-300 with glow, label is "Not configured"', () => {
-    expect(STATUS_CONFIG.not_configured.dot).toBe("bg-amber-300");
-    expect(STATUS_CONFIG.not_configured.glow).toBe("shadow-amber-300/50");
-    expect(STATUS_CONFIG.not_configured.label).toBe("Not configured");
-  });
-
-  it("is a frozen static map — same key always returns same object reference", () => {
-    for (const key of ALL_STATUS_KEYS) {
-      expect(STATUS_CONFIG[key]).toBe(STATUS_CONFIG[key]);
-    }
-  });
-});

canvas/src/lib/__tests__/theme.test.ts

@@ -1,60 +0,0 @@
-// @vitest-environment jsdom
-/**
- * Tests for theme.ts — cssVar() function and ColorToken type.
- */
-import { describe, it, expect } from "vitest";
-import { cssVar, type ColorToken } from "../theme";
-
-describe("cssVar", () => {
-  it("wraps each known token in a var() reference", () => {
-    const tokens: ColorToken[] = [
-      "surface",
-      "surface-elevated",
-      "surface-sunken",
-      "surface-card",
-      "line",
-      "line-soft",
-      "ink",
-      "ink-mid",
-      "ink-soft",
-      "accent",
-      "accent-strong",
-      "warm",
-      "good",
-      "bad",
-      "bg",
-      "bg-elev",
-      "bg-card",
-      "line-strong",
-      "ink-mute",
-      "ink-dim",
-      "accent-dim",
-      "plasma",
-      "warn",
-    ];
-    for (const token of tokens) {
-      expect(cssVar(token)).toBe(`var(--color-${token})`);
-    }
-  });
-
-  it("is a pure function — same token always returns same value", () => {
-    for (let i = 0; i < 5; i++) {
-      expect(cssVar("accent")).toBe("var(--color-accent)");
-      expect(cssVar("surface")).toBe("var(--color-surface)");
-      expect(cssVar("good")).toBe("var(--color-good)");
-    }
-  });
-
-  it("handles hyphenated tokens correctly", () => {
-    expect(cssVar("surface-elevated")).toBe("var(--color-surface-elevated)");
-    expect(cssVar("line-soft")).toBe("var(--color-line-soft)");
-    expect(cssVar("ink-mute")).toBe("var(--color-ink-mute)");
-  });
-
-  it("produces a value usable as an inline style prop value", () => {
-    const result = cssVar("accent");
-    expect(typeof result).toBe("string");
-    expect(result.startsWith("var(--color-")).toBe(true);
-    expect(result.endsWith(")")).toBe(true);
-  });
-});

manifest.json

@@ -30,10 +30,7 @@
     {"name": "openclaw", "repo": "molecule-ai/molecule-ai-workspace-template-openclaw", "ref": "main"},
     {"name": "codex", "repo": "molecule-ai/molecule-ai-workspace-template-codex", "ref": "main"},
     {"name": "langgraph", "repo": "molecule-ai/molecule-ai-workspace-template-langgraph", "ref": "main"},
-    {"name": "crewai", "repo": "molecule-ai/molecule-ai-workspace-template-crewai", "ref": "main"},
-    {"name": "autogen", "repo": "molecule-ai/molecule-ai-workspace-template-autogen", "ref": "main"},
-    {"name": "deepagents", "repo": "molecule-ai/molecule-ai-workspace-template-deepagents", "ref": "main"},
-    {"name": "gemini-cli", "repo": "molecule-ai/molecule-ai-workspace-template-gemini-cli", "ref": "main"}
+    {"name": "autogen", "repo": "molecule-ai/molecule-ai-workspace-template-autogen", "ref": "main"}
   ],
   "org_templates": [
     {"name": "molecule-dev", "repo": "molecule-ai/molecule-ai-org-template-molecule-dev", "ref": "main"},

tests/e2e/test_peer_visibility_mcp_staging.sh

@@ -0,0 +1,376 @@
+#!/usr/bin/env bash
+# Staging E2E — fresh-provision peer-visibility gate via the LITERAL MCP path.
+#
+# WHY THIS EXISTS
+# ---------------
+# Hermes and OpenClaw were repeatedly reported "fleet-verified / cascade-
+# complete" because the *proxy* signals were green:
+#   - registry-registration + heartbeat (Hermes), and
+#   - model round-trip 200 (OpenClaw).
+# But a freshly-provisioned workspace, asked on canvas "can you see your
+# peers", actually FAILS:
+#   - Hermes: 401 on the molecule MCP `list_peers` call,
+#   - OpenClaw: falls back to native `sessions_list`, sees no platform peers.
+# Tasks #142/#159 were even marked "completed" under this same proxy flaw.
+#
+# This script codifies the LITERAL user-facing path so it can never silently
+# regress: it provisions a brand-new throwaway org + sibling workspaces via
+# the real control-plane provisioning path, then for each runtime that should
+# have platform peer-visibility it drives the EXACT MCP call the canvas agent
+# makes — `POST /workspaces/:id/mcp` JSON-RPC tools/call name=list_peers,
+# authenticated by that workspace's own bearer token through the real
+# WorkspaceAuth + MCPRateLimiter middleware chain. It then asserts:
+#   (1) HTTP 200,
+#   (2) JSON-RPC `result` present (NOT an `error` object — a -32000
+#       "tool call failed" or a 401 from WorkspaceAuth fails here),
+#   (3) the returned peer set CONTAINS the other provisioned sibling
+#       workspace IDs — not an empty list, not a native-sessions fallback.
+#
+# This is NOT a proxy. It does not look at a registry row, /health, the
+# heartbeat table, or `GET /registry/:id/peers`. It drives the byte-for-byte
+# JSON-RPC envelope that mcp_molecule_list_peers issues from a real agent.
+#
+# It is written to FAIL on today's broken Hermes/OpenClaw behavior and go
+# green only when the in-flight root-cause fixes (Hermes-401, OpenClaw MCP
+# wiring) actually land. That is the point: it is the objective proof gate.
+#
+# AUTH MODEL (mirrors tests/e2e/test_staging_full_saas.sh)
+# --------------------------------------------------------
+#   Single MOLECULE_ADMIN_TOKEN (= CP_ADMIN_API_TOKEN on Railway staging)
+#   drives: POST /cp/admin/orgs (provision), GET
+#   /cp/admin/orgs/:slug/admin-token (per-tenant token), DELETE
+#   /cp/admin/tenants/:slug (teardown). The per-tenant admin token drives
+#   tenant workspace creation; each workspace's OWN auth_token (returned by
+#   POST /workspaces) drives its MCP call.
+#
+# Required env:
+#   MOLECULE_ADMIN_TOKEN   CP admin bearer — Railway staging CP_ADMIN_API_TOKEN
+# Optional env:
+#   MOLECULE_CP_URL        default https://staging-api.moleculesai.app
+#   E2E_RUN_ID             slug suffix; CI passes ${GITHUB_RUN_ID}
+#   PV_RUNTIMES            space list; default "hermes openclaw claude-code"
+#   E2E_PROVISION_TIMEOUT_SECS  default 1800 (hermes/openclaw cold EC2 budget)
+#   E2E_MINIMAX_API_KEY / E2E_ANTHROPIC_API_KEY / E2E_OPENAI_API_KEY
+#                          LLM provider key injected so the runtime can boot
+#   E2E_KEEP_ORG           1 → skip teardown (local debugging only)
+#
+# Exit codes:
+#   0  every runtime saw its peers via the literal MCP call
+#   1  generic failure
+#   2  missing required env
+#   3  provisioning timed out
+#   4  teardown left orphan resources
+#   10 peer-visibility regression reproduced (the gate firing as designed)
+
+set -uo pipefail
+
+CP_URL="${MOLECULE_CP_URL:-https://staging-api.moleculesai.app}"
+ADMIN_TOKEN="${MOLECULE_ADMIN_TOKEN:?MOLECULE_ADMIN_TOKEN required — Railway staging CP_ADMIN_API_TOKEN}"
+RUN_ID_SUFFIX="${E2E_RUN_ID:-$(date +%H%M%S)-$$}"
+PV_RUNTIMES="${PV_RUNTIMES:-hermes openclaw claude-code}"
+PROVISION_TIMEOUT_SECS="${E2E_PROVISION_TIMEOUT_SECS:-1800}"
+
+# Slug MUST start with 'e2e-' so the sweep-stale-e2e-orgs safety net
+# (EPHEMERAL_PREFIXES) catches any leak this run fails to tear down.
+SLUG="e2e-pv-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
+SLUG=$(echo "$SLUG" | tr '[:upper:]' '[:lower:]' | tr -cd 'a-z0-9-' | head -c 32)
+
+ORG_ID=""
+TENANT_URL=""
+TENANT_TOKEN=""
+
+log()  { echo "[$(date +%H:%M:%S)] $*"; }
+fail() { echo "[$(date +%H:%M:%S)] ❌ $*" >&2; exit 1; }
+ok()   { echo "[$(date +%H:%M:%S)] ✅ $*"; }
+
+admin_call() {
+  local method="$1" path="$2"; shift 2
+  curl -sS -X "$method" "$CP_URL$path" \
+    -H "Authorization: Bearer $ADMIN_TOKEN" \
+    -H "Content-Type: application/json" "$@"
+}
+tenant_call() {
+  local method="$1" path="$2"; shift 2
+  curl -sS -X "$method" "$TENANT_URL$path" \
+    -H "Authorization: Bearer $TENANT_TOKEN" \
+    -H "X-Molecule-Org-Id: $ORG_ID" \
+    -H "Content-Type: application/json" "$@"
+}
+
+# ─── Scoped teardown ───────────────────────────────────────────────────
+# Deletes ONLY the org this run created (DELETE /cp/admin/tenants/$SLUG
+# with the {"confirm":$SLUG} fat-finger guard). Never a cluster-wide
+# sweep — honors feedback_cleanup_after_each_test and
+# feedback_never_run_cluster_cleanup_tests_on_live_platform. The
+# workflow's always() step + sweep-stale-e2e-orgs are the outer nets.
+teardown() {
+  local rc=$?
+  set +e
+  if [ "${E2E_KEEP_ORG:-0}" = "1" ]; then
+    echo ""
+    log "[teardown] E2E_KEEP_ORG=1 — leaving $SLUG for debugging (REMEMBER TO DELETE)"
+    exit $rc
+  fi
+  echo ""
+  log "[teardown] DELETE /cp/admin/tenants/$SLUG (scoped to this run only)"
+  admin_call DELETE "/cp/admin/tenants/$SLUG" --max-time 120 \
+    -d "{\"confirm\":\"$SLUG\"}" >/dev/null 2>&1
+  for j in $(seq 1 24); do
+    LIST=$(admin_call GET "/cp/admin/orgs?limit=500" 2>/dev/null)
+    LEAK=$(echo "$LIST" | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: print(1); sys.exit(0)
+orgs = d if isinstance(d, list) else d.get('orgs', [])
+print(sum(1 for o in orgs if o.get('slug') == '$SLUG' and o.get('instance_status') not in ('purged',) and o.get('status') != 'purged'))
+" 2>/dev/null || echo 1)
+    if [ "$LEAK" = "0" ]; then
+      log "[teardown] ✓ $SLUG purged (after ${j}x5s)"
+      exit $rc
+    fi
+    sleep 5
+  done
+  echo "::warning::[teardown] $SLUG still present after 120s — sweep-stale-e2e-orgs will catch it within MAX_AGE_MINUTES" >&2
+  [ $rc -eq 0 ] && rc=4
+  exit $rc
+}
+trap teardown EXIT INT TERM
+
+# ─── 1. Provision the throwaway org ────────────────────────────────────
+log "1/6 POST /cp/admin/orgs — slug=$SLUG"
+CREATE=$(admin_call POST /cp/admin/orgs \
+  -d "{\"slug\":\"$SLUG\",\"name\":\"E2E peer-visibility $SLUG\",\"owner_user_id\":\"e2e-runner:$SLUG\"}")
+ORG_ID=$(echo "$CREATE" | python3 -c "import sys,json; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
+[ -n "$ORG_ID" ] || fail "org creation failed: $(echo "$CREATE" | head -c 300)"
+log "    ORG_ID=$ORG_ID"
+
+# ─── 2. Wait for tenant EC2 + DNS ──────────────────────────────────────
+log "2/6 waiting for tenant instance_status=running (cold EC2 + cloudflared)..."
+DEADLINE=$(( $(date +%s) + PROVISION_TIMEOUT_SECS ))
+while true; do
+  [ "$(date +%s)" -gt "$DEADLINE" ] && fail "tenant never came up within ${PROVISION_TIMEOUT_SECS}s"
+  STATUS=$(admin_call GET "/cp/admin/orgs?limit=500" 2>/dev/null | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: sys.exit(0)
+orgs = d if isinstance(d, list) else d.get('orgs', [])
+for o in orgs:
+    if o.get('slug') == '$SLUG':
+        print(o.get('instance_status') or o.get('status') or 'unknown'); break
+" 2>/dev/null)
+  case "$STATUS" in running|online|ready) break ;; esac
+  sleep 10
+done
+log "    tenant status=$STATUS"
+
+# ─── 3. Per-tenant admin token + tenant URL ────────────────────────────
+log "3/6 fetching per-tenant admin token..."
+TT_RESP=$(admin_call GET "/cp/admin/orgs/$SLUG/admin-token")
+TENANT_TOKEN=$(echo "$TT_RESP" | python3 -c "import sys,json; print(json.load(sys.stdin).get('admin_token',''))" 2>/dev/null)
+[ -n "$TENANT_TOKEN" ] || fail "tenant token fetch failed: $(echo "$TT_RESP" | head -c 200)"
+
+CP_HOST=$(echo "$CP_URL" | sed -E 's#^https?://##; s#/.*$##')
+case "$CP_HOST" in
+  api.*)         DERIVED_DOMAIN="${CP_HOST#api.}" ;;
+  staging-api.*) DERIVED_DOMAIN="staging.${CP_HOST#staging-api.}" ;;
+  *)             DERIVED_DOMAIN="$CP_HOST" ;;
+esac
+TENANT_URL="https://${SLUG}.${DERIVED_DOMAIN}"
+log "    tenant url: $TENANT_URL"
+
+log "3b. waiting for tenant /health (TLS/DNS, up to 10min)..."
+for i in $(seq 1 120); do
+  curl -fsS "$TENANT_URL/health" -m 5 -k >/dev/null 2>&1 && { log "    /health ok (attempt $i)"; break; }
+  sleep 5
+done
+
+# ─── 4. Provision the parent + one sibling per runtime under test ──────
+# Inject the LLM provider key so each runtime can authenticate at boot.
+# Priority: MiniMax → direct-Anthropic → OpenAI (mirrors
+# test_staging_full_saas.sh's secrets-injection chain).
+SECRETS_JSON='{}'
+if [ -n "${E2E_MINIMAX_API_KEY:-}" ]; then
+  SECRETS_JSON=$(python3 -c "import json,os;k=os.environ['E2E_MINIMAX_API_KEY'];print(json.dumps({'ANTHROPIC_BASE_URL':'https://api.minimax.io/anthropic','ANTHROPIC_AUTH_TOKEN':k,'MINIMAX_API_KEY':k}))")
+elif [ -n "${E2E_ANTHROPIC_API_KEY:-}" ]; then
+  SECRETS_JSON=$(python3 -c "import json,os;k=os.environ['E2E_ANTHROPIC_API_KEY'];print(json.dumps({'ANTHROPIC_API_KEY':k}))")
+elif [ -n "${E2E_OPENAI_API_KEY:-}" ]; then
+  SECRETS_JSON=$(python3 -c "import json,os;k=os.environ['E2E_OPENAI_API_KEY'];print(json.dumps({'OPENAI_API_KEY':k,'OPENAI_BASE_URL':'https://api.openai.com/v1','MODEL_PROVIDER':'openai:gpt-4o','HERMES_INFERENCE_PROVIDER':'custom','HERMES_CUSTOM_BASE_URL':'https://api.openai.com/v1','HERMES_CUSTOM_API_KEY':k,'HERMES_CUSTOM_API_MODE':'chat_completions'}))")
+fi
+
+log "4/6 provisioning parent (claude-code) + one sibling per runtime under test..."
+P_RESP=$(tenant_call POST /workspaces \
+  -d "{\"name\":\"pv-parent\",\"runtime\":\"claude-code\",\"tier\":3,\"secrets\":$SECRETS_JSON}")
+PARENT_ID=$(echo "$P_RESP" | python3 -c "import sys,json; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
+[ -n "$PARENT_ID" ] || fail "parent create failed: $(echo "$P_RESP" | head -c 300)"
+log "    PARENT_ID=$PARENT_ID"
+
+# WS_IDS[runtime]=id ; WS_TOKENS[runtime]=auth_token (the MCP bearer)
+declare -A WS_IDS WS_TOKENS
+ALL_WS_IDS="$PARENT_ID"
+for rt in $PV_RUNTIMES; do
+  R=$(tenant_call POST /workspaces \
+    -d "{\"name\":\"pv-$rt\",\"runtime\":\"$rt\",\"tier\":2,\"parent_id\":\"$PARENT_ID\",\"secrets\":$SECRETS_JSON}")
+  WID=$(echo "$R" | python3 -c "import sys,json; print(json.load(sys.stdin).get('id',''))" 2>/dev/null)
+  # auth_token is top-level for container runtimes; external-like nest it
+  # under connection.auth_token (verified vs staging response shape).
+  WTOK=$(echo "$R" | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: print(''); sys.exit(0)
+print(d.get('auth_token') or d.get('connection', {}).get('auth_token') or '')
+" 2>/dev/null)
+  [ -n "$WID" ] || fail "$rt workspace create failed: $(echo "$R" | head -c 300)"
+  [ -n "$WTOK" ] || fail "$rt workspace did not return an auth_token — cannot drive its MCP call (resp: $(echo "$R" | head -c 300))"
+  WS_IDS[$rt]="$WID"
+  WS_TOKENS[$rt]="$WTOK"
+  ALL_WS_IDS="$ALL_WS_IDS $WID"
+  log "    $rt → $WID"
+done
+
+# ─── 5. Wait for every sibling online ──────────────────────────────────
+log "5/6 waiting for all workspaces status=online (up to ${PROVISION_TIMEOUT_SECS}s — cold boot)..."
+WS_DEADLINE=$(( $(date +%s) + PROVISION_TIMEOUT_SECS ))
+for rt in $PV_RUNTIMES; do
+  wid="${WS_IDS[$rt]}"
+  LAST=""
+  while true; do
+    [ "$(date +%s)" -gt "$WS_DEADLINE" ] && fail "$rt ($wid) never reached online (last=$LAST)"
+    S=$(tenant_call GET "/workspaces/$wid" 2>/dev/null | python3 -c "
+import sys, json
+try: d = json.load(sys.stdin)
+except Exception: sys.exit(0)
+w = d.get('workspace') if isinstance(d.get('workspace'), dict) else d
+print(w.get('status') or '')
+" 2>/dev/null)
+    [ "$S" != "$LAST" ] && { log "    $rt → $S"; LAST="$S"; }
+    case "$S" in
+      online) break ;;
+      failed) sleep 10 ;;   # transient: bootstrap-watcher 5-min deadline, heartbeat recovers
+      *)      sleep 10 ;;
+    esac
+  done
+  ok "    $rt online"
+done
+
+# ─── 6. THE GATE — literal mcp_molecule_list_peers via POST /:id/mcp ────
+# This is the byte-for-byte user-facing call. NOT GET /registry/:id/peers,
+# NOT /health, NOT the heartbeat table. JSON-RPC 2.0 tools/call,
+# name=list_peers, authenticated by the workspace's OWN bearer token
+# through WorkspaceAuth + MCPRateLimiter.
+log "6/6 driving the LITERAL list_peers MCP call per runtime..."
+echo ""
+RPC_BODY='{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_peers","arguments":{}}}'
+REGRESSED=0
+declare -A VERDICT
+
+for rt in $PV_RUNTIMES; do
+  wid="${WS_IDS[$rt]}"
+  wtok="${WS_TOKENS[$rt]}"
+  # The expected peer set = every OTHER provisioned workspace (parent +
+  # the sibling runtimes), excluding the caller itself.
+  EXPECT_IDS=$(echo "$ALL_WS_IDS" | tr ' ' '\n' | grep -v "^${wid}$" | grep -v '^$')
+
+  set +e
+  RESP=$(curl -sS -X POST "$TENANT_URL/workspaces/$wid/mcp" \
+    -H "Authorization: Bearer $wtok" \
+    -H "X-Molecule-Org-Id: $ORG_ID" \
+    -H "Content-Type: application/json" \
+    -d "$RPC_BODY" \
+    -o /tmp/pv_mcp_body.json -w "%{http_code}" 2>/dev/null)
+  set -e
+  HTTP_CODE="$RESP"
+  BODY=$(cat /tmp/pv_mcp_body.json 2>/dev/null || echo '')
+
+  echo "--- $rt (ws=$wid) ---"
+  echo "    HTTP $HTTP_CODE"
+  echo "    body: $(echo "$BODY" | head -c 600)"
+
+  # (1) HTTP 200 — a 401 (WorkspaceAuth reject, the Hermes symptom) fails here.
+  if [ "$HTTP_CODE" != "200" ]; then
+    echo "  ✗ $rt: list_peers MCP call returned HTTP $HTTP_CODE (expected 200)"
+    VERDICT[$rt]="FAIL(http=$HTTP_CODE)"
+    REGRESSED=1
+    continue
+  fi
+
+  # (2) JSON-RPC result present, not an error object.
+  PARSE=$(echo "$BODY" | python3 -c "
+import sys, json
+expect = set(filter(None, '''$EXPECT_IDS'''.split()))
+try:
+    d = json.load(sys.stdin)
+except Exception as e:
+    print('PARSE_ERROR:' + str(e)); sys.exit(0)
+if isinstance(d, dict) and d.get('error') is not None:
+    print('RPC_ERROR:' + json.dumps(d['error'])[:200]); sys.exit(0)
+res = d.get('result') if isinstance(d, dict) else None
+if res is None:
+    print('NO_RESULT'); sys.exit(0)
+# MCP tools/call result shape: {content:[{type:text,text:'<json or prose>'}]}
+text = ''
+if isinstance(res, dict):
+    for c in res.get('content', []):
+        if c.get('type') == 'text':
+            text += c.get('text', '')
+text_l = text.lower()
+# Native-sessions fallback signature (the OpenClaw symptom): the agent
+# answered from its own runtime session list, not the platform peer set.
+if 'sessions_list' in text_l or 'no platform peers' in text_l or 'native session' in text_l:
+    print('NATIVE_FALLBACK:' + text[:200]); sys.exit(0)
+# The expected sibling IDs must literally appear in the returned peer text.
+found = sorted(i for i in expect if i in text)
+missing = sorted(expect - set(found))
+if not expect:
+    print('NO_EXPECTED_PEERS_CONFIGURED'); sys.exit(0)
+if missing:
+    print('MISSING_PEERS:found=%d/%d missing=%s' % (len(found), len(expect), ','.join(m[:8] for m in missing)))
+    sys.exit(0)
+print('OK:found=%d/%d' % (len(found), len(expect)))
+" 2>/dev/null)
+
+  case "$PARSE" in
+    OK:*)
+      echo "  ✓ $rt: list_peers returned 200 and contains all expected peers ($PARSE)"
+      VERDICT[$rt]="OK"
+      ;;
+    NATIVE_FALLBACK:*)
+      echo "  ✗ $rt: list_peers fell back to NATIVE sessions — sees no platform peers ($PARSE)"
+      VERDICT[$rt]="FAIL(native-fallback)"
+      REGRESSED=1
+      ;;
+    RPC_ERROR:*|NO_RESULT|PARSE_ERROR:*)
+      echo "  ✗ $rt: list_peers MCP call did not return a usable result ($PARSE)"
+      VERDICT[$rt]="FAIL(rpc=$PARSE)"
+      REGRESSED=1
+      ;;
+    MISSING_PEERS:*)
+      echo "  ✗ $rt: list_peers returned 200 but peer set is wrong/empty ($PARSE)"
+      VERDICT[$rt]="FAIL(peers=$PARSE)"
+      REGRESSED=1
+      ;;
+    *)
+      echo "  ✗ $rt: unexpected verdict '$PARSE'"
+      VERDICT[$rt]="FAIL(unknown)"
+      REGRESSED=1
+      ;;
+  esac
+  echo ""
+done
+
+echo "=== SUMMARY — fresh-provision peer-visibility (literal MCP list_peers) ==="
+for rt in $PV_RUNTIMES; do
+  printf '  %-14s %s\n' "$rt" "${VERDICT[$rt]:-NO_RUN}"
+done
+echo ""
+
+if [ "$REGRESSED" -ne 0 ]; then
+  echo "✗ GATE FAILED — at least one runtime cannot see its peers via the"
+  echo "  literal mcp_molecule_list_peers call. This is the real user-facing"
+  echo "  failure the proxy signals (registry row / heartbeat / model 200)"
+  echo "  were hiding. Expected RED until the Hermes-401 + OpenClaw-MCP-wiring"
+  echo "  root-cause fixes land; goes green only when they actually do."
+  exit 10
+fi
+
+ok "GATE PASSED — every runtime under test sees its platform peers via the literal MCP call."
+exit 0

workspace-server/internal/handlers/a2a_poll_ingest_persist_test.go

@@ -0,0 +1,160 @@
+package handlers
+
+// Regression coverage for the POLL-mode arm of the canvas user-message
+// data-loss bug (internal#470 sibling — tracked on internal#471).
+//
+// Bug (reported 2026-05-16 by CTO Hongming): "in canvas i sometimes lose
+// my own message when i exit chat". The push-mode arm was fixed by
+// #1347 (persistUserMessageAtIngest — a SYNCHRONOUS, before-dispatch,
+// context.WithoutCancel INSERT). #1347's framing asserted "poll-mode
+// workspaces were never affected — logA2AReceiveQueued already persists
+// at ingest". That assertion is OVERSTATED.
+//
+// Hongming's tenant (slug `hongming`, org 2c940477-...) has 4 workspaces,
+// ALL runtime=external with empty URL → ALL delivery_mode=poll (proven
+// empirically: a benign A2A probe returns the synthetic
+// {"delivery_mode":"poll","status":"queued"} envelope for every one).
+// So his reported loss is the POLL path, NOT the push path #1347 fixes.
+//
+// Root cause (poll arm): the poll-mode short-circuit (a2a_proxy.go ~402)
+// calls logA2AReceiveQueued and then IMMEDIATELY returns the synthetic
+// 200 {status:"queued"} to the canvas. But logA2AReceiveQueued's durable
+// INSERT runs inside h.goAsync(...) — a DETACHED goroutine with NO
+// happens-before barrier against the HTTP response. The canvas sees 200
+// ("message accepted") while the activity_logs row may not yet be — and,
+// on a workspace-server restart / deploy / OOM / EC2 hibernation between
+// the 200 and the goroutine's commit, NEVER will be — durable. There is
+// also no fallback (unlike push-mode's legacy-INSERT fallback): a
+// swallowed LogActivity error loses the message with only a log line.
+// Chat-history reads activity_logs (postgres_store.go:165-187); a missing
+// row = message gone on reopen. That is exactly Hongming's symptom.
+//
+// Fix (parity with push-mode): the poll-mode ingest persist of the
+// canvas user message must be SYNCHRONOUS — committed before the queued
+// 200 is returned — on a context.WithoutCancel derived context, so a
+// client disconnect on chat-exit and a post-response restart cannot lose
+// it. Behavior is never worse than today (best-effort; a persist error
+// still returns queued).
+//
+// TEST DESIGN NOTE: sqlmock.ExpectationsWereMet() hangs indefinitely if
+// the expected query never fires. We use a select+default+time.After
+// pattern so the test FAILS fast (not hangs) when the production code
+// regresses to async (the INSERT never fires before handler returns),
+// while still returning promptly when all expectations are met. The
+// insertDelay is kept small (50ms) to minimise suite-level timing
+// impact under -race detection, where mock delays are amplified by
+// the instrumenter's goroutine overhead.
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/DATA-DOG/go-sqlmock"
+	"github.com/gin-gonic/gin"
+)
+
+// TestProxyA2A_PollMode_PersistsUserMessageSynchronouslyBeforeQueuedResponse
+// is the defining contract: for a poll-mode workspace, the canvas user
+// message MUST be durably INSERTed into activity_logs BEFORE the synthetic
+// queued 200 is returned to the client — with NO reliance on a detached
+// async goroutine completing later.
+//
+// The test proves the ordering by making the INSERT block briefly and
+// asserting the handler does NOT return until the INSERT has completed.
+// Pre-fix (INSERT in h.goAsync, response returned immediately) the
+// handler returns ~instantly while the INSERT is still pending in the
+// goroutine → the elapsed time is far below the injected INSERT delay and
+// ExpectationsWereMet() is racy/unmet at return. Post-fix (synchronous
+// persist before the queued response) the handler return is gated on the
+// INSERT, so elapsed >= the injected delay and the expectation is met
+// deterministically at return WITHOUT any waitAsyncForTest()/sleep.
+func TestProxyA2A_PollMode_PersistsUserMessageSynchronouslyBeforeQueuedResponse(t *testing.T) {
+	mock := setupTestDB(t)
+	setupTestRedis(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
+
+	const wsID = "ws-poll-sync-persist"
+	// Keep delay small: -race detection amplifies mock delays significantly.
+	// A 50ms delay is sufficient to prove synchronous blocking (~50× the
+	// normal INSERT latency) without bloating the full ./... suite runtime.
+	const insertDelay = 50 * time.Millisecond
+
+	expectBudgetCheck(mock, wsID)
+
+	// lookupDeliveryMode → poll, triggering the short-circuit.
+	mock.ExpectQuery("SELECT delivery_mode FROM workspaces WHERE id").
+		WithArgs(wsID).
+		WillReturnRows(sqlmock.NewRows([]string{"delivery_mode"}).AddRow("poll"))
+
+	// workspace-name lookup inside logA2AReceiveQueued.
+	mock.ExpectQuery(`SELECT name FROM workspaces WHERE id`).
+		WithArgs(wsID).
+		WillReturnRows(sqlmock.NewRows([]string{"name"}).AddRow("Poll WS"))
+
+	// The durable user-message write. We delay it so a synchronous
+	// persist visibly gates the handler return; a detached-goroutine
+	// persist (pre-fix) does not. The fix must keep using
+	// context.WithoutCancel so this write survives a chat-exit cancel.
+	mock.ExpectExec("INSERT INTO activity_logs").
+		WillDelayFor(insertDelay).
+		WillReturnResult(sqlmock.NewResult(0, 1))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: wsID}}
+
+	// callerID == "" (no X-Workspace-ID) → this is a canvas_user message,
+	// exactly Hongming's case.
+	body := `{"jsonrpc":"2.0","id":"poll-canvas-1","method":"message/send","params":{"message":{"role":"user","parts":[{"text":"my own message"}]}}}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+wsID+"/a2a", bytes.NewBufferString(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	start := time.Now()
+	handler.ProxyA2A(c)
+	elapsed := time.Since(start)
+
+	// Defining assertion #1: the handler must not have returned the
+	// queued response before the durable INSERT committed. Pre-fix this
+	// fails (elapsed ≈ 0, INSERT still racing in goAsync).
+	if elapsed < insertDelay {
+		t.Fatalf("poll-mode queued response returned in %v, before the %v user-message INSERT — "+
+			"the message is not durable when the client/process goes away (DATA LOSS). "+
+			"Persist must be synchronous before the queued 200.", elapsed, insertDelay)
+	}
+
+	// Defining assertion #2: the durable write actually happened by the
+	// time the handler returned. ExpectionsWereMet() hangs indefinitely if
+	// the mock never fires (e.g. production code regressed to async),
+	// so we check it in a goroutine with a hard 2s timeout — fails fast
+	// (no CI hang) on regression while returning promptly on success.
+	expectDone := make(chan error, 1)
+	go func() { expectDone <- mock.ExpectationsWereMet() }()
+	select {
+	case err := <-expectDone:
+		if err != nil {
+			t.Fatalf("user-message INSERT was not durable at handler return (unmet sqlmock expectations): %v", err)
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatalf("ExpectationsWereMet() hung for >2s — INSERT mock never fired. " +
+			"Likely cause: production code regressed logA2AReceiveQueued to goAsync " +
+			"(INSERT fires after handler returns, not before).")
+	}
+
+	// Sanity: still the correct poll-mode envelope + status.
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200 (queued), got %d: %s", w.Code, w.Body.String())
+	}
+	var resp map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("response is not valid JSON: %v", err)
+	}
+	if resp["status"] != "queued" || resp["delivery_mode"] != "poll" {
+		t.Errorf("poll envelope changed: got status=%v delivery_mode=%v, want queued/poll",
+			resp["status"], resp["delivery_mode"])
+	}
+}

workspace-server/internal/handlers/a2a_proxy.go

@@ -97,28 +97,28 @@ const maxProxyResponseBody = 10 << 20
 //
 // Timeout model — three independent budgets, none of which gets in each other's way:
 //
-//   1. Client.Timeout — DELIBERATELY UNSET. Client.Timeout is a hard wall on
-//      the entire request including streamed body reads, and would pre-empt
-//      legitimate slow cold-start flows (Claude Code first-token over OAuth
-//      can take 30-60s on boot; long-running agent synthesis can stream
-//      tokens for minutes). Total-request budget is enforced per-request
-//      via context deadline (canvas = idle-only, agent-to-agent = 30 min ceiling).
+//  1. Client.Timeout — DELIBERATELY UNSET. Client.Timeout is a hard wall on
+//     the entire request including streamed body reads, and would pre-empt
+//     legitimate slow cold-start flows (Claude Code first-token over OAuth
+//     can take 30-60s on boot; long-running agent synthesis can stream
+//     tokens for minutes). Total-request budget is enforced per-request
+//     via context deadline (canvas = idle-only, agent-to-agent = 30 min ceiling).
 //
-//   2. Transport.DialContext — 10s connect timeout. When a workspace's EC2
-//      black-holes TCP connects (instance terminated mid-flight, security group
-//      flipped, NACL bug), the OS default is 75s on Linux / 21s on macOS — long
-//      enough that Cloudflare's ~100s edge timeout can fire first and surface
-//      a generic 502 page to canvas. 10s is well above realistic intra-region
-//      latencies and well below CF's edge timeout.
+//  2. Transport.DialContext — 10s connect timeout. When a workspace's EC2
+//     black-holes TCP connects (instance terminated mid-flight, security group
+//     flipped, NACL bug), the OS default is 75s on Linux / 21s on macOS — long
+//     enough that Cloudflare's ~100s edge timeout can fire first and surface
+//     a generic 502 page to canvas. 10s is well above realistic intra-region
+//     latencies and well below CF's edge timeout.
 //
-//   3. Transport.ResponseHeaderTimeout — 180s default. From request-body-end
-//      to response-headers-start. Configurable via
-//      A2A_PROXY_RESPONSE_HEADER_TIMEOUT (envx.Duration). Covers cold-start
-//      first-byte (30-60s OAuth flow above) with enough room for Opus agent
-//      turns (big context + internal delegate_task round-trips routinely exceed
-//      the old 60s ceiling). Body streaming after headers is governed by the
-//      per-request context deadline, NOT this timeout — so multi-minute agent
-//      responses still work fine.
+//  3. Transport.ResponseHeaderTimeout — 180s default. From request-body-end
+//     to response-headers-start. Configurable via
+//     A2A_PROXY_RESPONSE_HEADER_TIMEOUT (envx.Duration). Covers cold-start
+//     first-byte (30-60s OAuth flow above) with enough room for Opus agent
+//     turns (big context + internal delegate_task round-trips routinely exceed
+//     the old 60s ceiling). Body streaming after headers is governed by the
+//     per-request context deadline, NOT this timeout — so multi-minute agent
+//     responses still work fine.
 //
 // The point of (2) and (3) is to surface a *structured* 503 from
 // handleA2ADispatchError when the workspace agent is unreachable, so canvas
@@ -399,7 +399,21 @@ func (h *WorkspaceHandler) proxyA2ARequest(ctx context.Context, workspaceID stri
 	// (no Do(), no maybeMarkContainerDead). The response is a synthetic
 	// {status:"queued"} envelope so the caller (canvas, another workspace)
 	// knows delivery is acknowledged but pending consumption.
-	if lookupDeliveryMode(ctx, workspaceID) == models.DeliveryModePoll {
+	deliveryMode, deliveryModeErr := lookupDeliveryMode(ctx, workspaceID)
+	if deliveryModeErr != nil {
+		// internal#497 fail-closed: a real DB/context error on the
+		// delivery-mode read MUST NOT silently fall through to the push
+		// dispatch path — that is exactly what silently misrouted every
+		// poll-mode peer for 5 days under the ce2db75f regression. Surface
+		// a structured error so the delegation is marked failed (loud +
+		// retryable) instead of dispatched to the wrong path.
+		log.Printf("ProxyA2A: delivery-mode lookup failed for %s: %v — failing closed", workspaceID, deliveryModeErr)
+		return 0, nil, &proxyA2AError{
+			Status:   http.StatusServiceUnavailable,
+			Response: gin.H{"error": "delivery-mode lookup failed; refusing to dispatch to avoid silent misrouting"},
+		}
+	}
+	if deliveryMode == models.DeliveryModePoll {
 		if logActivity {
 			h.logA2AReceiveQueued(ctx, workspaceID, callerID, body, a2aMethod)
 		}

workspace-server/internal/handlers/a2a_proxy_helpers.go

@@ -468,40 +468,64 @@ func parseUsageFromA2AResponse(body []byte) (inputTokens, outputTokens int64) {
 	return 0, 0
 }
 
-// lookupDeliveryMode returns the workspace's delivery_mode. On any DB
-// error or missing row it returns DeliveryModePush — the fail-closed
-// default. "Closed" here means "fall back to today's behavior (synchronous
-// dispatch)" rather than "fall back to drop the request silently into
-// activity_logs where the agent might never see it." A poll-mode workspace
-// that briefly reads as push will get its A2A request dispatched to the
-// stored URL (or a 502 if no URL); a push-mode workspace that briefly
-// reads as poll would get its request silently queued with no dispatch.
-// The first failure is loud + recoverable; the second is silent.
+// lookupDeliveryMode returns the workspace's delivery_mode.
+//
+// internal#497 / RFC#497 fail-closed (SURGICAL scope): the *specific*
+// failure mode that hid the ce2db75f regression for 5 days is now
+// propagated instead of silently swallowed — a CONTEXT error
+// (context.Canceled / context.DeadlineExceeded). Under ce2db75f the
+// detached delegation goroutine ran on a cancelled request context, every
+// `SELECT delivery_mode` failed `context canceled`, this function returned
+// push, the poll-mode short-circuit in proxyA2ARequest was skipped, and
+// poll-mode peers (e.g. an operator laptop on molecule-mcp-claude-channel)
+// silently never got their a2a_receive inbox row. A transient,
+// systematic-once-triggered context cancellation became permanent
+// invisible misrouting. Returning that error lets the caller fail loud
+// (mark the delegation failed) instead of mis-dispatching.
+//
+// Scope is deliberately narrow: only ctx errors propagate. Other DB
+// errors retain the long-standing documented "fall back to push (today's
+// synchronous behavior)" contract — that path is loud + recoverable
+// (502 / SSRF reject / restart), unlike the silent poll-mode drop, and
+// the surrounding proxy (incl. the sibling checkWorkspaceBudget) is
+// intentionally built around that fail-open-to-push behavior. Widening
+// further is an RFC#497 follow-up, not part of this P0 fix.
+//
+// A genuinely *absent* configuration is NOT an error and still resolves to
+// push (the safe synchronous default): sql.ErrNoRows, a NULL/empty column,
+// or an unrecognised value all return (push, nil).
 //
 // The function is intentionally lookup-only — it never mutates the row.
 // The register handler (registry.go) is the only writer for delivery_mode.
 //
 // See #2339 PR 1 for the column + register-flow side; this is the
 // proxy-side read used for the short-circuit in proxyA2ARequest.
-func lookupDeliveryMode(ctx context.Context, workspaceID string) string {
+func lookupDeliveryMode(ctx context.Context, workspaceID string) (string, error) {
 	var mode sql.NullString
 	err := db.DB.QueryRowContext(ctx,
 		`SELECT delivery_mode FROM workspaces WHERE id = $1`, workspaceID,
 	).Scan(&mode)
 	if err != nil {
-		if !errors.Is(err, sql.ErrNoRows) {
-			log.Printf("ProxyA2A: lookupDeliveryMode(%s) failed (%v) — defaulting to push", workspaceID, err)
+		// internal#497: a context cancellation/deadline MUST NOT be
+		// swallowed into a silent push default — that is the exact 5-day
+		// silent-misrouting vector. Propagate so the caller fails closed.
+		if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+			log.Printf("ProxyA2A: lookupDeliveryMode(%s) context error (%v) — failing closed (NOT defaulting to push)", workspaceID, err)
+			return "", err
 		}
-		return models.DeliveryModePush
+		if !errors.Is(err, sql.ErrNoRows) {
+			log.Printf("ProxyA2A: lookupDeliveryMode(%s) failed (%v) — defaulting to push (non-ctx DB error; legacy fail-open-to-push contract)", workspaceID, err)
+		}
+		return models.DeliveryModePush, nil
 	}
 	if !mode.Valid || mode.String == "" {
-		return models.DeliveryModePush
+		return models.DeliveryModePush, nil
 	}
 	if !models.IsValidDeliveryMode(mode.String) {
 		log.Printf("ProxyA2A: workspace %s has invalid delivery_mode=%q — defaulting to push", workspaceID, mode.String)
-		return models.DeliveryModePush
+		return models.DeliveryModePush, nil
 	}
-	return mode.String
+	return mode.String, nil
 }
 
 // logA2AReceiveQueued records a poll-mode "queued" A2A receive into
@@ -514,26 +538,49 @@ func lookupDeliveryMode(ctx context.Context, workspaceID string) string {
 // reads in PR 3 — that's how a poll-mode workspace receives inbound A2A
 // without a public URL.
 func (h *WorkspaceHandler) logA2AReceiveQueued(ctx context.Context, workspaceID, callerID string, body []byte, a2aMethod string) {
+	// DATA-LOSS FIX (internal#471 — poll-mode sibling of #1347/internal#470):
+	// this is the ONLY durable write of a poll-mode inbound message,
+	// including a canvas_user message (callerID == "") typed in the canvas
+	// chat. It MUST be SYNCHRONOUS and complete BEFORE the caller returns
+	// the synthetic {status:"queued"} 200 — otherwise the canvas sees the
+	// send acknowledged while the activity_logs row is still racing in a
+	// detached goroutine, and a workspace-server restart / deploy / OOM /
+	// EC2 hibernation between the 200 and the goroutine's commit loses the
+	// user's message permanently (chat-history reads activity_logs, so a
+	// missing row = message gone on reopen). Hongming's tenant is entirely
+	// poll-mode (4 external workspaces, no URL — verified empirically), so
+	// his reported loss is THIS path; #1347 (push-mode, persists AFTER the
+	// poll short-circuit) structurally cannot cover it.
+	//
+	// Mirrors persistUserMessageAtIngest's discipline:
+	//   - context.WithoutCancel: a client disconnect on chat-exit (which
+	//     cancels the inbound request ctx) MUST NOT abort this write.
+	//   - SYNCHRONOUS (no goAsync): the row must be durable before the
+	//     queued 200 is returned to the caller.
+	//   - Best-effort: LogActivity already logs+swallows INSERT errors, so
+	//     a hiccup never blocks or fails the user's send (behavior for
+	//     that one request is never worse than the pre-fix async path).
+	// The post-commit broadcast still fires inside LogActivity; a missed
+	// WebSocket event is not data loss (the durable row is the truth the
+	// canvas re-reads on reopen).
+	insCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), 30*time.Second)
+	defer cancel()
+
 	var wsName string
-	db.DB.QueryRowContext(ctx, `SELECT name FROM workspaces WHERE id = $1`, workspaceID).Scan(&wsName)
+	db.DB.QueryRowContext(insCtx, `SELECT name FROM workspaces WHERE id = $1`, workspaceID).Scan(&wsName)
 	if wsName == "" {
 		wsName = workspaceID
 	}
 	summary := a2aMethod + " → " + wsName + " (queued for poll)"
-	parent := ctx
-	h.goAsync(func() {
-		logCtx, cancel := context.WithTimeout(context.WithoutCancel(parent), 30*time.Second)
-		defer cancel()
-		LogActivity(logCtx, h.broadcaster, ActivityParams{
-			WorkspaceID:  workspaceID,
-			ActivityType: "a2a_receive",
-			SourceID:     nilIfEmpty(callerID),
-			TargetID:     &workspaceID,
-			Method:       &a2aMethod,
-			Summary:      &summary,
-			RequestBody:  json.RawMessage(body),
-			Status:       "ok",
-		})
+	LogActivity(insCtx, h.broadcaster, ActivityParams{
+		WorkspaceID:  workspaceID,
+		ActivityType: "a2a_receive",
+		SourceID:     nilIfEmpty(callerID),
+		TargetID:     &workspaceID,
+		Method:       &a2aMethod,
+		Summary:      &summary,
+		RequestBody:  json.RawMessage(body),
+		Status:       "ok",
 	})
 }
 

workspace-server/internal/handlers/a2a_proxy_preflight_test.go

@@ -54,6 +54,7 @@ func TestPreflight_ContainerRunning_ReturnsNil(t *testing.T) {
 	_ = setupTestDB(t)
 	stub := &preflightLocalProv{running: true, err: nil}
 	h := NewWorkspaceHandler(newTestBroadcaster(), nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, h)
 	h.provisioner = stub
 
 	if err := h.preflightContainerHealth(context.Background(), "ws-running-123"); err != nil {
@@ -186,8 +187,8 @@ func TestProxyA2A_Preflight_RoutesThroughProvisionerSSOT(t *testing.T) {
 	}
 
 	var (
-		callsIsRunning             bool
-		callsContainerInspectRaw   bool
+		callsIsRunning                  bool
+		callsContainerInspectRaw        bool
 		callsRunningContainerNameDirect bool
 	)
 	ast.Inspect(fn.Body, func(n ast.Node) bool {

workspace-server/internal/handlers/a2a_proxy_test.go

@@ -262,6 +262,7 @@ func TestProxyA2A_Upstream502_TriggersContainerDeadCheck(t *testing.T) {
 	allowLoopbackForTest(t)
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 	cp := &fakeCPProv{running: false}
 	handler.SetCPProvisioner(cp)
 
@@ -324,6 +325,7 @@ func TestProxyA2A_Upstream502_AliveAgent_PropagatesAsIs(t *testing.T) {
 	allowLoopbackForTest(t)
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 	cp := &fakeCPProv{running: true}
 	handler.SetCPProvisioner(cp)
 
@@ -513,6 +515,7 @@ func TestProxyA2A_AllowedSelf_SkipsAccessCheck(t *testing.T) {
 	allowLoopbackForTest(t)
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 
 	agentServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		w.Header().Set("Content-Type", "application/json")
@@ -661,18 +664,18 @@ func TestProxyA2A_CallerIDDerivedFromBearer(t *testing.T) {
 	//    (column order: workspace_id, activity_type, source_id, target_id, ...)
 	mock.ExpectExec("INSERT INTO activity_logs").
 		WithArgs(
-			"ws-target",                       // $1 workspace_id
-			"a2a_receive",                     // $2 activity_type
-			sqlmock.AnyArg(),                  // $3 source_id — *string("ws-caller"), checked below
-			sqlmock.AnyArg(),                  // $4 target_id
-			sqlmock.AnyArg(),                  // $5 method
-			sqlmock.AnyArg(),                  // $6 summary
-			sqlmock.AnyArg(),                  // $7 request_body
-			sqlmock.AnyArg(),                  // $8 response_body
-			sqlmock.AnyArg(),                  // $9 tool_trace
-			sqlmock.AnyArg(),                  // $10 duration_ms
-			sqlmock.AnyArg(),                  // $11 status
-			sqlmock.AnyArg(),                  // $12 error_detail
+			"ws-target",      // $1 workspace_id
+			"a2a_receive",    // $2 activity_type
+			sqlmock.AnyArg(), // $3 source_id — *string("ws-caller"), checked below
+			sqlmock.AnyArg(), // $4 target_id
+			sqlmock.AnyArg(), // $5 method
+			sqlmock.AnyArg(), // $6 summary
+			sqlmock.AnyArg(), // $7 request_body
+			sqlmock.AnyArg(), // $8 response_body
+			sqlmock.AnyArg(), // $9 tool_trace
+			sqlmock.AnyArg(), // $10 duration_ms
+			sqlmock.AnyArg(), // $11 status
+			sqlmock.AnyArg(), // $12 error_detail
 		).
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
@@ -1716,7 +1719,6 @@ func TestDispatchA2A_RejectsUnsafeURL(t *testing.T) {
 	}
 }
 
-
 // --- handleA2ADispatchError ---
 
 func TestHandleA2ADispatchError_ContextDeadline(t *testing.T) {
@@ -1803,6 +1805,7 @@ func TestMaybeMarkContainerDead_CPOnly_NotRunning(t *testing.T) {
 	mock := setupTestDB(t)
 	setupTestRedis(t)
 	handler := NewWorkspaceHandler(newTestBroadcaster(), nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 	cp := &fakeCPProv{running: false}
 	handler.SetCPProvisioner(cp)
 
@@ -1955,6 +1958,7 @@ func TestLogA2AFailure_Smoke(t *testing.T) {
 	mock := setupTestDB(t)
 	setupTestRedis(t)
 	handler := NewWorkspaceHandler(newTestBroadcaster(), nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 
 	// Sync workspace-name lookup (called in the caller goroutine).
 	mock.ExpectQuery(`SELECT name FROM workspaces WHERE id =`).
@@ -1973,6 +1977,7 @@ func TestLogA2AFailure_EmptyNameFallback(t *testing.T) {
 	mock := setupTestDB(t)
 	setupTestRedis(t)
 	handler := NewWorkspaceHandler(newTestBroadcaster(), nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 
 	// Empty name from DB → summary uses the workspaceID as the name.
 	mock.ExpectQuery(`SELECT name FROM workspaces WHERE id =`).
@@ -1989,6 +1994,7 @@ func TestLogA2ASuccess_Smoke(t *testing.T) {
 	mock := setupTestDB(t)
 	setupTestRedis(t)
 	handler := NewWorkspaceHandler(newTestBroadcaster(), nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 
 	mock.ExpectQuery(`SELECT name FROM workspaces WHERE id =`).
 		WithArgs("ws-ok").
@@ -2005,6 +2011,7 @@ func TestLogA2ASuccess_ErrorStatus(t *testing.T) {
 	mock := setupTestDB(t)
 	setupTestRedis(t)
 	handler := NewWorkspaceHandler(newTestBroadcaster(), nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, handler)
 
 	mock.ExpectQuery(`SELECT name FROM workspaces WHERE id =`).
 		WithArgs("ws-err").
@@ -2228,12 +2235,18 @@ func TestProxyA2A_PushMode_NoShortCircuit(t *testing.T) {
 	}
 }
 
-// TestProxyA2A_PollMode_FailsClosedToPush verifies the safety contract:
-// a DB error reading delivery_mode must default to push (the existing
-// behavior), NOT poll. Failing to push means a poll-mode workspace
-// briefly attempts a real dispatch — visible failure (502 / SSRF
-// rejection / restart cascade), not a silent drop into activity_logs
-// where the agent might never look. Loud > silent, recoverable > lost.
+// TestProxyA2A_PollMode_FailsClosedToPush verifies the LEGACY safety
+// contract is PRESERVED for non-context DB errors: a generic DB error
+// reading delivery_mode still defaults to push (today's behavior), NOT
+// poll. Failing to push means a poll-mode workspace briefly attempts a
+// real dispatch — visible failure (502 / SSRF rejection / restart
+// cascade), not a silent drop into activity_logs where the agent might
+// never look. Loud > silent, recoverable > lost.
+//
+// internal#497 narrows the fail-closed change to *context* errors only
+// (the actual ce2db75f regression vector); generic DB errors keep this
+// long-standing fail-open-to-push contract. The ctx-error fail-closed is
+// covered by TestLookupDeliveryMode_ContextCanceled_FailsClosed.
 func TestProxyA2A_PollMode_FailsClosedToPush(t *testing.T) {
 	mock := setupTestDB(t)
 	setupTestRedis(t) // empty Redis — forces resolveAgentURL DB lookup
@@ -2244,7 +2257,8 @@ func TestProxyA2A_PollMode_FailsClosedToPush(t *testing.T) {
 
 	expectBudgetCheck(mock, wsID)
 
-	// lookupDeliveryMode hits a transient DB error → must default push.
+	// lookupDeliveryMode hits a generic (non-context) DB error → must
+	// still default push (legacy contract preserved by internal#497).
 	mock.ExpectQuery("SELECT delivery_mode FROM workspaces WHERE id").
 		WithArgs(wsID).
 		WillReturnError(sql.ErrConnDone)
@@ -2268,7 +2282,7 @@ func TestProxyA2A_PollMode_FailsClosedToPush(t *testing.T) {
 		var resp map[string]interface{}
 		_ = json.Unmarshal(w.Body.Bytes(), &resp)
 		if resp["status"] == "queued" {
-			t.Errorf("DB error on delivery_mode lookup silently queued the request — must fail-closed-to-push, got body: %s", w.Body.String())
+			t.Errorf("generic DB error on delivery_mode lookup silently queued the request — must fail-open-to-push, got body: %s", w.Body.String())
 		}
 	}
 
@@ -2277,6 +2291,37 @@ func TestProxyA2A_PollMode_FailsClosedToPush(t *testing.T) {
 	}
 }
 
+// TestLookupDeliveryMode_ContextCanceled_FailsClosed is the internal#497
+// regression test for the SECONDARY defect. It pins the exact invariant
+// that hid the ce2db75f regression for 5 days: when the delivery_mode read
+// fails because the context was cancelled (precisely what happened in the
+// detached delegation goroutine running on a returned request context),
+// lookupDeliveryMode MUST return an error and MUST NOT silently return
+// "push". Returning push there is what skipped the poll-mode short-circuit
+// and silently dropped 100% of poll-mode peer deliveries.
+//
+// A pre-cancelled context makes QueryRowContext fail with
+// context.Canceled deterministically — no DB rows are mocked because the
+// query never reaches a result.
+func TestLookupDeliveryMode_ContextCanceled_FailsClosed(t *testing.T) {
+	mock := setupTestDB(t)
+	// The query fails on the cancelled ctx before matching; provide a
+	// permissive expectation so sqlmock doesn't complain about the attempt.
+	mock.ExpectQuery("SELECT delivery_mode FROM workspaces WHERE id").
+		WillReturnError(context.Canceled)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel() // simulate the HTTP handler having returned (request ctx dead)
+
+	mode, err := lookupDeliveryMode(ctx, "ws-poll-peer")
+	if err == nil {
+		t.Fatalf("internal#497 regression: lookupDeliveryMode swallowed a context error and returned mode=%q with nil err — this is the exact 5-day silent-misrouting vector", mode)
+	}
+	if mode == models.DeliveryModePush {
+		t.Errorf("internal#497 regression: context error must NOT default to push (got mode=%q)", mode)
+	}
+}
+
 // ==================== a2aClient ResponseHeaderTimeout config ====================
 
 func TestA2AClientResponseHeaderTimeout(t *testing.T) {

workspace-server/internal/handlers/a2a_queue_test.go

@@ -26,6 +26,10 @@ import (
 // setupTestDBForQueueTests creates a sqlmock DB using QueryMatcherEqual (exact
 // string matching) so that ExpectQuery/ExpectExec patterns are compared verbatim.
 // Uses the same global db.DB as setupTestDB so the handler can use it.
+//
+// IMPORTANT: db.DB is saved before assignment and restored via t.Cleanup so
+// that tests running after this one are not polluted by a closed mock.
+// Same fix as setupTestDB (handlers_test.go); same root cause as mc#975.
 func setupTestDBForQueueTests(t *testing.T) sqlmock.Sqlmock {
 	t.Helper()
 	mockDB, mock, err := sqlmock.New(sqlmock.QueryMatcherOption(sqlmock.QueryMatcherEqual))

workspace-server/internal/handlers/admin_schedules_health_test.go

@@ -5,6 +5,9 @@ import (
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strconv"
 	"testing"
 	"time"
 
@@ -444,3 +447,178 @@ func TestAdminSchedulesHealth_ResponseFields(t *testing.T) {
 		t.Fatalf("unmet expectations: %v", err)
 	}
 }
+
+// ── classifyScheduleStatus — additional edge cases ─────────────────────────────────
+
+func TestClassifyScheduleStatus_ZeroThreshold(t *testing.T) {
+	now := time.Now()
+	lastRun := now.Add(-365 * 24 * time.Hour) // very old
+	result := classifyScheduleStatus(&lastRun, 0, now)
+	if result != "ok" {
+		t.Errorf("classifyScheduleStatus(threshold=0) = %q; want 'ok'", result)
+	}
+}
+
+func TestClassifyScheduleStatus_NegativeThreshold(t *testing.T) {
+	now := time.Now()
+	lastRun := now.Add(-24 * time.Hour)
+	result := classifyScheduleStatus(&lastRun, -1*time.Hour, now)
+	if result != "ok" {
+		t.Errorf("classifyScheduleStatus(threshold=-1h) = %q; want 'ok'", result)
+	}
+}
+
+func TestClassifyScheduleStatus_ExactlyAtThreshold(t *testing.T) {
+	// Strict >: if now.Sub(lastRun) == threshold, it is NOT stale
+	now := time.Date(2026, 5, 18, 12, 0, 0, 0, time.UTC)
+	lastRun := time.Date(2026, 5, 18, 10, 0, 0, 0, time.UTC) // exactly 2h ago
+	result := classifyScheduleStatus(&lastRun, 2*time.Hour, now)
+	if result != "ok" {
+		t.Errorf("classifyScheduleStatus(exactly at threshold) = %q; want 'ok'", result)
+	}
+}
+
+// ── loadRuntimeProvisionTimeouts (runtime_provision_timeouts.go) ─────────────────
+
+func writeRuntimeConfigYAML(t *testing.T, tmpDir, templateName, runtime string, timeoutSecs int) {
+	t.Helper()
+	dir := filepath.Join(tmpDir, templateName)
+	if err := os.MkdirAll(dir, 0755); err != nil {
+		t.Fatalf("MkdirAll(%s): %v", dir, err)
+	}
+	yamlContent := "runtime: " + runtime + "\nruntime_config:\n  provision_timeout_seconds: " + strconv.Itoa(timeoutSecs) + "\n"
+	if err := os.WriteFile(filepath.Join(dir, "config.yaml"), []byte(yamlContent), 0644); err != nil {
+		t.Fatalf("WriteFile: %v", err)
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_EmptyDir(t *testing.T) {
+	tmpDir := t.TempDir()
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if len(result) != 0 {
+		t.Errorf("loadRuntimeProvisionTimeouts(empty dir) len = %d; want 0", len(result))
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresNonDirEntries(t *testing.T) {
+	tmpDir := t.TempDir()
+	if err := os.WriteFile(filepath.Join(tmpDir, "not-a-dir.yaml"), []byte("runtime: hermes\n"), 0644); err != nil {
+		t.Fatal(err)
+	}
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if len(result) != 0 {
+		t.Errorf("loadRuntimeProvisionTimeouts(file-only dir) len = %d; want 0", len(result))
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_SingleTemplate(t *testing.T) {
+	tmpDir := t.TempDir()
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-hermes", "hermes", 300)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if v, ok := result["hermes"]; !ok || v != 300 {
+		t.Errorf("loadRuntimeProvisionTimeouts → hermes = %d; want 300", v)
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_MultipleTemplatesSameRuntime(t *testing.T) {
+	// Two templates using the same runtime — takes the MAX timeout
+	tmpDir := t.TempDir()
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-hermes-slow", "hermes", 600)
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-hermes-fast", "hermes", 120)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if v, ok := result["hermes"]; !ok || v != 600 {
+		t.Errorf("loadRuntimeProvisionTimeouts → hermes = %d; want 600 (max of 600, 120)", v)
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_MultipleRuntimes(t *testing.T) {
+	tmpDir := t.TempDir()
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-hermes", "hermes", 300)
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-claude-code", "claude-code", 420)
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-deepagents", "deepagents", 180)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	want := map[string]int{
+		"hermes":      300,
+		"claude-code": 420,
+		"deepagents":  180,
+	}
+	for runtime, wantSecs := range want {
+		if got, ok := result[runtime]; !ok || got != wantSecs {
+			t.Errorf("loadRuntimeProvisionTimeouts → %s = %d; want %d", runtime, got, wantSecs)
+		}
+	}
+	if len(result) != len(want) {
+		t.Errorf("loadRuntimeProvisionTimeouts → len = %d; want %d", len(result), len(want))
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresZeroTimeout(t *testing.T) {
+	tmpDir := t.TempDir()
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-zero", "zero-runtime", 0)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if _, ok := result["zero-runtime"]; ok {
+		t.Errorf("loadRuntimeProvisionTimeouts → 'zero-runtime' present; want absent (timeout=0)")
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresNegativeTimeout(t *testing.T) {
+	tmpDir := t.TempDir()
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-negative", "neg-runtime", -60)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if _, ok := result["neg-runtime"]; ok {
+		t.Errorf("loadRuntimeProvisionTimeouts → 'neg-runtime' present; want absent (timeout<0)")
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresMissingRuntimeField(t *testing.T) {
+	tmpDir := t.TempDir()
+	dir := filepath.Join(tmpDir, "tmpl-no-runtime")
+	if err := os.MkdirAll(dir, 0755); err != nil {
+		t.Fatal(err)
+	}
+	yamlContent := "template_name: no-runtime-template\nruntime_config:\n  provision_timeout_seconds: 300\n"
+	if err := os.WriteFile(filepath.Join(dir, "config.yaml"), []byte(yamlContent), 0644); err != nil {
+		t.Fatal(err)
+	}
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if len(result) != 0 {
+		t.Errorf("loadRuntimeProvisionTimeouts → len = %d; want 0 (runtime field absent)", len(result))
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresMalformedYAML(t *testing.T) {
+	tmpDir := t.TempDir()
+	dir := filepath.Join(tmpDir, "tmpl-bad-yaml")
+	if err := os.MkdirAll(dir, 0755); err != nil {
+		t.Fatal(err)
+	}
+	badYAML := "runtime: bad\n  provision_timeout_seconds: not a number\n"
+	if err := os.WriteFile(filepath.Join(dir, "config.yaml"), []byte(badYAML), 0644); err != nil {
+		t.Fatal(err)
+	}
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if len(result) != 0 {
+		t.Errorf("loadRuntimeProvisionTimeouts → len = %d; want 0 (malformed YAML)", len(result))
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresMissingConfig(t *testing.T) {
+	tmpDir := t.TempDir()
+	if err := os.MkdirAll(filepath.Join(tmpDir, "tmpl-no-config"), 0755); err != nil {
+		t.Fatal(err)
+	}
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-good", "good-runtime", 300)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if v, ok := result["good-runtime"]; !ok || v != 300 {
+		t.Errorf("loadRuntimeProvisionTimeouts → good-runtime = %d; want 300", v)
+	}
+}
+
+func TestLoadRuntimeProvisionTimeouts_IgnoresEmptyRuntime(t *testing.T) {
+	tmpDir := t.TempDir()
+	writeRuntimeConfigYAML(t, tmpDir, "tmpl-empty", "", 300)
+	result := loadRuntimeProvisionTimeouts(tmpDir)
+	if len(result) != 0 {
+		t.Errorf("loadRuntimeProvisionTimeouts → len = %d; want 0 (empty runtime)", len(result))
+	}
+}

workspace-server/internal/handlers/admin_workspace_images.go

@@ -44,8 +44,8 @@ func NewWorkspaceImageService(docker *dockerclient.Client) *WorkspaceImageServic
 // AllRuntimes is the canonical list mirroring docs/workspace-runtime-package.md.
 // Update both when a new template is added.
 var AllRuntimes = []string{
-	"claude-code", "langgraph", "crewai", "autogen",
-	"deepagents", "hermes", "gemini-cli", "openclaw",
+	"claude-code", "langgraph", "autogen",
+	"hermes", "openclaw",
 }
 
 // RefreshResult is the per-call outcome surfaced to HTTP callers AND logged

workspace-server/internal/handlers/agent_card_reconcile.go

@@ -0,0 +1,113 @@
+package handlers
+
+import "encoding/json"
+
+// agent_card_reconcile.go — server-side repair for the fleet-wide
+// agent-card identity gap.
+//
+// Root cause: the runtime builds its AgentCard from config.name
+// (workspace/main.py:198), and config.name is read from the
+// CP-regenerated /configs/config.yaml whose `name:` field is the raw
+// workspace UUID — NOT the friendly name the operator sees. The friendly
+// name IS captured: POST /workspaces and PATCH /workspaces/:id (the
+// canvas Details tab) write it to the trusted workspaces.name DB column.
+// But /registry/register stores the runtime-supplied card verbatim
+// (registry.go: `agent_card = EXCLUDED.agent_card`), so the stored card
+// served at /.well-known/agent-card.json and returned to peers via
+// agent_card_url ends up with name = UUID, description = "", role = null.
+//
+// Fix shape (deliberately minimal, no contract weakening): when the
+// runtime-supplied card's `name` is empty or equals the workspace UUID
+// (the placeholder the runtime had no better value for), the PLATFORM —
+// not the agent — substitutes the friendly value from the trusted
+// workspaces row. Identity stays platform-controlled: the agent never
+// gains the ability to self-set its own name/role; the platform sources
+// it from the operator-controlled DB column. We only ever FILL gaps
+// (empty / UUID-placeholder); a card that already carries a real
+// friendly name is never downgraded.
+//
+// list_peers / the /registry/:id/peers endpoint already resolve display
+// names from workspaces.name directly (discovery.go / mcp_tools.go
+// `SELECT w.id, w.name, ...`), so peer_name in delivered message tags
+// was already correct — this fix closes the remaining surface: the
+// agent_card blob itself (canvas Agent Card / Skills view, peer
+// agent_card_url fetches, the well-known card).
+//
+// description / role degrade discovery the same way: an empty
+// description and null role give peers nothing to reason about. We
+// default description from the (now reconciled) name when blank and
+// role from workspaces.role when the operator set one.
+
+// reconcileAgentCardIdentity patches identity gaps in a runtime-supplied
+// agent card from the trusted workspace DB row. It returns the
+// (possibly rewritten) card bytes and whether anything changed. On any
+// failure (malformed JSON, nothing to fill) it returns the input bytes
+// unchanged with changed=false so the caller can store them verbatim —
+// this is strictly no-worse-than-before, never a regression.
+//
+// Pure function: no DB / HTTP / globals, so it is exhaustively
+// unit-testable (agent_card_reconcile_test.go) without booting the
+// handler or a sqlmock.
+func reconcileAgentCardIdentity(card json.RawMessage, workspaceID, dbName, dbRole string) (json.RawMessage, bool) {
+	var m map[string]any
+	if err := json.Unmarshal(card, &m); err != nil || m == nil {
+		// Malformed card — not this function's job to reject it (the
+		// upsert stores it as-is and downstream readers handle bad
+		// JSON). Return verbatim so byte-for-byte behaviour is
+		// preserved on the failure path.
+		return card, false
+	}
+
+	changed := false
+
+	// name: fill only when empty or the UUID placeholder. A dbName that
+	// is itself the UUID is a placeholder row (registry.go INSERT seeds
+	// name = id before the canvas sets a friendly one) — not a friendly
+	// name, so it is not an eligible source.
+	cardName, _ := m["name"].(string)
+	if (cardName == "" || cardName == workspaceID) &&
+		dbName != "" && dbName != workspaceID {
+		m["name"] = dbName
+		changed = true
+	}
+
+	// description: when blank, default to the (reconciled) name so peers
+	// and the canvas Agent Card view have a non-empty human label
+	// instead of "". Mirrors the runtime's own
+	// `config.description or config.name` fallback (main.py:199) but
+	// applied to the registry copy where the runtime's fallback was the
+	// UUID.
+	if desc, _ := m["description"].(string); desc == "" {
+		if n, _ := m["name"].(string); n != "" && n != workspaceID {
+			m["description"] = n
+			changed = true
+		}
+	}
+
+	// role: surface the operator-set workspaces.role when the card
+	// carries none. Discovery (peer_role) and the canvas Role row read
+	// workspaces.role directly; this just makes the standalone card
+	// self-describing too. Never overwrite a role the card already has.
+	if dbRole != "" {
+		if r, ok := m["role"].(string); !ok || r == "" {
+			m["role"] = dbRole
+			changed = true
+		}
+	}
+
+	if !changed {
+		// No-op: return the original bytes untouched so callers that
+		// compare/store get byte-identical input (re-marshalling would
+		// reorder keys for no reason).
+		return card, false
+	}
+
+	out, err := json.Marshal(m)
+	if err != nil {
+		// Re-marshal of a map we just unmarshalled should never fail;
+		// if it somehow does, fall back to the verbatim input rather
+		// than storing nothing.
+		return card, false
+	}
+	return out, true
+}

workspace-server/internal/handlers/agent_card_reconcile_test.go

@@ -0,0 +1,166 @@
+package handlers
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+// TestReconcileAgentCardIdentity covers the server-side backfill that
+// repairs the fleet-wide agent-card identity gap (internal#XXX): the
+// runtime POSTs /registry/register with agent_card.name = the workspace
+// UUID (because the CP-regenerated /configs/config.yaml sets name: <uuid>)
+// while the trusted workspaces.name DB column — the value the canvas
+// Details tab shows and lets the operator edit — holds the friendly
+// name ("Claude Code Agent"). The platform reconciles them from the DB
+// row (NOT from the agent — identity stays platform-controlled, not
+// self-mutable).
+func TestReconcileAgentCardIdentity(t *testing.T) {
+	const wsID = "3b81321b-1ec7-488c-96f7-72c42a968da6"
+
+	tests := []struct {
+		name        string
+		card        string
+		dbName      string
+		dbRole      string
+		wantName    string
+		wantDesc    string
+		wantRole    string
+		wantChanged bool
+	}{
+		{
+			name:        "name is the workspace UUID — backfill from DB",
+			card:        `{"name":"3b81321b-1ec7-488c-96f7-72c42a968da6","description":"","capabilities":{"streaming":true}}`,
+			dbName:      "Claude Code Agent",
+			dbRole:      "",
+			wantName:    "Claude Code Agent",
+			wantDesc:    "Claude Code Agent",
+			wantRole:    "",
+			wantChanged: true,
+		},
+		{
+			name:        "empty name — backfill from DB",
+			card:        `{"name":"","description":"x"}`,
+			dbName:      "ops-agent",
+			dbRole:      "sre",
+			wantName:    "ops-agent",
+			wantDesc:    "x",
+			wantRole:    "sre",
+			wantChanged: true,
+		},
+		{
+			name:        "role null in card, DB has role — backfill role only",
+			card:        `{"name":"Reviewer","description":"Senior reviewer"}`,
+			dbName:      "Reviewer",
+			dbRole:      "code-reviewer",
+			wantName:    "Reviewer",
+			wantDesc:    "Senior reviewer",
+			wantRole:    "code-reviewer",
+			wantChanged: true,
+		},
+		{
+			name: "card already has a real friendly name — do NOT clobber it",
+			// A richer card (e.g. an external channel agent) must win;
+			// the platform only fills gaps, never downgrades.
+			card:        `{"name":"Claude Code (channel)","description":"Local Claude Code session bridged","role":"assistant"}`,
+			dbName:      "hongming-pc",
+			dbRole:      "operator",
+			wantName:    "Claude Code (channel)",
+			wantDesc:    "Local Claude Code session bridged",
+			wantRole:    "assistant",
+			wantChanged: false,
+		},
+		{
+			name:        "no DB name available — leave UUID name untouched (no worse than before)",
+			card:        `{"name":"3b81321b-1ec7-488c-96f7-72c42a968da6","description":""}`,
+			dbName:      "",
+			dbRole:      "",
+			wantName:    "3b81321b-1ec7-488c-96f7-72c42a968da6",
+			wantDesc:    "",
+			wantRole:    "",
+			wantChanged: false,
+		},
+		{
+			name:        "dbName equals UUID (placeholder row) — not a friendly name, leave untouched",
+			card:        `{"name":"3b81321b-1ec7-488c-96f7-72c42a968da6"}`,
+			dbName:      "3b81321b-1ec7-488c-96f7-72c42a968da6",
+			dbRole:      "",
+			wantName:    "3b81321b-1ec7-488c-96f7-72c42a968da6",
+			wantDesc:    "",
+			wantRole:    "",
+			wantChanged: false,
+		},
+		{
+			name:        "malformed card JSON — return unchanged, no panic",
+			card:        `{not json`,
+			dbName:      "Claude Code Agent",
+			dbRole:      "",
+			wantChanged: false,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			out, changed := reconcileAgentCardIdentity(
+				json.RawMessage(tc.card), wsID, tc.dbName, tc.dbRole,
+			)
+			if changed != tc.wantChanged {
+				t.Fatalf("changed = %v, want %v", changed, tc.wantChanged)
+			}
+			if !tc.wantChanged {
+				// Unchanged path must return the input bytes verbatim.
+				if string(out) != tc.card {
+					t.Fatalf("unchanged path mutated bytes:\n got  %s\n want %s", out, tc.card)
+				}
+				return
+			}
+			var got map[string]any
+			if err := json.Unmarshal(out, &got); err != nil {
+				t.Fatalf("output not valid JSON: %v (%s)", err, out)
+			}
+			if g, _ := got["name"].(string); g != tc.wantName {
+				t.Errorf("name = %q, want %q", g, tc.wantName)
+			}
+			if g, _ := got["description"].(string); g != tc.wantDesc {
+				t.Errorf("description = %q, want %q", g, tc.wantDesc)
+			}
+			if tc.wantRole != "" {
+				if g, _ := got["role"].(string); g != tc.wantRole {
+					t.Errorf("role = %q, want %q", g, tc.wantRole)
+				}
+			}
+		})
+	}
+}
+
+// TestReconcileAgentCardIdentity_PreservesOtherFields ensures the
+// reconcile is a minimal in-place patch — capabilities, version,
+// skills and any unknown future fields survive untouched.
+func TestReconcileAgentCardIdentity_PreservesOtherFields(t *testing.T) {
+	card := `{"name":"ws-uuid","description":"","version":"1.0.0",` +
+		`"capabilities":{"streaming":true,"pushNotifications":true},` +
+		`"skills":[{"id":"a","name":"a"}],"configuration_status":"ready"}`
+	out, changed := reconcileAgentCardIdentity(
+		json.RawMessage(card), "ws-uuid", "Friendly Name", "",
+	)
+	if !changed {
+		t.Fatal("expected changed = true")
+	}
+	var got map[string]any
+	if err := json.Unmarshal(out, &got); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
+	}
+	if got["version"] != "1.0.0" {
+		t.Errorf("version not preserved: %v", got["version"])
+	}
+	if got["configuration_status"] != "ready" {
+		t.Errorf("configuration_status not preserved: %v", got["configuration_status"])
+	}
+	caps, ok := got["capabilities"].(map[string]any)
+	if !ok || caps["streaming"] != true {
+		t.Errorf("capabilities not preserved: %v", got["capabilities"])
+	}
+	skills, ok := got["skills"].([]any)
+	if !ok || len(skills) != 1 {
+		t.Errorf("skills not preserved: %v", got["skills"])
+	}
+}

workspace-server/internal/handlers/delegation.go

@@ -2,6 +2,7 @@ package handlers
 
 import (
 	"context"
+	"database/sql"
 	"encoding/json"
 	"log"
 	"net/http"
@@ -162,8 +163,32 @@ func (h *DelegationHandler) Delegate(c *gin.Context) {
 		},
 	})
 
-	// Fire-and-forget: send A2A in background goroutine
-	go h.executeDelegation(ctx, sourceID, body.TargetID, delegationID, a2aBody)
+	// Fire-and-forget: send A2A in a background goroutine.
+	//
+	// internal#497 — the goroutine MUST NOT inherit the HTTP request's
+	// cancellation. `ctx` here is c.Request.Context(); the handler returns
+	// 202 a few lines below, which cancels that context immediately. Before
+	// this fix (regression ce2db75f) executeDelegation ran on the
+	// request-scoped ctx, so every DB op + proxy call in the detached
+	// goroutine failed `context canceled` the instant the 202 was written.
+	// That silently broke 100% of A2A peer delegations fleet-wide since
+	// 2026-05-12 (poll-mode peers never got their a2a_receive inbox row;
+	// lookupDeliveryMode swallowed the ctx error and defaulted to push).
+	//
+	// context.WithoutCancel detaches cancellation/deadline while PRESERVING
+	// all context values (trace/correlation/tenant ids that proxyA2ARequest
+	// and the broadcaster read off ctx) — this is the established pattern in
+	// this package (a2a_proxy.go:850, a2a_proxy_helpers.go:525,
+	// registry.go:822). The 30-minute ceiling matches the prior internal
+	// budget executeDelegation used before ce2db75f and the proxy's own
+	// absolute agent-dispatch ceiling (a2a_proxy.go forwardCtx).
+	delegationCtx, cancelDelegation := context.WithTimeout(
+		context.WithoutCancel(ctx), 30*time.Minute,
+	)
+	go func() {
+		defer cancelDelegation()
+		h.executeDelegation(delegationCtx, sourceID, body.TargetID, delegationID, a2aBody)
+	}()
 
 	// Broadcast event so canvas shows delegation in real-time
 	h.broadcaster.RecordAndBroadcast(ctx, string(events.EventDelegationSent), sourceID, map[string]interface{}{
@@ -698,7 +723,8 @@ func (h *DelegationHandler) listDelegationsFromLedger(ctx context.Context, works
 
 	var result []map[string]interface{}
 	for rows.Next() {
-		var delegationID, callerID, calleeID, taskPreview, status, resultPreview, errorDetail string
+		var delegationID, callerID, calleeID, taskPreview, status string
+		var resultPreview, errorDetail sql.NullString
 		var lastHeartbeat, deadline, createdAt, updatedAt *time.Time
 		if err := rows.Scan(
 			&delegationID, &callerID, &calleeID, &taskPreview,
@@ -717,11 +743,11 @@ func (h *DelegationHandler) listDelegationsFromLedger(ctx context.Context, works
 			"updated_at":    updatedAt,
 			"_ledger":       true, // marker so callers know this row is from the ledger
 		}
-		if resultPreview != "" {
-			entry["response_preview"] = textutil.TruncateBytes(resultPreview, 300)
+		if resultPreview.Valid && resultPreview.String != "" {
+			entry["response_preview"] = textutil.TruncateBytes(resultPreview.String, 300)
 		}
-		if errorDetail != "" {
-			entry["error"] = errorDetail
+		if errorDetail.Valid && errorDetail.String != "" {
+			entry["error"] = errorDetail.String
 		}
 		if lastHeartbeat != nil {
 			entry["last_heartbeat"] = lastHeartbeat

workspace-server/internal/handlers/delegation_list_test.go

@@ -145,6 +145,54 @@ func TestListDelegationsFromLedger_MultipleRows(t *testing.T) {
 	}
 }
 
+func TestListDelegationsFromLedger_NullsOmitted(t *testing.T) {
+	// last_heartbeat, deadline, result_preview, error_detail are all NULL.
+	// Handler must not panic and must omit those keys from the map.
+	mockDB, mock, err := sqlmock.New()
+	if err != nil {
+		t.Fatalf("failed to create sqlmock: %v", err)
+	}
+	prevDB := db.DB
+	db.DB = mockDB
+	t.Cleanup(func() { mockDB.Close(); db.DB = prevDB })
+
+	now := time.Now()
+	rows := sqlmock.NewRows([]string{
+		"delegation_id", "caller_id", "callee_id", "task_preview",
+		"status", "result_preview", "error_detail",
+		"last_heartbeat", "deadline", "created_at", "updated_at",
+	}).
+		AddRow("del-1", "ws-1", "ws-2", "task", "queued", nil, nil, nil, nil, now, now)
+	mock.ExpectQuery("SELECT .+ FROM delegations").
+		WithArgs("ws-1").
+		WillReturnRows(rows)
+
+	broadcaster := newTestBroadcaster()
+	wh := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
+	dh := NewDelegationHandler(wh, broadcaster)
+
+	got := dh.listDelegationsFromLedger(context.Background(), "ws-1")
+	if len(got) != 1 {
+		t.Fatalf("expected 1 entry, got %d", len(got))
+	}
+	e := got[0]
+	if _, ok := e["last_heartbeat"]; ok {
+		t.Error("last_heartbeat should be absent when NULL")
+	}
+	if _, ok := e["deadline"]; ok {
+		t.Error("deadline should be absent when NULL")
+	}
+	if _, ok := e["response_preview"]; ok {
+		t.Error("response_preview should be absent when NULL result_preview")
+	}
+	if _, ok := e["error"]; ok {
+		t.Error("error should be absent when NULL error_detail")
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("sqlmock expectations: %v", err)
+	}
+}
+
 func TestListDelegationsFromLedger_QueryError(t *testing.T) {
 	// Query failure returns nil — graceful fallback, no panic.
 	mockDB, mock, err := sqlmock.New()

workspace-server/internal/handlers/delegation_test.go

@@ -16,6 +16,65 @@ import (
 	"github.com/gin-gonic/gin"
 )
 
+// ---------- internal#497 regression: detached goroutine ctx must outlive the handler ----------
+
+// TestDelegate_DetachedContext_SurvivesRequestCancellation pins the
+// load-bearing invariant that regression ce2db75f violated: the context
+// handed to executeDelegation in the fire-and-forget goroutine must NOT be
+// cancelled when the HTTP handler returns 202 (which cancels
+// c.Request.Context()). Before the fix, executeDelegation ran on the
+// request-scoped ctx, so every DB op + proxy call failed `context
+// canceled` the instant the 202 was written — silently breaking 100% of
+// A2A peer delegations fleet-wide since 2026-05-12.
+//
+// This test asserts the exact ctx-derivation contract used by Delegate
+// (context.WithoutCancel(parent) + a timeout budget): the derived context
+// (a) stays alive after the parent is cancelled, and (b) still carries
+// parent values (trace/correlation/tenant ids the downstream proxy +
+// broadcaster read off ctx). It is intentionally DB-free and fast.
+func TestDelegate_DetachedContext_SurvivesRequestCancellation(t *testing.T) {
+	type ctxKey string
+	const traceKey ctxKey = "trace-id"
+
+	// Simulate c.Request.Context() carrying a correlation value.
+	parent, cancelParent := context.WithCancel(
+		context.WithValue(context.Background(), traceKey, "trace-abc-123"),
+	)
+
+	// Exact derivation Delegate uses for the detached goroutine.
+	delegationCtx, cancelDelegation := context.WithTimeout(
+		context.WithoutCancel(parent), 30*time.Minute,
+	)
+	defer cancelDelegation()
+
+	// The HTTP handler "returns 202" → request context is cancelled.
+	cancelParent()
+
+	if err := parent.Err(); err == nil {
+		t.Fatal("precondition: parent context should be cancelled after the handler returns")
+	}
+
+	// (a) Cancellation MUST NOT propagate to the detached context.
+	select {
+	case <-delegationCtx.Done():
+		t.Fatalf("regression: detached delegation ctx was cancelled by the handler returning (err=%v) — executeDelegation would fail every DB op with `context canceled`", delegationCtx.Err())
+	default:
+		// alive — correct
+	}
+
+	// (b) Parent values MUST still be readable (WithoutCancel preserves
+	// values; trace/correlation/tenant ids the proxy + broadcaster use).
+	if got, _ := delegationCtx.Value(traceKey).(string); got != "trace-abc-123" {
+		t.Errorf("detached ctx lost the parent trace value: got %q, want %q", got, "trace-abc-123")
+	}
+
+	// And it still has a real deadline (the 30m budget), so it is not an
+	// unbounded background context.
+	if _, hasDeadline := delegationCtx.Deadline(); !hasDeadline {
+		t.Error("detached ctx must carry the 30-minute timeout budget, but has no deadline")
+	}
+}
+
 // ---------- Delegate: missing target_id → 400 ----------
 
 func TestDelegate_MissingTargetID(t *testing.T) {

workspace-server/internal/handlers/external_connection.go

@@ -646,8 +646,12 @@ const externalOpenClawTemplate = `# OpenClaw MCP config — outbound tool path.
 # external machine today, pair with the Python SDK tab.
 
 # 1. Install openclaw CLI + the workspace runtime wheel:
+#    The version pin (>=0.1.999) ensures the "molecule-mcp" console
+#    script is present — it is what keeps the workspace ALIVE on canvas
+#    (register-on-startup + 20s heartbeat). Older versions only ship
+#    a2a_mcp_server which does not heartbeat.
 npm install -g openclaw@latest
-pip install molecule-ai-workspace-runtime
+pip install "molecule-ai-workspace-runtime>=0.1.999"
 
 # 2. Onboard openclaw against your model provider (one-time setup).
 #    --non-interactive needs an explicit --provider + --model so it

workspace-server/internal/handlers/handlers_test.go

@@ -61,6 +61,11 @@ func drainTestAsync() {
 // setupTestDB creates a sqlmock DB and assigns it to the global db.DB.
 // It also disables the SSRF URL check so that httptest.NewServer loopback
 // URLs and fake hostnames (*.example) used in tests don't trigger rejections.
+//
+// IMPORTANT: db.DB is saved before assignment and restored via t.Cleanup so
+// that tests running after this one are not polluted by a closed mock.
+// This is the single root cause of the systemic CI/Platform (Go) failures on
+// main HEAD 8026f020 (mc#975).
 func setupTestDB(t *testing.T) sqlmock.Sqlmock {
 	t.Helper()
 	mockDB, mock, err := sqlmock.New()
@@ -98,6 +103,11 @@ func setupTestDB(t *testing.T) sqlmock.Sqlmock {
 	return mock
 }
 
+func waitForHandlerAsyncBeforeDBCleanup(t *testing.T, h *WorkspaceHandler) {
+	t.Helper()
+	t.Cleanup(h.waitAsyncForTest)
+}
+
 // setupTestRedis creates a miniredis instance and assigns it to the global db.RDB.
 func setupTestRedis(t *testing.T) *miniredis.Miniredis {
 	t.Helper()
@@ -397,6 +407,11 @@ func TestWorkspaceCreate(t *testing.T) {
 }
 
 func TestBuildProvisionerConfig_IncludesAwarenessSettings(t *testing.T) {
+	mock := setupTestDB(t)
+	mock.ExpectQuery(`SELECT digest FROM runtime_image_pins`).
+		WithArgs("claude-code").
+		WillReturnError(sql.ErrNoRows)
+
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", "/tmp/configs")
 

workspace-server/internal/handlers/instructions_test.go

@@ -2,10 +2,12 @@ package handlers
 
 import (
 	"bytes"
+	"context"
 	"encoding/json"
 	"errors"
 	"net/http"
 	"net/http/httptest"
+	"regexp"
 	"testing"
 	"time"
 
@@ -80,117 +82,135 @@ func TestInstructionsList_ByWorkspaceID(t *testing.T) {
 	if w.Code != http.StatusOK {
 		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var out []Instruction
-	if err := json.Unmarshal(w.Body.Bytes(), &out); err != nil {
-		t.Fatalf("response not valid JSON: %v", err)
+	var result []Instruction
+	if err := json.Unmarshal(w.Body.Bytes(), &result); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
 	}
-	if len(out) != 2 {
-		t.Errorf("expected 2 instructions, got %d", len(out))
+	if len(result) != 2 {
+		t.Fatalf("expected 2 instructions, got %d", len(result))
 	}
-	if out[0].Scope != "global" {
-		t.Errorf("first row scope: expected global, got %s", out[0].Scope)
+	if result[0].Scope != "global" || result[1].Scope != "workspace" {
+		t.Fatalf("expected global then workspace instructions, got %#v", result)
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Fatalf("unmet expectations: %v", err)
 	}
 }
 
-func TestInstructionsList_ByScope(t *testing.T) {
+func TestInstructionsHandler_List_WithScopeFilter(t *testing.T) {
 	mock := setupTestDB(t)
-	h := NewInstructionsHandler()
+	handler := NewInstructionsHandler()
 
-	w, c := newGetRequest("/instructions?scope=global")
-	c.Request = httptest.NewRequest(http.MethodGet, "/instructions?scope=global", nil)
+	rows := sqlmock.NewRows([]string{
+		"id", "scope", "scope_target", "title", "content", "priority", "enabled", "created_at", "updated_at",
+	}).AddRow("inst-1", "global", nil, "Be kind", "Always be kind", 10, true,
+		time.Now(), time.Now())
 
-	rows := sqlmock.NewRows(instructionCols).
-		AddRow("inst-g", "global", nil, "Global Rule", "Follow policy.", 10, true, time.Now(), time.Now())
-	mock.ExpectQuery("SELECT id, scope, scope_target, title, content, priority, enabled, created_at, updated_at FROM platform_instructions WHERE 1=1").
+	mock.ExpectQuery(regexp.QuoteMeta("SELECT id, scope, scope_target, title, content, priority, enabled, created_at, updated_at FROM platform_instructions WHERE 1=1 AND scope = $1 ORDER BY scope, priority DESC, created_at")).
 		WithArgs("global").
 		WillReturnRows(rows)
 
-	h.List(c)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("GET", "/instructions?scope=global", nil)
+
+	handler.List(c)
 
 	if w.Code != http.StatusOK {
-		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
+		t.Fatalf("expected 200, got %d", w.Code)
 	}
-	var out []Instruction
-	if err := json.Unmarshal(w.Body.Bytes(), &out); err != nil {
-		t.Fatalf("response not valid JSON: %v", err)
+	var result []Instruction
+	if err := json.Unmarshal(w.Body.Bytes(), &result); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
 	}
-	if len(out) != 1 || out[0].Scope != "global" {
-		t.Errorf("unexpected response: %v", out)
+	if len(result) != 1 {
+		t.Fatalf("expected 1 instruction, got %d", len(result))
+	}
+	if result[0].Scope != "global" {
+		t.Errorf("expected scope 'global', got %q", result[0].Scope)
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Fatalf("unmet expectations: %v", err)
 	}
 }
 
-func TestInstructionsList_AllNoParams(t *testing.T) {
+func TestInstructionsHandler_List_WithWorkspaceID(t *testing.T) {
 	mock := setupTestDB(t)
-	h := NewInstructionsHandler()
+	handler := NewInstructionsHandler()
+	wsID := "ws-test-123"
 
-	w, c := newGetRequest("/instructions")
+	rows := sqlmock.NewRows([]string{
+		"id", "scope", "scope_target", "title", "content", "priority", "enabled", "created_at", "updated_at",
+	}).AddRow("inst-1", "global", nil, "Global rule", "Stay safe", 5, true,
+		time.Now(), time.Now()).
+		AddRow("inst-2", "workspace", &wsID, "WS rule", "Use HTTPS", 10, true,
+			time.Now(), time.Now())
 
-	rows := sqlmock.NewRows(instructionCols)
-	mock.ExpectQuery("SELECT id, scope, scope_target, title, content, priority, enabled, created_at, updated_at FROM platform_instructions WHERE 1=1").
+	mock.ExpectQuery("SELECT id, scope, scope_target, title, content, priority, enabled, created_at, updated_at FROM platform_instructions WHERE enabled = true AND \\(").
+		WithArgs(wsID).
 		WillReturnRows(rows)
 
-	h.List(c)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("GET", "/instructions?workspace_id="+wsID, nil)
+
+	handler.List(c)
 
 	if w.Code != http.StatusOK {
-		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
+		t.Fatalf("expected 200, got %d", w.Code)
 	}
-	var out []Instruction
-	if err := json.Unmarshal(w.Body.Bytes(), &out); err != nil {
-		t.Fatalf("response not valid JSON: %v", err)
+	var result []Instruction
+	if err := json.Unmarshal(w.Body.Bytes(), &result); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
 	}
-	// Empty slice, not nil
-	if out == nil {
-		t.Error("expected empty slice, got nil")
+	if len(result) != 2 {
+		t.Fatalf("expected 2 instructions, got %d", len(result))
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Fatalf("unmet expectations: %v", err)
 	}
 }
 
-func TestInstructionsList_DBError(t *testing.T) {
+func TestInstructionsHandler_List_QueryError(t *testing.T) {
 	mock := setupTestDB(t)
-	h := NewInstructionsHandler()
-
-	w, c := newGetRequest("/instructions")
-	c.Request = httptest.NewRequest(http.MethodGet, "/instructions", nil)
+	handler := NewInstructionsHandler()
 
 	mock.ExpectQuery("SELECT id, scope, scope_target, title, content, priority, enabled, created_at, updated_at FROM platform_instructions WHERE 1=1").
-		WillReturnError(errors.New("connection refused"))
+		WillReturnError(context.DeadlineExceeded)
 
-	h.List(c)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("GET", "/instructions", nil)
+
+	handler.List(c)
 
 	if w.Code != http.StatusInternalServerError {
-		t.Fatalf("expected 500, got %d: %s", w.Code, w.Body.String())
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Fatalf("expected 500, got %d", w.Code)
 	}
 }
 
-// ─── Create ───────────────────────────────────────────────────────────────────
+// ── Create ──────────────────────────────────────────────────────────────────────
 
-func TestInstructionsCreate_ValidGlobal(t *testing.T) {
+func TestInstructionsHandler_Create_Success(t *testing.T) {
 	mock := setupTestDB(t)
-	h := NewInstructionsHandler()
-
-	w, c := newPostRequest("/instructions", map[string]interface{}{
-		"scope":    "global",
-		"title":    "Be Helpful",
-		"content":  "Always be helpful to the user.",
-		"priority": 10,
-	})
+	handler := NewInstructionsHandler()
 
 	mock.ExpectQuery("INSERT INTO platform_instructions").
-		WithArgs("global", nil, "Be Helpful", "Always be helpful to the user.", 10).
-		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow("new-inst-1"))
+		WithArgs("global", nil, "Be kind", "Always be kind", 5).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow("new-inst-id"))
 
-	h.Create(c)
+	body, _ := json.Marshal(map[string]interface{}{
+		"scope":    "global",
+		"title":    "Be kind",
+		"content":  "Always be kind",
+		"priority": 5,
+	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/instructions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Create(c)
 
 	if w.Code != http.StatusCreated {
 		t.Fatalf("expected 201, got %d: %s", w.Code, w.Body.String())
@@ -199,8 +219,8 @@ func TestInstructionsCreate_ValidGlobal(t *testing.T) {
 	if err := json.Unmarshal(w.Body.Bytes(), &out); err != nil {
 		t.Fatalf("response not valid JSON: %v", err)
 	}
-	if out["id"] != "new-inst-1" {
-		t.Errorf("expected id new-inst-1, got %s", out["id"])
+	if out["id"] != "new-inst-id" {
+		t.Errorf("expected id new-inst-id, got %s", out["id"])
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
 		t.Errorf("unmet expectations: %v", err)
@@ -299,56 +319,65 @@ func TestInstructionsCreate_InvalidScope(t *testing.T) {
 	}
 }
 
-func TestInstructionsCreate_WorkspaceScopeNoTarget(t *testing.T) {
+func TestInstructionsHandler_Create_WorkspaceScopeMissingScopeTarget(t *testing.T) {
 	setupTestDB(t)
-	h := NewInstructionsHandler()
+	handler := NewInstructionsHandler()
 
-	w, c := newPostRequest("/instructions", map[string]interface{}{
+	body, _ := json.Marshal(map[string]interface{}{
 		"scope":   "workspace",
-		"title":   "Missing Target",
-		"content": "Workspace scope without scope_target.",
+		"title":   "Test",
+		"content": "Test content",
 	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/instructions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
 
-	h.Create(c)
+	handler.Create(c)
 
 	if w.Code != http.StatusBadRequest {
 		t.Fatalf("expected 400, got %d: %s", w.Code, w.Body.String())
 	}
 }
 
-func TestInstructionsCreate_ContentTooLong(t *testing.T) {
+func TestInstructionsHandler_Create_ContentTooLong(t *testing.T) {
 	setupTestDB(t)
-	h := NewInstructionsHandler()
+	handler := NewInstructionsHandler()
 
-	// Build a string longer than maxInstructionContentLen (8192).
-	longContent := string(make([]byte, maxInstructionContentLen+1))
-
-	w, c := newPostRequest("/instructions", map[string]interface{}{
+	longContent := string(bytes.Repeat([]byte("x"), 8193))
+	body, _ := json.Marshal(map[string]interface{}{
 		"scope":   "global",
-		"title":   "Too Long",
+		"title":   "Test",
 		"content": longContent,
 	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/instructions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
 
-	h.Create(c)
+	handler.Create(c)
 
 	if w.Code != http.StatusBadRequest {
 		t.Fatalf("expected 400, got %d: %s", w.Code, w.Body.String())
 	}
 }
 
-func TestInstructionsCreate_TitleTooLong(t *testing.T) {
+func TestInstructionsHandler_Create_TitleTooLong(t *testing.T) {
 	setupTestDB(t)
-	h := NewInstructionsHandler()
+	handler := NewInstructionsHandler()
 
-	longTitle := string(make([]byte, 201))
-
-	w, c := newPostRequest("/instructions", map[string]interface{}{
+	longTitle := string(bytes.Repeat([]byte("x"), 201))
+	body, _ := json.Marshal(map[string]interface{}{
 		"scope":   "global",
 		"title":   longTitle,
-		"content": "Short content.",
+		"content": "Short content",
 	})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("POST", "/instructions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
 
-	h.Create(c)
+	handler.Create(c)
 
 	if w.Code != http.StatusBadRequest {
 		t.Fatalf("expected 400, got %d: %s", w.Code, w.Body.String())
@@ -842,43 +871,250 @@ func TestInstructionsResolve_ScopeTransitionOnlyGlobal(t *testing.T) {
 	if w.Code != http.StatusOK {
 		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var out struct {
-		Instructions string `json:"instructions"`
-	}
-	if err := json.Unmarshal(w.Body.Bytes(), &out); err != nil {
-		t.Fatalf("response not valid JSON: %v", err)
-	}
-	// Two global instructions share one section header.
-	if bytes.Count([]byte(out.Instructions), []byte("Platform-Wide Rules")) != 1 {
-		t.Error("expect exactly one 'Platform-Wide Rules' header for consecutive global rows")
-	}
 	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Fatalf("unmet expectations: %v", err)
 	}
 }
 
-// ─── Update: empty body (all nil — no-op update) ─────────────────────────────
-
-func TestInstructionsUpdate_EmptyBody(t *testing.T) {
+func TestInstructionsHandler_Update_NotFound(t *testing.T) {
 	mock := setupTestDB(t)
-	h := NewInstructionsHandler()
+	handler := NewInstructionsHandler()
 
-	instID := "inst-empty-update"
-	w, c := newPutRequest("/instructions/"+instID, map[string]interface{}{})
-	c.Params = []gin.Param{{Key: "id", Value: instID}}
+	mock.ExpectExec(regexp.QuoteMeta("UPDATE platform_instructions SET\n\t\t\t\ttitle = COALESCE($2, title),\n\t\t\t\tcontent = COALESCE($3, content),\n\t\t\t\tpriority = COALESCE($4, priority),\n\t\t\t\tenabled = COALESCE($5, enabled),\n\t\t\t\tupdated_at = NOW()\n\t\t\t\tWHERE id = $1")).
+		WithArgs("nonexistent", sqlmock.AnyArg(), nil, nil, nil).
+		WillReturnResult(sqlmock.NewResult(0, 0))
 
-	// COALESCE(nil, ...) = unchanged; still updates updated_at.
-	// Args order: ($1=id, $2=title, $3=content, $4=priority, $5=enabled)
-	mock.ExpectExec("UPDATE platform_instructions SET").
-		WithArgs(instID, sqlmock.AnyArg(), sqlmock.AnyArg(), sqlmock.AnyArg(), sqlmock.AnyArg()).
+	body, _ := json.Marshal(map[string]interface{}{"title": "Updated title"})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "nonexistent"}}
+	c.Request = httptest.NewRequest("PUT", "/instructions/nonexistent", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Update(c)
+
+	if w.Code != http.StatusNotFound {
+		t.Fatalf("expected 404, got %d: %s", w.Code, w.Body.String())
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+func TestInstructionsHandler_Update_ContentTooLong(t *testing.T) {
+	setupTestDB(t)
+	handler := NewInstructionsHandler()
+
+	longContent := string(bytes.Repeat([]byte("x"), 8193))
+	body, _ := json.Marshal(map[string]interface{}{"content": longContent})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "inst-1"}}
+	c.Request = httptest.NewRequest("PUT", "/instructions/inst-1", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Update(c)
+
+	if w.Code != http.StatusBadRequest {
+		t.Fatalf("expected 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestInstructionsHandler_Update_TitleTooLong(t *testing.T) {
+	setupTestDB(t)
+	handler := NewInstructionsHandler()
+
+	longTitle := string(bytes.Repeat([]byte("x"), 201))
+	body, _ := json.Marshal(map[string]interface{}{"title": longTitle})
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "inst-1"}}
+	c.Request = httptest.NewRequest("PUT", "/instructions/inst-1", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Update(c)
+
+	if w.Code != http.StatusBadRequest {
+		t.Fatalf("expected 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+// ── Delete ─────────────────────────────────────────────────────────────────────
+
+func TestInstructionsHandler_Delete_Success(t *testing.T) {
+	mock := setupTestDB(t)
+	handler := NewInstructionsHandler()
+
+	mock.ExpectExec(regexp.QuoteMeta("DELETE FROM platform_instructions WHERE id = $1")).
+		WithArgs("inst-1").
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
-	h.Update(c)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "inst-1"}}
+	c.Request = httptest.NewRequest("DELETE", "/instructions/inst-1", nil)
+
+	handler.Delete(c)
 
 	if w.Code != http.StatusOK {
-		t.Fatalf("expected 200 for empty body, got %d: %s", w.Code, w.Body.String())
+		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+func TestInstructionsHandler_Delete_NotFound(t *testing.T) {
+	mock := setupTestDB(t)
+	handler := NewInstructionsHandler()
+
+	mock.ExpectExec(regexp.QuoteMeta("DELETE FROM platform_instructions WHERE id = $1")).
+		WithArgs("nonexistent").
+		WillReturnResult(sqlmock.NewResult(0, 0))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "nonexistent"}}
+	c.Request = httptest.NewRequest("DELETE", "/instructions/nonexistent", nil)
+
+	handler.Delete(c)
+
+	if w.Code != http.StatusNotFound {
+		t.Fatalf("expected 404, got %d: %s", w.Code, w.Body.String())
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+// ── Resolve ────────────────────────────────────────────────────────────────────
+
+func TestInstructionsHandler_Resolve_Empty(t *testing.T) {
+	mock := setupTestDB(t)
+	handler := NewInstructionsHandler()
+	wsID := "ws-resolve-1"
+
+	mock.ExpectQuery("SELECT scope, title, content FROM platform_instructions WHERE enabled = true AND").
+		WithArgs(wsID).
+		WillReturnRows(sqlmock.NewRows([]string{"scope", "title", "content"}))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: wsID}}
+	c.Request = httptest.NewRequest("GET", "/workspaces/"+wsID+"/instructions/resolve", nil)
+
+	handler.Resolve(c)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
+	}
+	var resp map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
+	}
+	if resp["workspace_id"] != wsID {
+		t.Errorf("expected workspace_id %q, got %v", wsID, resp["workspace_id"])
+	}
+	if resp["instructions"] != "" {
+		t.Errorf("expected empty instructions, got %q", resp["instructions"])
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+func TestInstructionsHandler_Resolve_WithInstructions(t *testing.T) {
+	mock := setupTestDB(t)
+	handler := NewInstructionsHandler()
+	wsID := "ws-resolve-2"
+
+	rows := sqlmock.NewRows([]string{"scope", "title", "content"}).
+		AddRow("global", "Be safe", "No SSRF").
+		AddRow("workspace", "WS Rule", "Use HTTPS")
+
+	mock.ExpectQuery("SELECT scope, title, content FROM platform_instructions WHERE enabled = true AND").
+		WithArgs(wsID).
+		WillReturnRows(rows)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: wsID}}
+	c.Request = httptest.NewRequest("GET", "/workspaces/"+wsID+"/instructions/resolve", nil)
+
+	handler.Resolve(c)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
+	}
+	var resp map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
+	}
+	instructions, ok := resp["instructions"].(string)
+	if !ok {
+		t.Fatalf("instructions field is not a string: %T", resp["instructions"])
+	}
+	if instructions == "" {
+		t.Fatalf("expected non-empty instructions")
+	}
+	// Verify scope headers are present
+	if !bytes.Contains([]byte(instructions), []byte("Platform-Wide Rules")) {
+		t.Errorf("expected 'Platform-Wide Rules' header in instructions")
+	}
+	if !bytes.Contains([]byte(instructions), []byte("Role-Specific Rules")) {
+		t.Errorf("expected 'Role-Specific Rules' header in instructions")
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+func TestInstructionsHandler_Resolve_MissingWorkspaceID(t *testing.T) {
+	setupTestDB(t)
+	handler := NewInstructionsHandler()
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: ""}}
+	c.Request = httptest.NewRequest("GET", "/workspaces//instructions/resolve", nil)
+
+	handler.Resolve(c)
+
+	if w.Code != http.StatusBadRequest {
+		t.Fatalf("expected 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+// scanInstructions is called by the List handler — verify it handles
+// rows.Err() gracefully without panicking.
+func TestInstructionsHandler_List_ScanErrorContinues(t *testing.T) {
+	mock := setupTestDB(t)
+	handler := NewInstructionsHandler()
+
+	rows := sqlmock.NewRows([]string{
+		"id", "scope", "scope_target", "title", "content", "priority", "enabled", "created_at", "updated_at",
+	}).AddRow("inst-1", "global", nil, "Good", "Content here", 5, true, time.Now(), time.Now()).
+		RowError(1, context.DeadlineExceeded) // error on row 2 (if it existed)
+
+	mock.ExpectQuery("SELECT id, scope, scope_target, title, content, priority, enabled, created_at, updated_at FROM platform_instructions WHERE 1=1").
+		WillReturnRows(rows)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequest("GET", "/instructions", nil)
+
+	handler.List(c)
+
+	// Should still return 200 and the one valid row
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d", w.Code)
+	}
+	var result []Instruction
+	if err := json.Unmarshal(w.Body.Bytes(), &result); err != nil {
+		t.Fatalf("invalid JSON: %v", err)
+	}
+	// The valid row should still be returned (error is logged, not fatal)
+	if len(result) != 1 {
+		t.Fatalf("expected 1 instruction despite row error, got %d", len(result))
 	}
 }

workspace-server/internal/handlers/org_helpers.go

@@ -15,6 +15,7 @@ import (
 
 	"gopkg.in/yaml.v3"
 )
+
 // resolvePromptRef reads a prompt body from either an inline string or a
 // file ref relative to the workspace's files_dir. Inline always wins when
 // both are non-empty (caller-provided inline is more authoritative than a
@@ -176,7 +177,7 @@ func isEnvIdentPart(c byte) bool {
 	return isEnvIdentStart(c) || (c >= '0' && c <= '9')
 }
 
-// loadWorkspaceEnv reads the org root .env and the workspace-specific .env .env and the workspace-specific .env
+// loadWorkspaceEnv reads the org root .env and the workspace-specific .env
 // (workspace overrides org root). Used by both secret injection and channel
 // config expansion.
 //

workspace-server/internal/handlers/org_helpers_pure_test.go

@@ -104,8 +104,8 @@ func TestHasUnresolvedVarRef_Resolved(t *testing.T) {
 		// documents this design choice; callers who need empty=resolved should
 		// pre-process the output before calling hasUnresolvedVarRef.
 		{"${VAR}", "", true},
-		{"${VAR}", "value", false},                    // var replaced
-		{"$VAR", "value", false},                      // bare var replaced
+		{"${VAR}", "value", false}, // var replaced
+		{"$VAR", "value", false},   // bare var replaced
 		{"prefix${VAR}suffix", "prefixvaluesuffix", false},
 		{"${A}${B}", "ab", false},
 		// FOO=FOO and BAR=BAR — both vars found and replaced. Expanded output
@@ -125,14 +125,14 @@ func TestHasUnresolvedVarRef_Resolved(t *testing.T) {
 func TestHasUnresolvedVarRef_Unresolved(t *testing.T) {
 	// Expansion left the refs intact → unresolved.
 	cases := []struct {
-		orig    string
+		orig     string
 		expanded string
 	}{
-		{"${VAR}", "${VAR}"},       // untouched
-		{"$VAR", "$VAR"},           // bare untouched
+		{"${VAR}", "${VAR}"}, // untouched
+		{"$VAR", "$VAR"},     // bare untouched
 		{"prefix${VAR}suffix", "prefix${VAR}suffix"},
-		{"${A}${B}", "${A}${B}"},   // both unresolved
-		{"${FOO}", ""},             // empty result with var ref in original
+		{"${A}${B}", "${A}${B}"}, // both unresolved
+		{"${FOO}", ""},           // empty result with var ref in original
 	}
 	for _, tc := range cases {
 		t.Run(tc.orig, func(t *testing.T) {
@@ -205,8 +205,8 @@ func TestMergeCategoryRouting_WorkspaceOverrides(t *testing.T) {
 		"ui":       {"Frontend Engineer"},
 	}
 	ws := map[string][]string{
-		"security": {"SRE Team"}, // narrows
-		"ui":       {},           // drops
+		"security": {"SRE Team"},      // narrows
+		"ui":       {},                // drops
 		"infra":    {"Platform Team"}, // adds
 	}
 	r := mergeCategoryRouting(defaults, ws)
@@ -467,6 +467,44 @@ func TestExpandWithEnv_PartiallyPresent(t *testing.T) {
 	assert.Equal(t, "yes and ${NOT_SET}", result)
 }
 
+func TestExpandWithEnv_EmbeddedMissingProcessEnvStaysLiteral(t *testing.T) {
+	t.Setenv("MOL_TEST_EMBEDDED_MISSING", "")
+
+	result := expandWithEnv("prefix/${MOL_TEST_EMBEDDED_MISSING}/suffix", map[string]string{})
+	assert.Equal(t, "prefix/${MOL_TEST_EMBEDDED_MISSING}/suffix", result)
+}
+
+// POSIX identifier guard regression tests (CWE-78 fix).
+// Keys not starting with [a-zA-Z_] must not be looked up in env or os.Getenv.
+func TestExpandWithEnv_DigitPrefix_NotExpanded(t *testing.T) {
+	// ${0}, ${5}, ${1VAR} — numeric prefix → not a valid shell identifier.
+	// Guard must return "$0", "$5", "$1VAR" literally; no env lookup.
+	cases := []struct {
+		input string
+		want  string
+	}{
+		{"${0}", "$0"},
+		{"${5}", "$5"},
+		{"${1VAR}", "$1VAR"},
+		{"prefix ${0} suffix", "prefix $0 suffix"},
+		{"$0", "$0"},
+		{"$5", "$5"},
+		{"HOME=${HOME}", "HOME=${HOME}"}, // HOME is valid but embedded in larger string
+	}
+	for _, tc := range cases {
+		t.Run(tc.input, func(t *testing.T) {
+			got := expandWithEnv(tc.input, map[string]string{})
+			assert.Equal(t, tc.want, got)
+		})
+	}
+}
+
+func TestExpandWithEnv_EmptyKey_ReturnsDollar(t *testing.T) {
+	// ${} → "$" (empty key, guard returns "$")
+	result := expandWithEnv("value=${}", map[string]string{})
+	assert.Equal(t, "value=$", result)
+}
+
 // mergeCategoryRouting tests — unions defaults with per-workspace routing.
 
 // ── Additional coverage: mergeCategoryRouting ──────────────────────
@@ -546,8 +584,8 @@ func TestRenderCategoryRoutingYAML_SingleCategory(t *testing.T) {
 
 func TestRenderCategoryRoutingYAML_MultipleCategoriesSorted(t *testing.T) {
 	routing := map[string][]string{
-		"zebra":   {"RoleZ"},
-		"alpha":   {"RoleA"},
+		"zebra":      {"RoleZ"},
+		"alpha":      {"RoleA"},
 		"middleware": {"RoleM"},
 	}
 	result, err := renderCategoryRoutingYAML(routing)

workspace-server/internal/handlers/plugins_install_eic_test.go

@@ -342,6 +342,11 @@ func TestPluginInstall_InstanceLookupError_Returns503(t *testing.T) {
 // ---------- dispatch: uninstall ----------
 
 func TestPluginUninstall_SaaS_DispatchesToEIC(t *testing.T) {
+	mock := setupTestDB(t)
+	mock.ExpectExec("DELETE FROM workspace_plugins WHERE workspace_id").
+		WithArgs("ws-1", "browser-automation").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+
 	stubReadPluginManifestViaEIC(t, func(ctx context.Context, instanceID, runtime, pluginName string) ([]byte, error) {
 		return []byte("name: browser-automation\nskills:\n  - browse\n"), nil
 	})

workspace-server/internal/handlers/plugins_install_test.go

@@ -0,0 +1,53 @@
+package handlers
+
+// plugins_install_test.go — additional coverage for plugins_install.go.
+//
+// Gaps filled vs. existing test files:
+//   - plugins_install_external_test.go: Install + Uninstall 422 (external runtime) ✓ covered
+//   - plugins_test.go:                 Install 400 (missing source, invalid body, etc.) ✓ covered
+//                                    Uninstall 400 (invalid plugin name, empty name) ✓ covered
+//                                    Download auth gate ✓ covered
+//   - org_import_helpers_test.go:      countWorkspaces, envRequirementKey, sanitizeEnvMembers,
+//                                        flattenAndSortRequirements, collectOrgEnv ✓ covered
+//
+// New test added here:
+//   - Uninstall 503: container not running, no SaaS dispatch.
+//
+// NOTE: validateWorkspaceID is not called inside the Install/Uninstall handlers.
+// UUID validation is the responsibility of the WorkspaceAuth middleware, so no
+// 400 test is needed here for UUID format.
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+// TestPluginUninstall_ContainerNotRunning_Returns503 exercises the 503 path
+// where neither a local Docker container nor a SaaS instance-id dispatch
+// resolves. The handler must return "workspace container not running" — NOT a
+// generic 500 or a misleading 422 (external-runtime) message.
+func TestPluginUninstall_ContainerNotRunning_Returns503(t *testing.T) {
+	// No docker client + no instance-id lookup → falls through to 503.
+	h := NewPluginsHandler(t.TempDir(), nil, nil)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{
+		{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"},
+		{Key: "name", Value: "some-plugin"},
+	}
+	c.Request = httptest.NewRequest("DELETE",
+		"/workspaces/550e8400-e29b-41d4-a716-446655440000/plugins/some-plugin", nil)
+
+	h.Uninstall(c)
+
+	require.Equal(t, http.StatusServiceUnavailable, w.Code)
+	var body map[string]string
+	json.Unmarshal(w.Body.Bytes(), &body)
+	require.Equal(t, "workspace container not running", body["error"])
+}

workspace-server/internal/handlers/plugins_sources_test.go

@@ -1,55 +0,0 @@
-package handlers
-
-import (
-	"encoding/json"
-	"net/http"
-	"net/http/httptest"
-	"testing"
-
-	"github.com/gin-gonic/gin"
-)
-
-// ListSources is the only exported function in plugins_sources.go.
-// It calls h.sources.Schemes() and returns the result verbatim,
-// so the test verifies the handler correctly serialises whatever
-// the real registry provides.
-func TestListSources_ReturnsSchemes(t *testing.T) {
-	// Use a real handler — the registry is deterministic (local + github).
-	h := NewPluginsHandler(t.TempDir(), nil, nil)
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Request = httptest.NewRequest("GET", "/plugins/sources", nil)
-
-	h.ListSources(c)
-
-	if w.Code != http.StatusOK {
-		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
-	}
-
-	var body struct {
-		Schemes []string `json:"schemes"`
-	}
-	if err := json.Unmarshal(w.Body.Bytes(), &body); err != nil {
-		t.Fatalf("failed to unmarshal response: %v", err)
-	}
-
-	// The default registry registers local + github resolvers.
-	if len(body.Schemes) < 1 {
-		t.Fatalf("expected at least 1 scheme, got %d: %v", len(body.Schemes), body.Schemes)
-	}
-
-	// Verify stability — same call always returns same result.
-	w2 := httptest.NewRecorder()
-	c2, _ := gin.CreateTestContext(w2)
-	c2.Request = httptest.NewRequest("GET", "/plugins/sources", nil)
-	h.ListSources(c2)
-
-	var body2 struct {
-		Schemes []string `json:"schemes"`
-	}
-	json.Unmarshal(w2.Body.Bytes(), &body2)
-	if len(body.Schemes) != len(body2.Schemes) {
-		t.Errorf("Schemes() is not stable: first=%v, second=%v", body.Schemes, body2.Schemes)
-	}
-}

workspace-server/internal/handlers/plugins_test.go

@@ -629,6 +629,9 @@ func TestPluginInstall_RejectsUnknownScheme(t *testing.T) {
 }
 
 func TestPluginInstall_LocalSourceReachesContainerLookup(t *testing.T) {
+	mock := setupTestDB(t)
+	expectAllowlistAllowAll(mock)
+
 	base := t.TempDir()
 	pluginDir := filepath.Join(base, "demo")
 	_ = os.MkdirAll(pluginDir, 0o755)
@@ -955,14 +958,14 @@ func TestLogInstallLimitsOnce(t *testing.T) {
 
 func TestRegexpEscapeForAwk(t *testing.T) {
 	cases := map[string]string{
-		"my-plugin":                 `my-plugin`,
-		"# Plugin: foo /":           `# Plugin: foo \/`,
-		"# Plugin: a.b /":           `# Plugin: a\.b \/`,
-		"foo[bar]":                  `foo\[bar\]`,
-		"a*b+c?":                    `a\*b\+c\?`,
-		"path|with|pipes":           `path\|with\|pipes`,
-		`back\slash`:                `back\\slash`,
-		"":                          ``,
+		"my-plugin":       `my-plugin`,
+		"# Plugin: foo /": `# Plugin: foo \/`,
+		"# Plugin: a.b /": `# Plugin: a\.b \/`,
+		"foo[bar]":        `foo\[bar\]`,
+		"a*b+c?":          `a\*b\+c\?`,
+		"path|with|pipes": `path\|with\|pipes`,
+		`back\slash`:      `back\\slash`,
+		"":                ``,
 	}
 	for in, want := range cases {
 		got := regexpEscapeForAwk(in)
@@ -1247,7 +1250,7 @@ func TestPluginDownload_GithubSchemeStreamsTarball(t *testing.T) {
 		scheme: "github",
 		fetchFn: func(_ context.Context, _ string, dst string) (string, error) {
 			files := map[string]string{
-				"plugin.yaml":            "name: remote-plugin\nversion: 1.0.0\n",
+				"plugin.yaml":             "name: remote-plugin\nversion: 1.0.0\n",
 				"skills/x/SKILL.md":       "---\nname: x\n---\n",
 				"adapters/claude_code.py": "from plugins_registry.builtins import AgentskillsAdaptor as Adaptor\n",
 			}

workspace-server/internal/handlers/registry.go

@@ -327,7 +327,33 @@ func (h *RegistryHandler) Register(c *gin.Context) {
 		}
 	}
 
-	agentCardStr := string(payload.AgentCard)
+	// Reconcile the runtime-supplied card's identity fields against the
+	// trusted workspaces row before storing. The runtime builds its card
+	// from config.name, which the CP-regenerated /configs/config.yaml
+	// sets to the workspace UUID — so without this the stored card
+	// served at /.well-known/agent-card.json and returned to peers via
+	// agent_card_url has name = UUID, description = "", role = null even
+	// though the operator-controlled workspaces.name holds the friendly
+	// name the canvas shows. We only FILL gaps from the DB (never
+	// downgrade a card that already carries a real name); identity stays
+	// platform-controlled — the agent cannot self-set these. Best-effort:
+	// a lookup failure leaves the card exactly as the runtime sent it
+	// (no-worse-than-before). See agent_card_reconcile.go.
+	reconciledCard := payload.AgentCard
+	{
+		var dbName, dbRole sql.NullString
+		if qErr := db.DB.QueryRowContext(ctx,
+			`SELECT name, role FROM workspaces WHERE id = $1`, payload.ID,
+		).Scan(&dbName, &dbRole); qErr == nil {
+			if rc, did := reconcileAgentCardIdentity(
+				payload.AgentCard, payload.ID, dbName.String, dbRole.String,
+			); did {
+				reconciledCard = rc
+				log.Printf("Registry register: reconciled agent_card identity for %s from workspaces row", payload.ID)
+			}
+		}
+	}
+	agentCardStr := string(reconciledCard)
 
 	// urlForUpsert: poll-mode workspaces don't need a URL. Empty input
 	// becomes NULL via sql.NullString so the row's URL stays clean (the
@@ -413,10 +439,12 @@ func (h *RegistryHandler) Register(c *gin.Context) {
 		}
 	}
 
-	// Broadcast WORKSPACE_ONLINE
+	// Broadcast WORKSPACE_ONLINE — use the reconciled card so the canvas
+	// Agent Card view live-updates with the friendly name, matching what
+	// was just persisted (not the runtime's raw UUID-name card).
 	if err := h.broadcaster.RecordAndBroadcast(ctx, string(events.EventWorkspaceOnline), payload.ID, map[string]interface{}{
 		"url":           cachedURL,
-		"agent_card":    payload.AgentCard,
+		"agent_card":    reconciledCard,
 		"delivery_mode": effectiveMode,
 	}); err != nil {
 		log.Printf("Registry broadcast error: %v", err)

workspace-server/internal/handlers/restart_signals_test.go

@@ -271,6 +271,7 @@ func TestGracefulPreRestart_URLResolutionError(t *testing.T) {
 		WorkspaceHandler: newHandlerWithTestDeps(t),
 		errToReturn:      context.DeadlineExceeded,
 	}
+	waitForHandlerAsyncBeforeDBCleanup(t, hWrapper.WorkspaceHandler)
 
 	hWrapper.gracefulPreRestart(context.Background(), "ws-url-err-111")
 	time.Sleep(200 * time.Millisecond)

workspace-server/internal/handlers/template_files_eic.go

@@ -19,6 +19,7 @@ package handlers
 import (
 	"bytes"
 	"context"
+	"errors"
 	"fmt"
 	"log"
 	"os"
@@ -357,6 +358,28 @@ func writeFileViaEIC(ctx context.Context, instanceID, runtime, root, relPath str
 		var stderr bytes.Buffer
 		sshCmd.Stderr = &stderr
 		if err := sshCmd.Run(); err != nil {
+			// When the per-op context deadline (eicFileOpTimeout) fires,
+			// exec.CommandContext SIGKILLs the ssh subprocess and Run()
+			// returns the bare "signal: killed" with empty stderr. That
+			// surfaced to the canvas as an opaque
+			// `500 {"error":"ssh install: signal: killed ()"}` which gave
+			// the operator no idea the workspace was simply mid-provision
+			// with a slow/unready EIC tunnel (internal#423). Detect the
+			// deadline explicitly and return an actionable message instead
+			// — the EIC mechanism, timeout value, and success path are all
+			// unchanged; this only improves the error a stuck write emits.
+			if cerr := ctx.Err(); cerr != nil {
+				reason := "timed out after " + eicFileOpTimeout.String()
+				if errors.Is(cerr, context.Canceled) && !errors.Is(cerr, context.DeadlineExceeded) {
+					reason = "was cancelled"
+				}
+				return fmt.Errorf(
+					"ssh install: EIC tunnel to workspace %s — "+
+						"the workspace may still be provisioning (slow/unready SSH); "+
+						"retry once it is online, or apply provider credentials via "+
+						"Settings → Secrets (encrypted, does not use this file-write path)",
+					reason)
+			}
 			return fmt.Errorf("ssh install: %w (%s)", err, strings.TrimSpace(stderr.String()))
 		}
 		log.Printf("writeFileViaEIC: ws instance=%s runtime=%s root=%s wrote %d bytes → %s",

workspace-server/internal/handlers/template_files_eic_write_timeout_test.go

@@ -0,0 +1,71 @@
+package handlers
+
+// template_files_eic_write_timeout_test.go — pins the actionable-error
+// behavior added for internal#423.
+//
+// When the per-op context deadline (eicFileOpTimeout) fires,
+// exec.CommandContext SIGKILLs the ssh subprocess and Run() returns the
+// bare "signal: killed" with empty stderr. Before the fix that surfaced
+// to the canvas as an opaque `500 {"error":"ssh install: signal:
+// killed ()"}` — useless to an operator whose workspace was simply
+// mid-provision with a slow/unready EIC tunnel. The fix detects the
+// deadline explicitly (errors.Is(ctx.Err(), context.DeadlineExceeded))
+// and returns a message that names the cause and the
+// Settings → Secrets workaround.
+
+import (
+	"context"
+	"strings"
+	"testing"
+	"time"
+)
+
+// TestWriteFileViaEIC_DeadlineExceeded_ActionableError stubs
+// withEICTunnel so the *real* inner closure runs against a context that
+// has already exceeded its deadline. The ssh subprocess fails (no real
+// sshd on the fake port) and ctx.Err() == DeadlineExceeded, so the new
+// branch must fire and produce an actionable message — NOT the opaque
+// "signal: killed ()" string the canvas used to show.
+func TestWriteFileViaEIC_DeadlineExceeded_ActionableError(t *testing.T) {
+	prev := withEICTunnel
+	withEICTunnel = func(_ context.Context, instanceID string, fn func(s eicSSHSession) error) error {
+		// Run the real inner closure. It closes over the ctx that
+		// writeFileViaEIC derived from our already-cancelled parent, so
+		// the ssh subprocess is killed immediately and ctx.Err()
+		// resolves — exactly the eicFileOpTimeout-expiry shape.
+		return fn(eicSSHSession{
+			instanceID: instanceID,
+			osUser:     "ubuntu",
+			localPort:  1, // nothing listening → ssh fails fast
+			keyPath:    "/nonexistent/key",
+		})
+	}
+	t.Cleanup(func() { withEICTunnel = prev })
+
+	// Drive the real writeFileViaEIC. Pass a parent whose deadline has
+	// already passed: the context.WithTimeout(ctx, eicFileOpTimeout)
+	// derived inside writeFileViaEIC inherits the expired parent
+	// deadline, so ctx.Err() == context.DeadlineExceeded by the time
+	// the killed ssh subprocess returns — the exact production shape
+	// (eicFileOpTimeout expiry), exercised deterministically.
+	parent, cancel := context.WithDeadline(context.Background(), time.Now().Add(-time.Second))
+	defer cancel()
+
+	err := writeFileViaEIC(parent, "i-test", "claude-code", "/configs", "config.yaml", []byte("model: sonnet\n"))
+	if err == nil {
+		t.Fatalf("expected an error from a killed ssh subprocess, got nil")
+	}
+	msg := err.Error()
+
+	// Must NOT leak the opaque bare-signal string to the operator.
+	if strings.Contains(msg, "signal: killed ()") {
+		t.Fatalf("error still surfaces the opaque %q form: %q", "signal: killed ()", msg)
+	}
+	// Must name the cause and the Secrets workaround so the canvas
+	// shows something actionable.
+	for _, want := range []string{"timed out", "provisioning", "Settings", "Secrets"} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("actionable error missing %q; got: %q", want, msg)
+		}
+	}
+}

workspace-server/internal/handlers/templates.go

@@ -210,11 +210,16 @@ func (h *TemplatesHandler) List(c *gin.Context) {
 			model = raw.RuntimeConfig.Model
 		}
 
+		tier := raw.Tier
+		if h.wh != nil && h.wh.IsSaaS() {
+			tier = h.wh.DefaultTier()
+		}
+
 		templates = append(templates, templateSummary{
 			ID:                      id,
 			Name:                    raw.Name,
 			Description:             raw.Description,
-			Tier:                    raw.Tier,
+			Tier:                    tier,
 			Runtime:                 raw.Runtime,
 			Model:                   model,
 			Models:                  raw.RuntimeConfig.Models,
@@ -371,6 +376,11 @@ func (h *TemplatesHandler) ListFiles(c *gin.Context) {
 		if err != nil || path == walkRoot {
 			return nil
 		}
+		// Skip symlinks to prevent path traversal via malicious symlinks
+		// inside the workspace config directory (OFFSEC-010).
+		if info.Mode()&os.ModeSymlink != 0 {
+			return nil
+		}
 		rel, _ := filepath.Rel(walkRoot, path)
 		// Enforce depth limit
 		if strings.Count(rel, string(filepath.Separator))+1 > depth {

workspace-server/internal/handlers/templates_test.go

@@ -847,6 +847,58 @@ func TestListFiles_FallbackToHost_WithTemplate(t *testing.T) {
 	}
 }
 
+func TestListFiles_FallbackToHost_SkipsSymlinks(t *testing.T) {
+	mock := setupTestDB(t)
+	setupTestRedis(t)
+
+	tmpDir := t.TempDir()
+	tmplDir := filepath.Join(tmpDir, "test-agent")
+	if err := os.MkdirAll(tmplDir, 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(tmplDir, "config.yaml"), []byte("name: Test Agent\n"), 0644); err != nil {
+		t.Fatal(err)
+	}
+	secret := filepath.Join(t.TempDir(), "secret.txt")
+	if err := os.WriteFile(secret, []byte("do-not-list"), 0600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Symlink(secret, filepath.Join(tmplDir, "leaked-secret")); err != nil {
+		t.Fatal(err)
+	}
+
+	handler := NewTemplatesHandler(tmpDir, nil, nil)
+
+	mock.ExpectQuery(`SELECT name, COALESCE\(instance_id, ''\), COALESCE\(runtime, ''\) FROM workspaces WHERE id =`).
+		WithArgs("ws-tmpl").
+		WillReturnRows(sqlmock.NewRows([]string{"name", "instance_id", "runtime"}).AddRow("Test Agent", "", ""))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "ws-tmpl"}}
+	c.Request = httptest.NewRequest("GET", "/workspaces/ws-tmpl/files", nil)
+
+	handler.ListFiles(c)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var resp []map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatal(err)
+	}
+	for _, file := range resp {
+		if file["path"] == "leaked-secret" {
+			t.Fatalf("symlink should not be listed: %#v", resp)
+		}
+	}
+
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("unmet sqlmock expectations: %v", err)
+	}
+}
+
 // ==================== GET /workspaces/:id/files/*path ====================
 
 func TestReadFile_PathTraversal(t *testing.T) {
@@ -1200,4 +1252,3 @@ func TestCWE78_DeleteFile_TraversalVariants(t *testing.T) {
 		})
 	}
 }
-

workspace-server/internal/handlers/terminal_test.go

@@ -340,6 +340,11 @@ func TestSSHCommandCmd_BuildsArgv(t *testing.T) {
 // a workspace must still be able to access its own terminal. The CanCommunicate
 // fast-path returns true when callerID == targetID.
 func TestTerminalConnect_KI005_AllowsOwnTerminal(t *testing.T) {
+	mock := setupTestDB(t)
+	mock.ExpectQuery("SELECT COALESCE").
+		WithArgs("ws-alice").
+		WillReturnRows(sqlmock.NewRows([]string{"instance_id"}).AddRow(""))
+
 	// CanCommunicate fast-path: callerID == targetID → returns true without DB.
 	prev := canCommunicateCheck
 	canCommunicateCheck = func(callerID, targetID string) bool { return callerID == targetID }
@@ -367,6 +372,11 @@ func TestTerminalConnect_KI005_AllowsOwnTerminal(t *testing.T) {
 // skip the CanCommunicate check entirely and fall through to the Docker auth path.
 // We assert they get the nil-docker 503 instead of 403.
 func TestTerminalConnect_KI005_SkipsCheckWithoutHeader(t *testing.T) {
+	mock := setupTestDB(t)
+	mock.ExpectQuery("SELECT COALESCE").
+		WithArgs("ws-any").
+		WillReturnRows(sqlmock.NewRows([]string{"instance_id"}).AddRow(""))
+
 	h := NewTerminalHandler(nil) // nil docker → 503 if reached
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
@@ -439,6 +449,9 @@ func TestTerminalConnect_KI005_AllowsSiblingWorkspace(t *testing.T) {
 	mock.ExpectExec(`UPDATE workspace_auth_tokens SET last_used_at`).
 		WithArgs(sqlmock.AnyArg()).
 		WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectQuery("SELECT COALESCE").
+		WithArgs("ws-dev").
+		WillReturnRows(sqlmock.NewRows([]string{"instance_id"}).AddRow(""))
 
 	h := NewTerminalHandler(nil)
 	w := httptest.NewRecorder()
@@ -463,7 +476,10 @@ func TestTerminalConnect_KI005_AllowsSiblingWorkspace(t *testing.T) {
 // introduced in GH#1885: internal routing uses org tokens which are not in
 // workspace_auth_tokens, so ValidateToken would always fail for them.
 func TestKI005_OrgToken_SkipsValidateToken(t *testing.T) {
-	setupTestDB(t) // no ValidateToken ExpectQuery — none should fire
+	mock := setupTestDB(t) // no ValidateToken ExpectQuery — none should fire
+	mock.ExpectQuery("SELECT COALESCE").
+		WithArgs("ws-target").
+		WillReturnRows(sqlmock.NewRows([]string{"instance_id"}).AddRow(""))
 	prev := canCommunicateCheck
 	canCommunicateCheck = func(callerID, targetID string) bool {
 		// Simulate platform agent → target workspace (same org).
@@ -544,4 +560,3 @@ func TestSSHCommandCmd_ConnectTimeoutPresent(t *testing.T) {
 			args)
 	}
 }
-

workspace-server/internal/handlers/tokens.go

@@ -10,8 +10,20 @@ import (
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/wsauth"
 	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
 )
 
+// validWorkspaceID returns true when id is a syntactically valid UUID.
+// workspace_id is a `uuid` column; passing a non-UUID (e.g. the canvas
+// "global" sentinel sent when no node is selected) makes Postgres raise
+// `invalid input syntax for type uuid`, which previously leaked as an
+// opaque 500. Reject up front with a clean 400 instead. Mirrors the
+// uuid.Parse guard already used in handlers/activity.go.
+func validWorkspaceID(id string) bool {
+	_, err := uuid.Parse(id)
+	return err == nil
+}
+
 // TokenHandler exposes user-facing token management for workspaces.
 // Routes: GET/POST/DELETE /workspaces/:id/tokens (behind WorkspaceAuth).
 type TokenHandler struct{}
@@ -31,6 +43,10 @@ type tokenListItem struct {
 // never the plaintext or hash).
 func (h *TokenHandler) List(c *gin.Context) {
 	workspaceID := c.Param("id")
+	if !validWorkspaceID(workspaceID) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid workspace id"})
+		return
+	}
 
 	limit := 50
 	if v := c.Query("limit"); v != "" {
@@ -53,6 +69,7 @@ func (h *TokenHandler) List(c *gin.Context) {
 		LIMIT $2 OFFSET $3
 	`, workspaceID, limit, offset)
 	if err != nil {
+		log.Printf("tokens: list query failed for workspace %s: %v", workspaceID, err)
 		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to list tokens"})
 		return
 	}
@@ -85,6 +102,10 @@ const maxTokensPerWorkspace = 50
 // exactly once in the response — it cannot be recovered afterwards.
 func (h *TokenHandler) Create(c *gin.Context) {
 	workspaceID := c.Param("id")
+	if !validWorkspaceID(workspaceID) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid workspace id"})
+		return
+	}
 
 	// Rate limit: max active tokens per workspace
 	var count int
@@ -117,6 +138,10 @@ func (h *TokenHandler) Create(c *gin.Context) {
 func (h *TokenHandler) Revoke(c *gin.Context) {
 	workspaceID := c.Param("id")
 	tokenID := c.Param("tokenId")
+	if !validWorkspaceID(workspaceID) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid workspace id"})
+		return
+	}
 
 	result, err := db.DB.ExecContext(c.Request.Context(), `
 		UPDATE workspace_auth_tokens

workspace-server/internal/handlers/tokens_sqlmock_test.go

@@ -41,6 +41,15 @@ import (
 
 func init() { gin.SetMode(gin.TestMode) }
 
+// Workspace IDs are validated as UUIDs up front (tokens.go validWorkspaceID),
+// so handler tests must pass syntactically valid UUIDs. Fixed values keep
+// sqlmock WithArgs assertions deterministic.
+const (
+	wsUUID1 = "11111111-1111-1111-1111-111111111111"
+	wsUUID2 = "22222222-2222-2222-2222-222222222222"
+	wsUUID3 = "33333333-3333-3333-3333-333333333333"
+)
+
 // withMockDB swaps `db.DB` for a sqlmock and returns the mock plus a
 // restore func. Tests use this in place of setupTokenTestDB which
 // skips on a missing real DB.
@@ -81,13 +90,13 @@ func TestTokenHandler_List_HappyPath(t *testing.T) {
 	created := time.Date(2026, 4, 1, 12, 0, 0, 0, time.UTC)
 	last := created.Add(time.Hour)
 	mock.ExpectQuery(`SELECT id, prefix, created_at, last_used_at\s+FROM workspace_auth_tokens`).
-		WithArgs("ws-1", 50, 0).
+		WithArgs(wsUUID1, 50, 0).
 		WillReturnRows(sqlmock.NewRows([]string{"id", "prefix", "created_at", "last_used_at"}).
 			AddRow("tok-1", "abc12345", created, last).
 			AddRow("tok-2", "def67890", created, nil))
 
 	w := makeReq(t, NewTokenHandler().List, "GET",
-		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: "ws-1"}})
+		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: wsUUID1}})
 
 	if w.Code != http.StatusOK {
 		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
@@ -121,7 +130,7 @@ func TestTokenHandler_List_EmptyResult(t *testing.T) {
 		WillReturnRows(sqlmock.NewRows([]string{"id", "prefix", "created_at", "last_used_at"}))
 
 	w := makeReq(t, NewTokenHandler().List, "GET",
-		"/workspaces/ws-2/tokens", gin.Params{{Key: "id", Value: "ws-2"}})
+		"/workspaces/ws-2/tokens", gin.Params{{Key: "id", Value: wsUUID2}})
 
 	if w.Code != http.StatusOK {
 		t.Fatalf("expected 200 on empty list, got %d", w.Code)
@@ -146,7 +155,7 @@ func TestTokenHandler_List_QueryError(t *testing.T) {
 		WillReturnError(errors.New("connection refused"))
 
 	w := makeReq(t, NewTokenHandler().List, "GET",
-		"/workspaces/ws-3/tokens", gin.Params{{Key: "id", Value: "ws-3"}})
+		"/workspaces/ws-3/tokens", gin.Params{{Key: "id", Value: wsUUID3}})
 
 	if w.Code != http.StatusInternalServerError {
 		t.Errorf("query error must surface as 500, got %d", w.Code)
@@ -158,13 +167,13 @@ func TestTokenHandler_List_RespectsLimit(t *testing.T) {
 	defer cleanup()
 
 	mock.ExpectQuery(`SELECT id, prefix, created_at, last_used_at`).
-		WithArgs("ws-1", 10, 5).
+		WithArgs(wsUUID1, 10, 5).
 		WillReturnRows(sqlmock.NewRows([]string{"id", "prefix", "created_at", "last_used_at"}))
 
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
 	c.Request = httptest.NewRequest("GET", "/workspaces/ws-1/tokens?limit=10&offset=5", nil)
-	c.Params = gin.Params{{Key: "id", Value: "ws-1"}}
+	c.Params = gin.Params{{Key: "id", Value: wsUUID1}}
 	NewTokenHandler().List(c)
 
 	if w.Code != http.StatusOK {
@@ -186,7 +195,7 @@ func TestTokenHandler_List_ScanError(t *testing.T) {
 			AddRow("tok-1", "abc", "not-a-timestamp", nil))
 
 	w := makeReq(t, NewTokenHandler().List, "GET",
-		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: "ws-1"}})
+		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: wsUUID1}})
 
 	if w.Code != http.StatusInternalServerError {
 		t.Errorf("scan error must surface as 500, got %d: %s", w.Code, w.Body.String())
@@ -201,11 +210,11 @@ func TestTokenHandler_Create_RateLimited(t *testing.T) {
 
 	// Count query returns 50 (== max) → 429.
 	mock.ExpectQuery(`SELECT COUNT\(\*\) FROM workspace_auth_tokens`).
-		WithArgs("ws-1").
+		WithArgs(wsUUID1).
 		WillReturnRows(sqlmock.NewRows([]string{"count"}).AddRow(50))
 
 	w := makeReq(t, NewTokenHandler().Create, "POST",
-		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: "ws-1"}})
+		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: wsUUID1}})
 
 	if w.Code != http.StatusTooManyRequests {
 		t.Errorf("max active tokens should 429, got %d", w.Code)
@@ -225,7 +234,7 @@ func TestTokenHandler_Create_IssueFails(t *testing.T) {
 		WillReturnError(errors.New("disk full"))
 
 	w := makeReq(t, NewTokenHandler().Create, "POST",
-		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: "ws-1"}})
+		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: wsUUID1}})
 
 	if w.Code != http.StatusInternalServerError {
 		t.Errorf("IssueToken DB error must 500, got %d", w.Code)
@@ -242,7 +251,7 @@ func TestTokenHandler_Create_HappyPath(t *testing.T) {
 		WillReturnResult(sqlmock.NewResult(1, 1))
 
 	w := makeReq(t, NewTokenHandler().Create, "POST",
-		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: "ws-1"}})
+		"/workspaces/ws-1/tokens", gin.Params{{Key: "id", Value: wsUUID1}})
 
 	if w.Code != http.StatusCreated {
 		t.Fatalf("expected 201, got %d: %s", w.Code, w.Body.String())
@@ -257,7 +266,7 @@ func TestTokenHandler_Create_HappyPath(t *testing.T) {
 	if body.AuthToken == "" {
 		t.Errorf("auth_token must be present and non-empty in response")
 	}
-	if body.WorkspaceID != "ws-1" {
+	if body.WorkspaceID != wsUUID1 {
 		t.Errorf("workspace_id mismatch: %q", body.WorkspaceID)
 	}
 }
@@ -269,12 +278,12 @@ func TestTokenHandler_Revoke_HappyPath(t *testing.T) {
 	defer cleanup()
 
 	mock.ExpectExec(`UPDATE workspace_auth_tokens\s+SET revoked_at = now\(\)`).
-		WithArgs("tok-1", "ws-1").
+		WithArgs("tok-1", wsUUID1).
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
 	w := makeReq(t, NewTokenHandler().Revoke, "DELETE",
 		"/workspaces/ws-1/tokens/tok-1", gin.Params{
-			{Key: "id", Value: "ws-1"},
+			{Key: "id", Value: wsUUID1},
 			{Key: "tokenId", Value: "tok-1"},
 		})
 
@@ -289,12 +298,12 @@ func TestTokenHandler_Revoke_NotFound(t *testing.T) {
 
 	// 0 rows affected → token not found OR already revoked.
 	mock.ExpectExec(`UPDATE workspace_auth_tokens`).
-		WithArgs("tok-ghost", "ws-1").
+		WithArgs("tok-ghost", wsUUID1).
 		WillReturnResult(sqlmock.NewResult(0, 0))
 
 	w := makeReq(t, NewTokenHandler().Revoke, "DELETE",
 		"/workspaces/ws-1/tokens/tok-ghost", gin.Params{
-			{Key: "id", Value: "ws-1"},
+			{Key: "id", Value: wsUUID1},
 			{Key: "tokenId", Value: "tok-ghost"},
 		})
 
@@ -312,7 +321,7 @@ func TestTokenHandler_Revoke_DBError(t *testing.T) {
 
 	w := makeReq(t, NewTokenHandler().Revoke, "DELETE",
 		"/workspaces/ws-1/tokens/tok-1", gin.Params{
-			{Key: "id", Value: "ws-1"},
+			{Key: "id", Value: wsUUID1},
 			{Key: "tokenId", Value: "tok-1"},
 		})
 
@@ -321,6 +330,59 @@ func TestTokenHandler_Revoke_DBError(t *testing.T) {
 	}
 }
 
+// ---- UUID validation (regression: "global" sentinel 500) ------------
+
+// The canvas Settings → Workspace Tokens tab sent the literal sentinel
+// "global" as the workspace id when no node was selected. workspace_id
+// is a `uuid` column, so the query raised
+// `invalid input syntax for type uuid: "global"` which leaked as an
+// opaque 500. List/Create/Revoke now reject any non-UUID id with a
+// clean 400 before touching the DB. No DB expectation is set on the
+// mock — a DB hit would fail ExpectationsWereMet, proving short-circuit.
+func TestTokenHandler_RejectsNonUUIDWorkspaceID(t *testing.T) {
+	h := NewTokenHandler()
+	cases := []struct {
+		name   string
+		run    func(c *gin.Context)
+		method string
+		params gin.Params
+	}{
+		{"List", h.List, "GET", gin.Params{{Key: "id", Value: "global"}}},
+		{"Create", h.Create, "POST", gin.Params{{Key: "id", Value: "global"}}},
+		{"Revoke", h.Revoke, "DELETE", gin.Params{
+			{Key: "id", Value: "global"},
+			{Key: "tokenId", Value: "tok-1"},
+		}},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			mock, cleanup := withMockDB(t)
+			defer cleanup()
+
+			w := makeReq(t, tc.run, tc.method,
+				"/workspaces/global/tokens", tc.params)
+
+			if w.Code != http.StatusBadRequest {
+				t.Fatalf("%s with non-UUID id must 400, got %d: %s",
+					tc.name, w.Code, w.Body.String())
+			}
+			var body struct {
+				Error string `json:"error"`
+			}
+			_ = json.Unmarshal(w.Body.Bytes(), &body)
+			if body.Error != "invalid workspace id" {
+				t.Errorf("%s: want error=%q, got %q",
+					tc.name, "invalid workspace id", body.Error)
+			}
+			// No query/exec was expected → if the handler hit the DB
+			// this fails, proving the guard short-circuits before SQL.
+			if err := mock.ExpectationsWereMet(); err != nil {
+				t.Errorf("%s leaked a DB call past the uuid guard: %v", tc.name, err)
+			}
+		})
+	}
+}
+
 // Compile-time noise removal: the imports list pulls in the sql /
 // driver packages and the silenced ctx so a future scenario that
 // needs them doesn't have to re-add the import. Documented here so

workspace-server/internal/handlers/tokens_test.go

@@ -11,6 +11,7 @@ import (
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/wsauth"
 	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
 )
 
 func init() { gin.SetMode(gin.TestMode) }
@@ -167,11 +168,14 @@ func TestTokenHandler_RevokeWrongWorkspace(t *testing.T) {
 
 	h := NewTokenHandler()
 
-	// Try to revoke with a different workspace ID — should 404
+	// Try to revoke with a different (valid-UUID) workspace ID that does
+	// not own the token — should 404. A valid UUID is required so this
+	// exercises the ownership branch, not the up-front uuid-shape 400.
+	otherWS := uuid.NewString()
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "wrong-workspace-id"}, {Key: "tokenId", Value: tokenID}}
-	c.Request = httptest.NewRequest("DELETE", "/workspaces/wrong/tokens/"+tokenID, nil)
+	c.Params = gin.Params{{Key: "id", Value: otherWS}, {Key: "tokenId", Value: tokenID}}
+	c.Request = httptest.NewRequest("DELETE", "/workspaces/"+otherWS+"/tokens/"+tokenID, nil)
 	h.Revoke(c)
 
 	if w.Code != http.StatusNotFound {

workspace-server/internal/handlers/workspace.go

@@ -176,15 +176,14 @@ func (h *WorkspaceHandler) Create(c *gin.Context) {
 
 	id := uuid.New().String()
 	awarenessNamespace := workspaceAwarenessNamespace(id)
-	if payload.Tier == 0 {
-		// SaaS-aware default. SaaS → T4 (full host access; each
-		// workspace runs on its own sibling EC2 so the tier boundary
-		// is a Docker resource limit on the only container present —
-		// no neighbour to protect from). Self-hosted → T3 (read-write
-		// workspace mount + Docker daemon access, most templates'
-		// baseline). Lower tiers (T1 sandboxed, T2 standard) remain
-		// explicit opt-ins for low-trust agents. Matches the canvas
-		// CreateWorkspaceDialog defaults so the API and the UI agree.
+	if h.IsSaaS() {
+		// SaaS hard gate: every hosted workspace gets its own sibling
+		// EC2 instance, so T4 is the only meaningful runtime boundary.
+		// Do not trust stale clients/templates that still send T1/T2/T3.
+		payload.Tier = 4
+	} else if payload.Tier == 0 {
+		// Self-hosted default remains T3. Lower tiers (T1 sandboxed,
+		// T2 standard) stay explicit opt-ins for low-trust local agents.
 		payload.Tier = h.DefaultTier()
 	}
 

workspace-server/internal/handlers/workspace_abilities_test.go

@@ -2,296 +2,192 @@ package handlers
 
 import (
 	"bytes"
-	"context"
 	"database/sql"
 	"encoding/json"
-	"errors"
 	"net/http"
 	"net/http/httptest"
 	"testing"
 
 	"github.com/DATA-DOG/go-sqlmock"
-	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
 	"github.com/gin-gonic/gin"
 )
 
-func setupAbilitiesTest(t *testing.T) (sqlmock.Sqlmock, func()) {
-	t.Helper()
-	mockDB, mock, err := sqlmock.New()
-	if err != nil {
-		t.Fatalf("failed to create sqlmock: %v", err)
-	}
-	prev := db.DB
-	db.DB = mockDB
-	return mock, func() {
-		db.DB = prev
-		mockDB.Close()
-	}
-}
-
-func TestPatchAbilities_InvalidWorkspaceID_Returns400(t *testing.T) {
-	_, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
-
+// patchReq builds a gin context for a PATCH request to /workspaces/:id/abilities.
+func patchReq(id, body string) (*http.Request, *httptest.ResponseRecorder, *gin.Context) {
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "not-a-valid-uuid"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/not-a-valid-uuid/abilities",
-		bytes.NewBufferString(`{"broadcast_enabled":true}`))
+	c.Params = gin.Params{{Key: "id", Value: id}}
+	c.Request = httptest.NewRequest("PATCH", "/workspaces/"+id+"/abilities", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
+	return c.Request, w, c
+}
 
+func TestPatchAbilities_InvalidWorkspaceID(t *testing.T) {
+	setupTestDB(t)
+
+	// "not-a-uuid" fails validateWorkspaceID
+	_, w, c := patchReq("not-a-uuid", `{"broadcast_enabled":true}`)
 	PatchAbilities(c)
 
 	if w.Code != http.StatusBadRequest {
 		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["error"] != "invalid workspace ID" {
-		t.Errorf("expected 'invalid workspace ID', got %q", body["error"])
-	}
 }
 
-func TestPatchAbilities_EmptyBody_Returns400(t *testing.T) {
-	_, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
+func TestPatchAbilities_EmptyBody(t *testing.T) {
+	setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000001"
 
+	// Empty JSON object — no ability fields present
+	_, w, c := patchReq(id, `{}`)
 	PatchAbilities(c)
 
 	if w.Code != http.StatusBadRequest {
 		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["error"] != "at least one ability field required" {
-		t.Errorf("expected 'at least one ability field required', got %q", body["error"])
+
+	var resp map[string]string
+	json.Unmarshal(w.Body.Bytes(), &resp)
+	if resp["error"] != "at least one ability field required" {
+		t.Errorf("expected 'at least one ability field required', got %v", resp["error"])
 	}
 }
 
-func TestPatchAbilities_InvalidJSON_Returns400(t *testing.T) {
-	_, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
+func TestPatchAbilities_WorkspaceNotFound(t *testing.T) {
+	mock := setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000002"
 
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{invalid json}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
-
-	PatchAbilities(c)
-
-	if w.Code != http.StatusBadRequest {
-		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
-	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["error"] != "invalid request body" {
-		t.Errorf("expected 'invalid request body', got %q", body["error"])
-	}
-}
-
-func TestPatchAbilities_WorkspaceNotFound_Returns404(t *testing.T) {
-	mock, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
-
-	mock.ExpectQuery("SELECT EXISTS").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnError(sql.ErrNoRows)
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{"broadcast_enabled":true}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
+	// SELECT EXISTS returns false (workspace does not exist)
+	mock.ExpectQuery(`SELECT EXISTS\(SELECT 1 FROM workspaces WHERE id = \$1 AND status != 'removed'\)`).
+		WithArgs(id).
+		WillReturnRows(sqlmock.NewRows([]string{"exists"}).AddRow(false))
 
+	_, w, c := patchReq(id, `{"broadcast_enabled":true}`)
 	PatchAbilities(c)
 
 	if w.Code != http.StatusNotFound {
 		t.Errorf("expected 404, got %d: %s", w.Code, w.Body.String())
 	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
-	}
 }
 
-func TestPatchAbilities_WorkspaceDBError_Returns404(t *testing.T) {
-	mock, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
+func TestPatchAbilities_SetBroadcastEnabledTrue(t *testing.T) {
+	mock := setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000003"
 
-	mock.ExpectQuery("SELECT EXISTS").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnError(errors.New("connection refused"))
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{"broadcast_enabled":true}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
-
-	PatchAbilities(c)
-
-	if w.Code != http.StatusNotFound {
-		t.Errorf("expected 404, got %d: %s", w.Code, w.Body.String())
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
-	}
-}
-
-func TestPatchAbilities_UpdateBroadcastEnabled_Returns200(t *testing.T) {
-	mock, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
-
-	mock.ExpectQuery("SELECT EXISTS").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
+	// SELECT EXISTS → true
+	mock.ExpectQuery(`SELECT EXISTS\(SELECT 1 FROM workspaces WHERE id = \$1 AND status != 'removed'\)`).
+		WithArgs(id).
 		WillReturnRows(sqlmock.NewRows([]string{"exists"}).AddRow(true))
-	mock.ExpectExec("UPDATE workspaces SET broadcast_enabled").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000", true).
+
+	// UPDATE broadcast_enabled = true
+	mock.ExpectExec(`UPDATE workspaces SET broadcast_enabled = \$2, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs(id, true).
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{"broadcast_enabled":true}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
-
+	_, w, c := patchReq(id, `{"broadcast_enabled":true}`)
 	PatchAbilities(c)
 
 	if w.Code != http.StatusOK {
 		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["status"] != "updated" {
-		t.Errorf("expected status=updated, got %v", body)
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+
+	var resp map[string]string
+	json.Unmarshal(w.Body.Bytes(), &resp)
+	if resp["status"] != "updated" {
+		t.Errorf("expected status=updated, got %v", resp["status"])
 	}
 }
 
-func TestPatchAbilities_UpdateTalkToUserEnabled_Returns200(t *testing.T) {
-	mock, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
+func TestPatchAbilities_SetTalkToUserEnabledFalse(t *testing.T) {
+	mock := setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000004"
 
-	mock.ExpectQuery("SELECT EXISTS").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
+	// SELECT EXISTS → true
+	mock.ExpectQuery(`SELECT EXISTS\(SELECT 1 FROM workspaces WHERE id = \$1 AND status != 'removed'\)`).
+		WithArgs(id).
 		WillReturnRows(sqlmock.NewRows([]string{"exists"}).AddRow(true))
-	mock.ExpectExec("UPDATE workspaces SET talk_to_user_enabled").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000", true).
+
+	// UPDATE talk_to_user_enabled = false
+	mock.ExpectExec(`UPDATE workspaces SET talk_to_user_enabled = \$2, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs(id, false).
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{"talk_to_user_enabled":true}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
-
+	_, w, c := patchReq(id, `{"talk_to_user_enabled":false}`)
 	PatchAbilities(c)
 
 	if w.Code != http.StatusOK {
 		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["status"] != "updated" {
-		t.Errorf("expected status=updated, got %v", body)
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
-	}
 }
 
-func TestPatchAbilities_UpdateBothAbilities_Returns200(t *testing.T) {
-	mock, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
+func TestPatchAbilities_BothFields(t *testing.T) {
+	mock := setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000005"
 
-	mock.ExpectQuery("SELECT EXISTS").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
+	// SELECT EXISTS → true
+	mock.ExpectQuery(`SELECT EXISTS\(SELECT 1 FROM workspaces WHERE id = \$1 AND status != 'removed'\)`).
+		WithArgs(id).
 		WillReturnRows(sqlmock.NewRows([]string{"exists"}).AddRow(true))
-	mock.ExpectExec("UPDATE workspaces SET broadcast_enabled").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000", true).
-		WillReturnResult(sqlmock.NewResult(0, 1))
-	mock.ExpectExec("UPDATE workspaces SET talk_to_user_enabled").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000", false).
+
+	// UPDATE broadcast_enabled = false
+	mock.ExpectExec(`UPDATE workspaces SET broadcast_enabled = \$2, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs(id, false).
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{"broadcast_enabled":true,"talk_to_user_enabled":false}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
+	// UPDATE talk_to_user_enabled = true
+	mock.ExpectExec(`UPDATE workspaces SET talk_to_user_enabled = \$2, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs(id, true).
+		WillReturnResult(sqlmock.NewResult(0, 1))
 
+	_, w, c := patchReq(id, `{"broadcast_enabled":false,"talk_to_user_enabled":true}`)
 	PatchAbilities(c)
 
 	if w.Code != http.StatusOK {
 		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["status"] != "updated" {
-		t.Errorf("expected status=updated, got %v", body)
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
-	}
 }
 
-func TestPatchAbilities_UpdateBroadcastDisabled_Returns200(t *testing.T) {
-	mock, cleanup := setupAbilitiesTest(t)
-	defer cleanup()
+func TestPatchAbilities_BroadcastUpdateFails(t *testing.T) {
+	mock := setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000006"
 
-	mock.ExpectQuery("SELECT EXISTS").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
+	// SELECT EXISTS → true
+	mock.ExpectQuery(`SELECT EXISTS\(SELECT 1 FROM workspaces WHERE id = \$1 AND status != 'removed'\)`).
+		WithArgs(id).
 		WillReturnRows(sqlmock.NewRows([]string{"exists"}).AddRow(true))
-	mock.ExpectExec("UPDATE workspaces SET broadcast_enabled").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000", false).
-		WillReturnResult(sqlmock.NewResult(0, 1))
 
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("PATCH",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/abilities",
-		bytes.NewBufferString(`{"broadcast_enabled":false}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
+	// UPDATE fails
+	mock.ExpectExec(`UPDATE workspaces SET broadcast_enabled = \$2, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs(id, true).
+		WillReturnError(sql.ErrConnDone)
 
+	_, w, c := patchReq(id, `{"broadcast_enabled":true}`)
 	PatchAbilities(c)
 
-	if w.Code != http.StatusOK {
-		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+	if w.Code != http.StatusInternalServerError {
+		t.Errorf("expected 500, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestPatchAbilities_TalkToUserUpdateFails(t *testing.T) {
+	mock := setupTestDB(t)
+	id := "00000000-0000-0000-0000-000000000007"
+
+	// SELECT EXISTS → true
+	mock.ExpectQuery(`SELECT EXISTS\(SELECT 1 FROM workspaces WHERE id = \$1 AND status != 'removed'\)`).
+		WithArgs(id).
+		WillReturnRows(sqlmock.NewRows([]string{"exists"}).AddRow(true))
+
+	// UPDATE broadcast_enabled skipped (not in payload)
+	// UPDATE talk_to_user_enabled fails
+	mock.ExpectExec(`UPDATE workspaces SET talk_to_user_enabled = \$2, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs(id, false).
+		WillReturnError(sql.ErrConnDone)
+
+	_, w, c := patchReq(id, `{"talk_to_user_enabled":false}`)
+	PatchAbilities(c)
+
+	if w.Code != http.StatusInternalServerError {
+		t.Errorf("expected 500, got %d: %s", w.Code, w.Body.String())
 	}
 }

workspace-server/internal/handlers/workspace_broadcast.go

@@ -3,7 +3,7 @@ package handlers
 // workspace_broadcast.go — POST /workspaces/:id/broadcast
 //
 // Allows a workspace with broadcast_enabled=true to send a message to every
-// non-removed agent workspace in the org.  The message is:
+// non-removed agent workspace in the SAME ORG.  The message is:
 //
 //   • Persisted in each recipient's activity_logs (type='broadcast_receive')
 //     so poll-mode agents pick it up via GET /activity.
@@ -16,6 +16,11 @@ package handlers
 // Auth: WorkspaceAuth (the agent triggers this with its own bearer token).
 // The handler re-validates broadcast_enabled inside the DB lookup to prevent
 // TOCTOU — the middleware only proved the token is valid, not the ability.
+//
+// Org isolation (OFFSEC-015): recipients are scoped to the sender's org using
+// a recursive CTE that walks the parent_id chain to find the org root. This
+// prevents a compromised or misconfigured workspace from broadcasting to
+// workspaces in other tenants' orgs.
 
 import (
 	"log"
@@ -29,11 +34,13 @@ import (
 
 // BroadcastHandler is constructed once and shared across requests.
 type BroadcastHandler struct {
-	broadcaster *events.Broadcaster
+	broadcaster events.EventEmitter
 }
 
 // NewBroadcastHandler creates a BroadcastHandler.
-func NewBroadcastHandler(b *events.Broadcaster) *BroadcastHandler {
+// The emitter is any EventEmitter — the concrete *Broadcaster in production,
+// or a test double in unit tests.
+func NewBroadcastHandler(b events.EventEmitter) *BroadcastHandler {
 	return &BroadcastHandler{broadcaster: b}
 }
 
@@ -74,11 +81,49 @@ func (h *BroadcastHandler) Broadcast(c *gin.Context) {
 		return
 	}
 
-	// Collect all non-removed agent workspaces (excludes the sender itself).
-	rows, err := db.DB.QueryContext(ctx,
-		`SELECT id FROM workspaces WHERE status != 'removed' AND id != $1`,
-		senderID,
-	)
+	// Find the sender's org root by walking the parent_id chain.
+	// Workspaces with parent_id = NULL are org roots; every other workspace
+	// belongs to the org identified by its topmost ancestor.
+	var orgRootID string
+	err = db.DB.QueryRowContext(ctx, `
+		WITH RECURSIVE org_chain AS (
+			SELECT id, parent_id, id AS root_id
+			FROM workspaces
+			WHERE id = $1
+			UNION ALL
+			SELECT w.id, w.parent_id, c.root_id
+			FROM workspaces w
+			JOIN org_chain c ON w.id = c.parent_id
+		)
+		SELECT root_id FROM org_chain WHERE parent_id IS NULL LIMIT 1
+	`, senderID).Scan(&orgRootID)
+	if err != nil {
+		log.Printf("Broadcast: org root lookup for %s: %v", senderID, err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "internal error"})
+		return
+	}
+
+	// Collect all non-removed agent workspaces in the SAME ORG (same root_id),
+	// excluding the sender itself.
+	rows, err := db.DB.QueryContext(ctx, `
+		WITH RECURSIVE org_chain AS (
+			SELECT id, parent_id, id AS root_id
+			FROM workspaces
+			WHERE parent_id IS NULL
+			UNION ALL
+			SELECT w.id, w.parent_id, c.root_id
+			FROM workspaces w
+			JOIN org_chain c ON w.parent_id = c.id
+		)
+		SELECT c.id
+		FROM org_chain c
+		WHERE c.root_id = $1
+		  AND c.id != $2
+		  AND EXISTS (
+			  SELECT 1 FROM workspaces w
+			  WHERE w.id = c.id AND w.status != 'removed'
+		  )
+	`, orgRootID, senderID)
 	if err != nil {
 		log.Printf("Broadcast: recipient query failed for %s: %v", senderID, err)
 		c.JSON(http.StatusInternalServerError, gin.H{"error": "internal error"})

workspace-server/internal/handlers/workspace_broadcast_test.go

@@ -3,7 +3,6 @@ package handlers
 import (
 	"bytes"
 	"context"
-	"database/sql"
 	"encoding/json"
 	"errors"
 	"net/http"
@@ -11,130 +10,187 @@ import (
 	"testing"
 
 	"github.com/DATA-DOG/go-sqlmock"
-	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
 	"github.com/gin-gonic/gin"
 )
 
-// -------------------------------------------------------------------------- //
-// broadcastTruncate
-// -------------------------------------------------------------------------- //
+// -------- Org-scoped recipient query tests (OFFSEC-015) --------
 
-func TestBroadcastTruncate_ShortString_ReturnsUnmodified(t *testing.T) {
-	result := broadcastTruncate("hello", 10)
-	if result != "hello" {
-		t.Errorf("expected 'hello', got %q", result)
-	}
-}
+// TestBroadcast_OrgScopedRecipients verifies that a broadcast from Org-A does
+// NOT reach workspaces belonging to Org-B. This is the core regression test
+// for OFFSEC-015: the original query had no org filter, so a workspace in
+// Org-A could broadcast to every non-removed workspace in the entire DB,
+// including workspaces owned by other tenants.
+func TestBroadcast_OrgScopedRecipients(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-func TestBroadcastTruncate_ExactlyMaxLength_ReturnsUnmodified(t *testing.T) {
-	result := broadcastTruncate("hello", 5)
-	if result != "hello" {
-		t.Errorf("expected 'hello', got %q", result)
-	}
-}
+	// Org-A structure:
+	//   org-a-root  (parent_id = NULL)  ← sender
+	//   ├── ws-a-child
+	// Org-B structure:
+	//   org-b-root  (parent_id = NULL)
+	//   └── ws-b-child
+	senderID := "00000000-0000-0000-0000-000000000001" // org-a-root
+	wsAChild := "00000000-0000-0000-0000-000000000002"
+	// ws-b-child is in Org-B (different root); the org-scoped query MUST NOT include it.
 
-func TestBroadcastTruncate_ExceedsMaxLength_TruncatesWithEllipsis(t *testing.T) {
-	result := broadcastTruncate("hello world", 5)
-	if result != "hello…" {
-		t.Errorf("expected 'hello…', got %q", result)
-	}
-}
+	// 1. Sender lookup
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Org-A Root", true))
 
-func TestBroadcastTruncate_Unicode_TruncatesAtRuneBoundary(t *testing.T) {
-	result := broadcastTruncate("日本語テスト", 2)
-	if result != "日本…" {
-		t.Errorf("expected '日本…', got %q", result)
-	}
-}
+	// 2. Org root lookup — sender is its own root (parent_id = NULL)
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(senderID))
 
-// -------------------------------------------------------------------------- //
-// BroadcastHandler
-// -------------------------------------------------------------------------- //
+	// 3. Org-scoped recipient query — MUST include org filter so ws-b-child is NOT included.
+	// The query joins on org_chain.root_id = orgRootID, which scopes to Org-A only.
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID, senderID). // orgRootID, senderID (EXCLUDED)
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(wsAChild)) // only Org-A child
 
-func setupBroadcastTest(t *testing.T) (sqlmock.Sqlmock, func()) {
-	t.Helper()
-	mockDB, mock, err := sqlmock.New()
-	if err != nil {
-		t.Fatalf("failed to create sqlmock: %v", err)
-	}
-	prev := db.DB
-	db.DB = mockDB
-	return mock, func() {
-		db.DB = prev
-		mockDB.Close()
-	}
-}
+	// Activity log inserts
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(wsAChild, senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
 
-func TestBroadcast_InvalidWorkspaceID_Returns400(t *testing.T) {
-	_, cleanup := setupBroadcastTest(t)
-	defer cleanup()
-
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "not-a-uuid"}}
-	c.Request = httptest.NewRequest("POST", "/workspaces/not-a-uuid/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"hello from org-a"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
 
-	if w.Code != http.StatusBadRequest {
-		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
+	if w.Code != http.StatusOK {
+		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["error"] != "invalid workspace ID" {
-		t.Errorf("expected 'invalid workspace ID', got %q", body["error"])
+	var resp map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal response: %v", err)
+	}
+	if resp["status"] != "sent" {
+		t.Errorf("expected status 'sent', got %v", resp["status"])
+	}
+	// ws-b-child is in a DIFFERENT org — the org-scoped query MUST NOT include it.
+	// If it were included, the mock would have an unmet expectation.
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("unmet mock expectations — cross-org workspace was included in broadcast: %v", err)
 	}
 }
 
-func TestBroadcast_MissingMessage_Returns400(t *testing.T) {
-	_, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+// TestBroadcast_OrgScoped_OrgRootSender verifies that when the sender IS the
+// org root (parent_id = NULL), broadcasts still reach sibling workspaces.
+func TestBroadcast_OrgScoped_OrgRootSender(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
+
+	senderID := "00000000-0000-0000-0000-000000000001" // org-a-root
+	siblingID := "00000000-0000-0000-0000-000000000002"
+
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Root Agent", true))
+
+	// Sender is the org root — CTE returns sender's own ID as root
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(senderID))
+
+	// Recipients in same org, excluding sender
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID, senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(siblingID))
+
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(siblingID, senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
 
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/broadcast",
-		bytes.NewBufferString(`{}`))
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"hello siblings"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
 
-	if w.Code != http.StatusBadRequest {
-		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
+	if w.Code != http.StatusOK {
+		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["error"] != "message is required" {
-		t.Errorf("expected 'message is required', got %q", body["error"])
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("unmet expectations: %v", err)
 	}
 }
 
-func TestBroadcast_WorkspaceNotFound_Returns404(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+// TestBroadcast_OrgScoped_ChildWorkspaceSender verifies that a non-root child
+// workspace can broadcast to siblings in the same org.
+func TestBroadcast_OrgScoped_ChildWorkspaceSender(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnError(sql.ErrNoRows)
+	orgRootID := "00000000-0000-0000-0000-000000000001"
+	senderID := "00000000-0000-0000-0000-000000000002" // child workspace
+	siblingID := "00000000-0000-0000-0000-000000000003"
+
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Child Agent", true))
+
+	// Org root lookup — walk up to find org-a-root
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(orgRootID))
+
+	// Recipients: same org, excluding sender
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(orgRootID, senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(siblingID))
+
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(siblingID, senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
 
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"child broadcasting"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("unmet expectations: %v", err)
+	}
+}
+
+// -------- Non-regression cases --------
+
+func TestBroadcast_NotFound(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
+
+	senderID := "00000000-0000-0000-0000-000000000099"
+	// UUID is valid, but no workspace row matches
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnError(errors.New("workspace not found"))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"test"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Broadcast(c)
 
 	if w.Code != http.StatusNotFound {
 		t.Errorf("expected 404, got %d: %s", w.Code, w.Body.String())
@@ -144,255 +200,343 @@ func TestBroadcast_WorkspaceNotFound_Returns404(t *testing.T) {
 	}
 }
 
-func TestBroadcast_BroadcastDisabled_Returns403(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+func TestBroadcast_Disabled(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).
-			AddRow("test-agent", false))
+	senderID := "00000000-0000-0000-0000-000000000003"
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Disabled Agent", false))
 
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"should not send"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
 
 	if w.Code != http.StatusForbidden {
 		t.Errorf("expected 403, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]string
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["error"] != "broadcast_disabled" {
-		t.Errorf("expected error='broadcast_disabled', got %v", body)
+	var resp map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
 	}
-	if _, ok := body["hint"]; !ok {
-		t.Errorf("expected hint field in 403 body, got %v", body)
+	if resp["error"] != "broadcast_disabled" {
+		t.Errorf("expected error 'broadcast_disabled', got %v", resp["error"])
+	}
+}
+
+func TestBroadcast_EmptyOrg_NoRecipients(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
+
+	senderID := "00000000-0000-0000-0000-000000000004" // org root, only workspace in org
+
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Lone Root", true))
+
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(senderID))
+
+	// No other workspaces in this org
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID, senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}))
+
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"hello org"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Broadcast(c)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
+	}
+	var resp map[string]interface{}
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("failed to unmarshal: %v", err)
+	}
+	if resp["delivered"] != float64(0) {
+		t.Errorf("expected delivered=0, got %v", resp["delivered"])
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
 		t.Errorf("unmet expectations: %v", err)
 	}
 }
 
-func TestBroadcast_RecipientQueryFails_Returns500(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+func TestBroadcast_InvalidWorkspaceID(t *testing.T) {
+	setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).
-			AddRow("test-agent", true))
-	mock.ExpectQuery("SELECT id FROM workspaces WHERE status != 'removed' AND id != ").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnError(errors.New("connection refused"))
-
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	c.Params = gin.Params{{Key: "id", Value: "not-a-uuid"}}
+	body := `{"message":"test"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/not-a-uuid/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestBroadcast_OrgRootLookupFails(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
+
+	senderID := "00000000-0000-0000-0000-000000000005"
+
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Root Agent", true))
+
+	// Org root CTE fails
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID).
+		WillReturnError(context.DeadlineExceeded)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: senderID}}
+	body := `{"message":"should not broadcast"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Broadcast(c)
 
 	if w.Code != http.StatusInternalServerError {
 		t.Errorf("expected 500, got %d: %s", w.Code, w.Body.String())
 	}
+	// The recipient query MUST NOT be called — it would broadcast cross-org
+	// if the org root lookup failed silently.
 	if err := mock.ExpectationsWereMet(); err != nil {
 		t.Errorf("unmet expectations: %v", err)
 	}
 }
 
-func TestBroadcast_NoRecipients_Returns200(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+func TestBroadcast_OrgScoped_SelfBroadcastExcluded(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).
-			AddRow("test-agent", true))
-	mock.ExpectQuery("SELECT id FROM workspaces WHERE status != 'removed' AND id != ").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000").
-		WillReturnRows(sqlmock.NewRows([]string{"id"}))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs("550e8400-e29b-41d4-a716-446655440000", "Broadcast sent to 0 workspace(s)").
-		WillReturnResult(sqlmock.NewResult(0, 1))
+	senderID := "00000000-0000-0000-0000-000000000006"
+	peerID := "00000000-0000-0000-0000-000000000007"
 
-	h := NewBroadcastHandler(newTestBroadcaster())
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "550e8400-e29b-41d4-a716-446655440000"}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/550e8400-e29b-41d4-a716-446655440000/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
-	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
-
-	h.Broadcast(c)
-
-	if w.Code != http.StatusOK {
-		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
-	}
-	var body map[string]interface{}
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if body["status"] != "sent" {
-		t.Errorf("expected status=sent, got %v", body)
-	}
-	if int(body["delivered"].(float64)) != 0 {
-		t.Errorf("expected delivered=0, got %v", body["delivered"])
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
-	}
-}
-
-func TestBroadcast_DeliversToOneRecipient_Returns200(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
-
-	senderID := "550e8400-e29b-41d4-a716-446655440000"
-	recipientID := "660e8400-e29b-41d4-a716-446655440001"
-	senderName := "test-agent"
-
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
 		WithArgs(senderID).
-		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).
-			AddRow(senderName, true))
-	mock.ExpectQuery("SELECT id FROM workspaces WHERE status != 'removed' AND id != ").
-		WithArgs(senderID).
-		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(recipientID))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs(recipientID, senderID, "Broadcast from "+senderName+": hello").
-		WillReturnResult(sqlmock.NewResult(0, 1))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs(senderID, "Broadcast sent to 1 workspace(s)").
-		WillReturnResult(sqlmock.NewResult(0, 1))
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Root Agent", true))
+
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(senderID))
+
+	// Recipient query MUST exclude sender via id != senderID
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID, senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(peerID))
+
+	// Peer receives broadcast_receive
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(peerID, senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
+	// Sender logs broadcast_sent (NOT broadcast_receive)
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).WillReturnResult(sqlmock.NewResult(0, 1))
 
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
 	c.Params = gin.Params{{Key: "id", Value: senderID}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/"+senderID+"/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	body := `{"message":"no echo to self"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
 
 	if w.Code != http.StatusOK {
 		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]interface{}
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if int(body["delivered"].(float64)) != 1 {
-		t.Errorf("expected delivered=1, got %v", body["delivered"])
-	}
 	if err := mock.ExpectationsWereMet(); err != nil {
 		t.Errorf("unmet expectations: %v", err)
 	}
 }
 
-func TestBroadcast_RecipientInsertFails_Continues_Returns200(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+// TestBroadcast_RecipientActivityLogFails_SkipsAndContinues: if one recipient's
+// activity_log insert fails, the handler logs the error and continues to the
+// next recipient rather than aborting the whole broadcast.
+func TestBroadcast_RecipientActivityLogFails_SkipsAndContinues(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-	senderID := "550e8400-e29b-41d4-a716-446655440000"
-	recipientID := "660e8400-e29b-41d4-a716-446655440001"
-	senderName := "test-agent"
+	senderID := "00000000-0000-0000-0000-000000000008"
+	peerA := "00000000-0000-0000-0000-000000000009"
+	peerB := "00000000-0000-0000-0000-00000000000a"
 
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
 		WithArgs(senderID).
-		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).
-			AddRow(senderName, true))
-	mock.ExpectQuery("SELECT id FROM workspaces WHERE status != 'removed' AND id != ").
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Resilient Agent", true))
+
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
 		WithArgs(senderID).
-		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(recipientID))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs(recipientID, senderID, "Broadcast from "+senderName+": hello").
-		WillReturnError(errors.New("connection refused"))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs(senderID, "Broadcast sent to 0 workspace(s)").
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(senderID))
+
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID, senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(peerA).AddRow(peerB))
+
+	// Peer A fails — handler logs and continues
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(peerA, senderID, sqlmock.AnyArg()).
+		WillReturnError(context.DeadlineExceeded)
+	// Peer B succeeds
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(peerB, senderID, sqlmock.AnyArg()).
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	// Sender log succeeds
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).
 		WillReturnResult(sqlmock.NewResult(0, 1))
 
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
 	c.Params = gin.Params{{Key: "id", Value: senderID}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/"+senderID+"/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	body := `{"message":"partial delivery"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
 
 	if w.Code != http.StatusOK {
-		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
+		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
 	}
-	var body map[string]interface{}
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if int(body["delivered"].(float64)) != 0 {
-		t.Errorf("expected delivered=0 (failed inserts don't count), got %v", body["delivered"])
+	var resp map[string]interface{}
+	json.Unmarshal(w.Body.Bytes(), &resp)
+	// Only peerB was delivered
+	if int(resp["delivered"].(float64)) != 1 {
+		t.Errorf("expected delivered=1, got %v", resp["delivered"])
 	}
 	if err := mock.ExpectationsWereMet(); err != nil {
 		t.Errorf("unmet expectations: %v", err)
 	}
 }
 
-func TestBroadcast_SenderLogFails_StillReturns200(t *testing.T) {
-	mock, cleanup := setupBroadcastTest(t)
-	defer cleanup()
+// TestBroadcast_SenderActivityLogFails_StillReturns200: if the sender's own
+// broadcast_sent activity_log insert fails, the handler still returns 200
+// so the caller doesn't retry a broadcast that already partially delivered.
+func TestBroadcast_SenderActivityLogFails_StillReturns200(t *testing.T) {
+	mock := setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
 
-	senderID := "550e8400-e29b-41d4-a716-446655440000"
-	recipientID := "660e8400-e29b-41d4-a716-446655440001"
-	senderName := "test-agent"
+	senderID := "00000000-0000-0000-0000-00000000000b"
+	peerA := "00000000-0000-0000-0000-00000000000c"
 
-	mock.ExpectQuery("SELECT name, broadcast_enabled FROM workspaces").
+	mock.ExpectQuery(`SELECT name, broadcast_enabled FROM workspaces WHERE id = \$1 AND status != 'removed'`).
 		WithArgs(senderID).
-		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).
-			AddRow(senderName, true))
-	mock.ExpectQuery("SELECT id FROM workspaces WHERE status != 'removed' AND id != ").
+		WillReturnRows(sqlmock.NewRows([]string{"name", "broadcast_enabled"}).AddRow("Log-Fail Agent", true))
+
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
 		WithArgs(senderID).
-		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(recipientID))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs(recipientID, senderID, "Broadcast from "+senderName+": hello").
+		WillReturnRows(sqlmock.NewRows([]string{"root_id"}).AddRow(senderID))
+
+	mock.ExpectQuery(`WITH RECURSIVE org_chain AS`).
+		WithArgs(senderID, senderID).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow(peerA))
+
+	// Peer log succeeds
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(peerA, senderID, sqlmock.AnyArg()).
 		WillReturnResult(sqlmock.NewResult(0, 1))
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WithArgs(senderID, "Broadcast sent to 1 workspace(s)").
-		WillReturnError(errors.New("connection refused"))
+	// Sender log FAILS
+	mock.ExpectExec(`INSERT INTO activity_logs`).WithArgs(senderID, sqlmock.AnyArg()).
+		WillReturnError(context.DeadlineExceeded)
 
-	h := NewBroadcastHandler(newTestBroadcaster())
 	w := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(w)
 	c.Params = gin.Params{{Key: "id", Value: senderID}}
-	c.Request = httptest.NewRequest("POST",
-		"/workspaces/"+senderID+"/broadcast",
-		bytes.NewBufferString(`{"message":"hello"}`))
+	body := `{"message":"log fail test"}`
+	c.Request = httptest.NewRequest("POST", "/workspaces/"+senderID+"/broadcast", bytes.NewBufferString(body))
 	c.Request.Header.Set("Content-Type", "application/json")
-	c.Request = c.Request.WithContext(context.Background())
 
-	h.Broadcast(c)
+	handler.Broadcast(c)
 
 	if w.Code != http.StatusOK {
-		t.Errorf("expected 200, got %d: %s", w.Code, w.Body.String())
-	}
-	var body map[string]interface{}
-	json.Unmarshal(w.Body.Bytes(), &body)
-	if int(body["delivered"].(float64)) != 1 {
-		t.Errorf("expected delivered=1, got %v", body["delivered"])
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet expectations: %v", err)
+		t.Errorf("expected 200 even on sender log failure, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestBroadcast_MissingMessage(t *testing.T) {
+	setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "00000000-0000-0000-0000-00000000000d"}}
+	c.Request = httptest.NewRequest("POST", "/workspaces/00000000-0000-0000-0000-00000000000d/broadcast", bytes.NewBufferString("{}"))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Broadcast(c)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+func TestBroadcast_MissingBody(t *testing.T) {
+	setupTestDB(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewBroadcastHandler(broadcaster)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "00000000-0000-0000-0000-00000000000e"}}
+	c.Request = httptest.NewRequest("POST", "/workspaces/00000000-0000-0000-0000-00000000000e/broadcast", nil)
+	// no Content-Type and no body
+
+	handler.Broadcast(c)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("expected 400, got %d: %s", w.Code, w.Body.String())
+	}
+}
+
+// TestBroadcast_Truncate tests that messages are truncated with the Unicode ellipsis
+// character (U+2026) when len(msg) > max. The truncated output is max runes + "…".
+func TestBroadcast_Truncate(t *testing.T) {
+	cases := []struct {
+		msg    string
+		max    int
+		expect string
+	}{
+		{"short", 120, "short"}, // under max — no truncation
+		// exactly 120 chars → unchanged
+		{"exactly120chars1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111", 120, "exactly120chars111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111…"},
+		// 21 runes at max=20 → 20 + "…" = 21 chars
+		{"this is a longer message that needs truncating", 20, "this is a longer mes…"},
+		// at-max boundary: 20 chars at max=20 → no truncation
+		{"exactly twenty chars", 20, "exactly twenty chars"},
+		// over max: 11 chars at max=10 → 10 + "…" = 11
+		{"hello world!", 10, "hello worl…"},
+		// Unicode: 3-rune string at max=3 → unchanged
+		{"日本語", 3, "日本語"},
+		// Empty string → unchanged
+		{"", 120, ""},
+	}
+	for _, tc := range cases {
+		result := broadcastTruncate(tc.msg, tc.max)
+		if result != tc.expect {
+			t.Errorf("broadcastTruncate(%q, %d) = %q; want %q", tc.msg, tc.max, result, tc.expect)
+		}
 	}
 }

workspace-server/internal/handlers/workspace_provision.go

@@ -15,6 +15,7 @@ import (
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/models"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/provisioner"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/wsauth"
+	"gopkg.in/yaml.v3"
 )
 
 // logProvisionPanic is the deferred recover at the top of every provision
@@ -472,9 +473,10 @@ func configDirName(workspaceID string) string {
 // runtime means bumping both this list and the Docker image tags.
 // knownRuntimes is populated from manifest.json at service init (see
 // runtime_registry.go). The package init order is:
-//   1. var knownRuntimes = fallbackRuntimes
-//   2. init() calls initKnownRuntimes() which replaces it if
-//      manifest.json is readable.
+//  1. var knownRuntimes = fallbackRuntimes
+//  2. init() calls initKnownRuntimes() which replaces it if
+//     manifest.json is readable.
+//
 // The fallback matters for unit tests that don't mount the manifest.
 //
 // "external" is a first-class runtime that intentionally does NOT
@@ -539,6 +541,9 @@ func (h *WorkspaceHandler) ensureDefaultConfig(workspaceID string, payload model
 		// org_import.go; consolidating prevents silent drift.
 		model = models.DefaultModel(runtime)
 	}
+	if runtime == "claude-code" {
+		model = normalizeClaudeCodeModel(model)
+	}
 
 	// Sanitize name/role/model for YAML safety — always double-quote so
 	// a crafted value with a newline or colon can't terminate the scalar
@@ -554,6 +559,11 @@ func (h *WorkspaceHandler) ensureDefaultConfig(workspaceID string, payload model
 	quoteModel := yamlQuote(model)
 	configYAML := fmt.Sprintf("name: %s\ndescription: %s\nversion: 1.0.0\ntier: %d\nruntime: %s\n",
 		quoteName, quoteRole, payload.Tier, runtime)
+	if runtime == "claude-code" {
+		if providersYAML := h.defaultTemplateProvidersYAML(runtime); providersYAML != "" {
+			configYAML += providersYAML + "\n"
+		}
+	}
 
 	// Model always at top level — config.py reads raw["model"] for all runtimes.
 	configYAML += fmt.Sprintf("model: %s\n", quoteModel)
@@ -563,7 +573,11 @@ func (h *WorkspaceHandler) ensureDefaultConfig(workspaceID string, payload model
 	// and preflight already validates that the env vars are present before
 	// the agent loop starts.  Hardcoding token names here caused #1028
 	// (expired CLAUDE_CODE_OAUTH_TOKEN baked into config.yaml).
-	configYAML += "runtime_config:\n  timeout: 0\n"
+	configYAML += "runtime_config:\n"
+	if runtime == "claude-code" {
+		configYAML += fmt.Sprintf("  model: %s\n", quoteModel)
+	}
+	configYAML += "  timeout: 0\n"
 
 	files["config.yaml"] = []byte(configYAML)
 
@@ -571,6 +585,60 @@ func (h *WorkspaceHandler) ensureDefaultConfig(workspaceID string, payload model
 	return files
 }
 
+func normalizeClaudeCodeModel(model string) string {
+	model = strings.TrimSpace(model)
+	if before, after, ok := strings.Cut(model, "/"); ok && before != "" && after != "" {
+		return after
+	}
+	return model
+}
+
+func (h *WorkspaceHandler) defaultTemplateProvidersYAML(runtime string) string {
+	if h.configsDir == "" {
+		return ""
+	}
+	templateName := runtime + "-default"
+	templatePath, err := resolveInsideRoot(h.configsDir, templateName)
+	if err != nil {
+		log.Printf("Provisioner: default template providers skipped for runtime %s: %v", runtime, err)
+		return ""
+	}
+	data, err := os.ReadFile(filepath.Join(templatePath, "config.yaml"))
+	if err != nil {
+		return ""
+	}
+
+	var root yaml.Node
+	if err := yaml.Unmarshal(data, &root); err != nil {
+		log.Printf("Provisioner: default template providers skipped for runtime %s: invalid YAML: %v", runtime, err)
+		return ""
+	}
+	if len(root.Content) == 0 || root.Content[0].Kind != yaml.MappingNode {
+		return ""
+	}
+
+	mapping := root.Content[0]
+	for i := 0; i+1 < len(mapping.Content); i += 2 {
+		if mapping.Content[i].Value != "providers" {
+			continue
+		}
+		out := yaml.Node{
+			Kind: yaml.MappingNode,
+			Content: []*yaml.Node{
+				{Kind: yaml.ScalarNode, Value: "providers"},
+				mapping.Content[i+1],
+			},
+		}
+		encoded, err := yaml.Marshal(&out)
+		if err != nil {
+			log.Printf("Provisioner: default template providers skipped for runtime %s: marshal failed: %v", runtime, err)
+			return ""
+		}
+		return strings.TrimRight(string(encoded), "\n")
+	}
+	return ""
+}
+
 // deriveProviderFromModelSlug maps a hermes-agent model slug prefix to
 // its provider name — a Go translation of the case statement in
 // workspace-configs-templates/hermes/scripts/derive-provider.sh that we

workspace-server/internal/handlers/workspace_provision_auto_test.go

@@ -144,6 +144,7 @@ func TestProvisionWorkspaceAuto_RoutesToCPWhenSet(t *testing.T) {
 	rec := &trackingCPProv{startErr: errors.New("simulated CP rejection")}
 	bcast := &concurrentSafeBroadcaster{}
 	h := NewWorkspaceHandler(bcast, nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, h)
 	h.SetCPProvisioner(rec)
 
 	wsID := "ws-routes-to-cp-0123456789abcdef"
@@ -595,6 +596,7 @@ func TestRestartWorkspaceAuto_RoutesToCPWhenSet(t *testing.T) {
 
 	// Mock DB so cpStopWithRetry can run without a real Postgres.
 	mock := setupTestDB(t)
+	waitForHandlerAsyncBeforeDBCleanup(t, h)
 	mock.MatchExpectationsInOrder(false)
 	// provisionWorkspaceCP runs in the goroutine and will hit secrets
 	// SELECTs + UPDATE workspace as failed (we make CP Start return
@@ -670,6 +672,7 @@ func TestRestartWorkspaceAuto_RoutesToDockerWhenOnlyDocker(t *testing.T) {
 
 	bcast := &concurrentSafeBroadcaster{}
 	h := NewWorkspaceHandler(bcast, nil, "http://localhost:8080", t.TempDir())
+	waitForHandlerAsyncBeforeDBCleanup(t, h)
 	stub := &stoppingLocalProv{}
 	h.provisioner = stub
 

workspace-server/internal/handlers/workspace_provision_test.go

@@ -2,6 +2,7 @@ package handlers
 
 import (
 	"context"
+	"database/sql"
 	"fmt"
 	"net/http"
 	"os"
@@ -260,6 +261,67 @@ func TestEnsureDefaultConfig_ClaudeCode(t *testing.T) {
 	}
 }
 
+func TestEnsureDefaultConfig_ClaudeCodeCopiesProviderRegistry(t *testing.T) {
+	broadcaster := newTestBroadcaster()
+	configsDir := t.TempDir()
+	templateDir := filepath.Join(configsDir, "claude-code-default")
+	if err := os.MkdirAll(templateDir, 0o755); err != nil {
+		t.Fatalf("mkdir template: %v", err)
+	}
+	if err := os.WriteFile(filepath.Join(templateDir, "config.yaml"), []byte(`
+name: Claude Code Agent
+runtime: claude-code
+providers:
+  - name: anthropic-oauth
+    auth_mode: oauth
+    model_aliases: [sonnet]
+    auth_env: [CLAUDE_CODE_OAUTH_TOKEN]
+  - name: minimax
+    auth_mode: third_party_anthropic_compat
+    model_prefixes: [minimax-]
+    base_url: https://api.minimax.io/anthropic
+    auth_env: [MINIMAX_API_KEY, ANTHROPIC_AUTH_TOKEN]
+runtime_config:
+  model: sonnet
+`), 0o644); err != nil {
+		t.Fatalf("write template: %v", err)
+	}
+	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", configsDir)
+
+	files := handler.ensureDefaultConfig("ws-code-123", models.CreateWorkspacePayload{
+		Name:    "Code Agent",
+		Tier:    4,
+		Runtime: "claude-code",
+		Model:   "minimax/MiniMax-M2.7",
+	})
+
+	var parsed struct {
+		Model     string `yaml:"model"`
+		Providers []struct {
+			Name          string   `yaml:"name"`
+			ModelPrefixes []string `yaml:"model_prefixes"`
+		} `yaml:"providers"`
+		RuntimeConfig struct {
+			Model string `yaml:"model"`
+		} `yaml:"runtime_config"`
+	}
+	if err := yaml.Unmarshal(files["config.yaml"], &parsed); err != nil {
+		t.Fatalf("generated YAML invalid: %v\n%s", err, files["config.yaml"])
+	}
+	if parsed.Model != "MiniMax-M2.7" {
+		t.Fatalf("top-level model = %q, want MiniMax-M2.7\n%s", parsed.Model, files["config.yaml"])
+	}
+	if parsed.RuntimeConfig.Model != "MiniMax-M2.7" {
+		t.Fatalf("runtime_config.model = %q, want MiniMax-M2.7\n%s", parsed.RuntimeConfig.Model, files["config.yaml"])
+	}
+	if len(parsed.Providers) != 2 {
+		t.Fatalf("providers len = %d, want 2\n%s", len(parsed.Providers), files["config.yaml"])
+	}
+	if parsed.Providers[1].Name != "minimax" || len(parsed.Providers[1].ModelPrefixes) != 1 || parsed.Providers[1].ModelPrefixes[0] != "minimax-" {
+		t.Fatalf("minimax provider registry not preserved: %+v\n%s", parsed.Providers, files["config.yaml"])
+	}
+}
+
 func TestEnsureDefaultConfig_CustomModel(t *testing.T) {
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
@@ -634,6 +696,11 @@ func TestSeedInitialMemories_EmptyMemoriesNil(t *testing.T) {
 // ==================== buildProvisionerConfig ====================
 
 func TestBuildProvisionerConfig_BasicFields(t *testing.T) {
+	mock := setupTestDB(t)
+	mock.ExpectQuery(`SELECT COALESCE\(workspace_dir`).
+		WithArgs("ws-basic").
+		WillReturnRows(sqlmock.NewRows([]string{"workspace_dir", "workspace_access"}).AddRow("", "none"))
+
 	broadcaster := newTestBroadcaster()
 	tmpDir := t.TempDir()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", tmpDir)
@@ -678,6 +745,14 @@ func TestBuildProvisionerConfig_BasicFields(t *testing.T) {
 }
 
 func TestBuildProvisionerConfig_WorkspacePathFromEnv(t *testing.T) {
+	mock := setupTestDB(t)
+	mock.ExpectQuery(`SELECT COALESCE\(workspace_dir`).
+		WithArgs("ws-env").
+		WillReturnError(sql.ErrNoRows)
+	mock.ExpectQuery(`SELECT digest FROM runtime_image_pins`).
+		WithArgs("claude-code").
+		WillReturnError(sql.ErrNoRows)
+
 	broadcaster := newTestBroadcaster()
 	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
 

workspace-server/internal/handlers/workspace_test.go

@@ -414,6 +414,44 @@ func TestWorkspaceCreate_DefaultsApplied(t *testing.T) {
 	}
 }
 
+func TestWorkspaceCreate_SaaSHardForcesTier4(t *testing.T) {
+	mock := setupTestDB(t)
+	setupTestRedis(t)
+	broadcaster := newTestBroadcaster()
+	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
+	handler.SetCPProvisioner(&trackingCPProv{})
+
+	mock.ExpectBegin()
+	mock.ExpectExec("INSERT INTO workspaces").
+		WithArgs(sqlmock.AnyArg(), "SaaS External Agent", nil, 4, "external", sqlmock.AnyArg(), (*string)(nil), nil, "none", (*int64)(nil), models.DefaultMaxConcurrentTasks, "push").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectCommit()
+	mock.ExpectExec("INSERT INTO canvas_layouts").
+		WithArgs(sqlmock.AnyArg(), float64(0), float64(0)).
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec("INSERT INTO structure_events").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec("UPDATE workspaces SET url").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec("INSERT INTO structure_events").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	body := `{"name":"SaaS External Agent","runtime":"external","external":true,"url":"https://example.com/agent","tier":2}`
+	c.Request = httptest.NewRequest("POST", "/workspaces", bytes.NewBufferString(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Create(c)
+
+	if w.Code != http.StatusCreated {
+		t.Errorf("expected status 201, got %d: %s", w.Code, w.Body.String())
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("unmet sqlmock expectations: %v", err)
+	}
+}
+
 // TestWorkspaceCreate_WithSecrets_Persists asserts that secrets in the create
 // payload are written to workspace_secrets inside the same transaction as the
 // workspace row, and that the handler returns 201.

workspace-server/internal/models/runtime_defaults.go

@@ -23,8 +23,8 @@ package models
 //   - claude-code: "sonnet" — Anthropic's CLI accepts the short
 //     name and resolves it via the operator's anthropic-oauth or
 //     ANTHROPIC_API_KEY chain.
-//   - everything else (hermes, langgraph, crewai, autogen, deepagents,
-//     codex, openclaw, gemini-cli, external, ""): a fully-qualified
+//   - everything else (hermes, langgraph, autogen, codex, openclaw,
+//     external, ""): a fully-qualified
 //     vendor:model slug that the universal MODEL_PROVIDER chain in
 //     molecule-core PR #247 can route via per-vendor required_env.
 //

workspace-server/internal/models/runtime_defaults_test.go

@@ -21,12 +21,9 @@ func TestDefaultModel(t *testing.T) {
 		// as a generic "unknown" failure.
 		{"hermes", "anthropic:claude-opus-4-7"},
 		{"langgraph", "anthropic:claude-opus-4-7"},
-		{"crewai", "anthropic:claude-opus-4-7"},
 		{"autogen", "anthropic:claude-opus-4-7"},
-		{"deepagents", "anthropic:claude-opus-4-7"},
 		{"codex", "anthropic:claude-opus-4-7"},
 		{"openclaw", "anthropic:claude-opus-4-7"},
-		{"gemini-cli", "anthropic:claude-opus-4-7"},
 		{"external", "anthropic:claude-opus-4-7"},
 
 		// Unknown / empty — fall through to universal default rather

workspace-server/internal/provisioner/cp_provisioner_test.go

@@ -217,6 +217,59 @@ func TestStart_HappyPath(t *testing.T) {
 	}
 }
 
+func TestStart_SendsTemplateAndGeneratedConfigFiles(t *testing.T) {
+	tmpl := t.TempDir()
+	if err := os.WriteFile(filepath.Join(tmpl, "config.yaml"), []byte("name: template\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(tmpl, "adapter.py"), bytes.Repeat([]byte("x"), cpConfigFilesMaxBytes), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Mkdir(filepath.Join(tmpl, "prompts"), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(tmpl, "prompts", "system.md"), []byte("hello"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	var body cpProvisionRequest
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			t.Errorf("decode request: %v", err)
+		}
+		w.WriteHeader(http.StatusCreated)
+		_, _ = io.WriteString(w, `{"instance_id":"i-abc123","state":"pending"}`)
+	}))
+	defer srv.Close()
+
+	p := &CPProvisioner{baseURL: srv.URL, orgID: "org-1", httpClient: srv.Client()}
+	_, err := p.Start(context.Background(), WorkspaceConfig{
+		WorkspaceID:  "ws-1",
+		Runtime:      "claude-code",
+		Tier:         4,
+		PlatformURL:  "http://tenant",
+		TemplatePath: tmpl,
+		ConfigFiles: map[string][]byte{
+			"config.yaml": []byte("name: generated\n"),
+		},
+	})
+	if err != nil {
+		t.Fatalf("Start: %v", err)
+	}
+
+	wantConfig := base64.StdEncoding.EncodeToString([]byte("name: generated\n"))
+	if got := body.ConfigFiles["config.yaml"]; got != wantConfig {
+		t.Errorf("config.yaml payload = %q, want generated override %q", got, wantConfig)
+	}
+	wantPrompt := base64.StdEncoding.EncodeToString([]byte("hello"))
+	if got := body.ConfigFiles["prompts/system.md"]; got != wantPrompt {
+		t.Errorf("prompt payload = %q, want %q", got, wantPrompt)
+	}
+	if _, ok := body.ConfigFiles["adapter.py"]; ok {
+		t.Error("non-config template file adapter.py must not be sent to CP")
+	}
+}
+
 // TestStart_Non201ReturnsStructuredError — when CP returns 401 with a
 // structured {"error":"..."} body, Start surfaces that error message.
 // Verifies the defense against log-leaking raw upstream bodies.
@@ -519,9 +572,9 @@ func TestStop_4xxResponseSurfacesError(t *testing.T) {
 func TestStop_2xxVariantsAllSucceed(t *testing.T) {
 	primeInstanceIDLookup(t, map[string]string{"ws-1": "i-ok"})
 	for _, code := range []int{
-		http.StatusOK,           // 200
-		http.StatusAccepted,     // 202
-		http.StatusNoContent,    // 204
+		http.StatusOK,        // 200
+		http.StatusAccepted,  // 202
+		http.StatusNoContent, // 204
 	} {
 		srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
 			w.WriteHeader(code)
@@ -589,11 +642,11 @@ func TestIsRunning_ParsesStateField(t *testing.T) {
 			_, _ = io.WriteString(w, `{"state":"`+state+`"}`)
 		}))
 		p := &CPProvisioner{
-			baseURL:    srv.URL,
-			orgID:      "org-1",
+			baseURL:      srv.URL,
+			orgID:        "org-1",
 			sharedSecret: "s3cret",
 			adminToken:   "tok-xyz",
-			httpClient: srv.Client(),
+			httpClient:   srv.Client(),
 		}
 		got, err := p.IsRunning(context.Background(), "ws-1")
 		srv.Close()

workspace-server/internal/provisioner/localbuild_test.go

@@ -190,7 +190,7 @@ func TestEnsureLocalImage_RepoNotFound(t *testing.T) {
 	opts.HTTPClient = srv.Client()
 	opts.remoteHeadSha = nil // exercise real HTTP path
 
-	_, err := ensureLocalImageWithOpts(context.Background(), "crewai", opts)
+	_, err := ensureLocalImageWithOpts(context.Background(), "hermes", opts)
 	if err == nil {
 		t.Fatalf("expected error, got nil")
 	}

workspace-server/internal/provisioner/provisioner.go

@@ -35,6 +35,19 @@ import (
 // drift-risk #6.
 var ErrNoBackend = errors.New("provisioner: no backend configured (zero-valued receiver)")
 
+// ErrUnresolvableRuntime is returned by selectImage when a workspace
+// names a runtime that has no resolvable image (not in RuntimeImages and
+// no operator-pinned cfg.Image). RFC internal#483 + security review 4269:
+// previously such a request silently fell through to DefaultImage
+// (langgraph) — a user asking for crewai would get a langgraph container
+// with no signal. The CTO standing directive
+// (feedback_platform_must_hardgate_base_contract) is fail-closed: a
+// named-but-unresolvable runtime must reject with a structured,
+// runtime-naming error so the existing provision-failed notify/log path
+// surfaces it, NOT silently degrade. The genuinely-unspecified (empty)
+// runtime is still a distinct, legitimate path that keeps DefaultImage.
+var ErrUnresolvableRuntime = errors.New("provisioner: requested runtime has no resolvable image")
+
 // RuntimeImages maps runtime names to their Docker image refs.
 // Each standalone template repo publishes its image via the reusable
 // publish-template-image workflow in molecule-ci on every main merge.
@@ -104,20 +117,33 @@ type WorkspaceConfig struct {
 // selectImage resolves the final Docker image ref for a workspace. The handler
 // layer is the source of truth — if it set cfg.Image (the digest-pinned form
 // from runtime_image_pins, #2272), honor that. Otherwise fall back to the
-// runtime→tag lookup in RuntimeImages (legacy `:latest` behavior). When the
-// runtime isn't recognized either, fall back to DefaultImage so Start() still
-// has something to hand Docker — surfacing a "No such image" later is more
-// actionable than a silent "" panic in ContainerCreate.
-func selectImage(cfg WorkspaceConfig) string {
+// runtime→tag lookup in RuntimeImages (legacy `:latest` behavior).
+//
+// Fail-closed contract (RFC internal#483 / security review 4269 /
+// feedback_platform_must_hardgate_base_contract): if the workspace NAMES a
+// runtime that resolves to no image (not in RuntimeImages, no pinned
+// cfg.Image), reject with ErrUnresolvableRuntime instead of silently
+// substituting DefaultImage. Pre-fix, removing crewai/deepagents/gemini-cli
+// from the catalog left those create requests silently provisioning a
+// langgraph container — the user asked for crewai and got langgraph with no
+// signal. The error propagates through Start → markProvisionFailed, which
+// already broadcasts WorkspaceProvisionFailed and records the message.
+//
+// The genuinely-unspecified runtime (empty cfg.Runtime, e.g. an org template
+// that doesn't pin one) is an intended distinct path and still resolves to
+// DefaultImage — only a NAMED-but-unresolvable runtime is rejected.
+func selectImage(cfg WorkspaceConfig) (string, error) {
 	if cfg.Image != "" {
-		return cfg.Image
+		return cfg.Image, nil
 	}
 	if cfg.Runtime != "" {
 		if img, ok := RuntimeImages[cfg.Runtime]; ok {
-			return img
+			return img, nil
 		}
+		return "", fmt.Errorf("%w: runtime %q (known runtimes: %v)",
+			ErrUnresolvableRuntime, cfg.Runtime, knownRuntimes)
 	}
-	return DefaultImage
+	return DefaultImage, nil
 }
 
 // Workspace-access constants for #65. Matches the CHECK constraint on
@@ -189,6 +215,24 @@ const containerNamePrefix = "ws-"
 // (the wiped-DB case after `docker compose down -v`).
 const LabelManaged = "molecule.platform.managed"
 
+// AgentUID / AgentGID are the uid/gid of the unprivileged `agent` user that
+// every workspace template creates and drops to via `gosu agent` before
+// exec'ing the runtime (the a2a_mcp_server runs under this uid). The value is
+// fixed at 1000:1000 across all templates — see:
+//   - workspace-configs-templates/claude-code-default/Dockerfile (`useradd -u 1000 ... agent`)
+//   - workspace-configs-templates/hermes/Dockerfile               (`useradd -u 1000 ... agent`)
+//   - workspace/entrypoint.sh                                     (`exec gosu agent` — "uid 1000")
+//
+// Files the platform injects into /configs AFTER the entrypoint's
+// `chown -R agent:agent /configs` (the post-start #418 re-injection and the
+// pre-start #1877 volume write) must be owned by this uid/gid, otherwise the
+// agent-uid MCP server hits EACCES reading /configs/.auth_token, sends an
+// empty bearer, and the platform 401s on /registry/{id}/peers (list_peers).
+const (
+	AgentUID = 1000
+	AgentGID = 1000
+)
+
 // managedLabels is the canonical label map applied to every workspace
 // container + volume. Pulled out so a future addition (e.g. instance
 // UUID for multi-platform-shared-daemon disambiguation) is one edit.
@@ -318,7 +362,15 @@ func (p *Provisioner) Start(ctx context.Context, cfg WorkspaceConfig) (string, e
 
 	env := buildContainerEnv(cfg)
 
-	image := selectImage(cfg)
+	image, imgErr := selectImage(cfg)
+	if imgErr != nil {
+		// Fail-closed: a named-but-unresolvable runtime must not silently
+		// become DefaultImage (RFC internal#483 / review 4269). The caller's
+		// error path (markProvisionFailed) broadcasts the failure + records
+		// the message so the canvas surfaces it.
+		log.Printf("Provisioner: refusing to start %s: %v", cfg.WorkspaceID, imgErr)
+		return "", imgErr
+	}
 
 	// Local-build mode (issue #63 / Task #194): when MOLECULE_IMAGE_REGISTRY
 	// is unset, the OSS contributor path skips the registry pull entirely
@@ -810,6 +862,15 @@ func ApplyTierConfig(hostCfg *container.HostConfig, cfg WorkspaceConfig, configM
 
 // CopyTemplateToContainer copies files from a host directory into /configs in the container.
 func (p *Provisioner) CopyTemplateToContainer(ctx context.Context, containerID, templatePath string) error {
+	buf, err := buildTemplateTar(templatePath)
+	if err != nil {
+		return err
+	}
+
+	return p.cli.CopyToContainer(ctx, containerID, "/configs", buf, container.CopyToContainerOptions{})
+}
+
+func buildTemplateTar(templatePath string) (*bytes.Buffer, error) {
 	// Resolve symlinks at the root before walking. filepath.Walk does
 	// NOT follow a symlink that IS the root — it Lstats the path, sees
 	// a symlink (non-directory), and emits exactly one entry without
@@ -832,6 +893,15 @@ func (p *Provisioner) CopyTemplateToContainer(ctx context.Context, containerID,
 		if err != nil {
 			return err
 		}
+		// OFFSEC-010: skip symlinks to prevent path traversal via malicious
+		// template symlinks (e.g. template/.ssh → /root/.ssh). filepath.Walk
+		// follows symlinks by default, so without this guard a crafted symlink
+		// inside the template directory could escape to include arbitrary host
+		// files in the tar archive. We intentionally skip rather than error so
+		// a broken symlink in an org template is a silent no-op.
+		if info.Mode()&os.ModeSymlink != 0 {
+			return nil
+		}
 		rel, err := filepath.Rel(templatePath, path)
 		if err != nil {
 			return err
@@ -872,17 +942,27 @@ func (p *Provisioner) CopyTemplateToContainer(ctx context.Context, containerID,
 		return nil
 	})
 	if err != nil {
-		return fmt.Errorf("failed to create tar from %s: %w", templatePath, err)
+		return nil, fmt.Errorf("failed to create tar from %s: %w", templatePath, err)
 	}
 	if err := tw.Close(); err != nil {
-		return fmt.Errorf("failed to close tar writer: %w", err)
+		return nil, fmt.Errorf("failed to close tar writer: %w", err)
 	}
 
-	return p.cli.CopyToContainer(ctx, containerID, "/configs", &buf, container.CopyToContainerOptions{})
+	return &buf, nil
 }
 
-// WriteFilesToContainer writes in-memory files into /configs in the container.
-func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID string, files map[string][]byte) error {
+// buildConfigFilesTar builds the tar stream that WriteFilesToContainer streams
+// into /configs via CopyToContainer. Every entry is stamped Uid/Gid = agent
+// (AgentUID/AgentGID) so the files land agent-owned after extraction. This is
+// the issue #418 post-start re-injection path: it runs AFTER the template
+// entrypoint's `chown -R agent:agent /configs`, so without explicit ownership
+// in the tar header the files extract as root:root (tar Uid/Gid default 0) and
+// the agent-uid MCP server can no longer read /configs/.auth_token (and
+// /configs/.platform_inbound_secret) → empty bearer → list_peers 401.
+//
+// Pulled out as a pure function so the ownership contract is unit-testable
+// without a live Docker daemon (mirrors buildTemplateTar).
+func buildConfigFilesTar(files map[string][]byte) (*bytes.Buffer, error) {
 	var buf bytes.Buffer
 	tw := tar.NewWriter(&buf)
 
@@ -895,8 +975,10 @@ func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID str
 				Typeflag: tar.TypeDir,
 				Name:     dir + "/",
 				Mode:     0755,
+				Uid:      AgentUID,
+				Gid:      AgentGID,
 			}); err != nil {
-				return fmt.Errorf("failed to write tar dir header for %s: %w", dir, err)
+				return nil, fmt.Errorf("failed to write tar dir header for %s: %w", dir, err)
 			}
 			createdDirs[dir] = true
 		}
@@ -905,19 +987,30 @@ func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID str
 			Name: name,
 			Mode: 0644,
 			Size: int64(len(data)),
+			Uid:  AgentUID,
+			Gid:  AgentGID,
 		}
 		if err := tw.WriteHeader(header); err != nil {
-			return fmt.Errorf("failed to write tar header for %s: %w", name, err)
+			return nil, fmt.Errorf("failed to write tar header for %s: %w", name, err)
 		}
 		if _, err := tw.Write(data); err != nil {
-			return fmt.Errorf("failed to write tar data for %s: %w", name, err)
+			return nil, fmt.Errorf("failed to write tar data for %s: %w", name, err)
 		}
 	}
 	if err := tw.Close(); err != nil {
-		return fmt.Errorf("failed to close tar writer: %w", err)
+		return nil, fmt.Errorf("failed to close tar writer: %w", err)
 	}
+	return &buf, nil
+}
 
-	return p.cli.CopyToContainer(ctx, containerID, "/configs", &buf, container.CopyToContainerOptions{})
+// WriteFilesToContainer writes in-memory files into /configs in the container,
+// agent-owned (see buildConfigFilesTar).
+func (p *Provisioner) WriteFilesToContainer(ctx context.Context, containerID string, files map[string][]byte) error {
+	buf, err := buildConfigFilesTar(files)
+	if err != nil {
+		return err
+	}
+	return p.cli.CopyToContainer(ctx, containerID, "/configs", buf, container.CopyToContainerOptions{})
 }
 
 // CopyToContainer exposes CopyToContainer from the Docker client for use by other packages.
@@ -1007,13 +1100,28 @@ func (p *Provisioner) ReadFromVolume(ctx context.Context, volumeName, filePath s
 	return clean, nil
 }
 
+// writeAuthTokenVolumeCmd is the shell command the throwaway alpine container
+// runs to seed /vol/.auth_token. alpine runs it as root, so without the
+// explicit `chown 1000:1000` the file stays root:root after the template
+// entrypoint's `chown -R agent:agent /configs` has already run — the agent-uid
+// (AgentUID) MCP server then gets EACCES reading it → empty bearer →
+// list_peers 401. Pulled out as a pure function so the ownership contract is
+// unit-testable without a live Docker daemon. Issue #1877.
+func writeAuthTokenVolumeCmd() string {
+	return fmt.Sprintf(
+		"mkdir -p /vol && printf '%%s' $TOKEN > /vol/.auth_token && chmod 0600 /vol/.auth_token && chown %d:%d /vol/.auth_token",
+		AgentUID, AgentGID,
+	)
+}
+
 // WriteAuthTokenToVolume writes the workspace auth token into the config volume
 // BEFORE the container starts, eliminating the token-injection race window where
 // a restarted container could read a stale token from /configs/.auth_token before
 // WriteFilesToContainer writes the new one. Issue #1877.
 //
 // Uses a throwaway alpine container to write directly to the named volume,
-// bypassing the container lifecycle entirely.
+// bypassing the container lifecycle entirely. The written file is chowned to
+// the agent uid/gid (see writeAuthTokenVolumeCmd).
 func (p *Provisioner) WriteAuthTokenToVolume(ctx context.Context, workspaceID, token string) error {
 	if p == nil || p.cli == nil {
 		return ErrNoBackend
@@ -1021,7 +1129,7 @@ func (p *Provisioner) WriteAuthTokenToVolume(ctx context.Context, workspaceID, t
 	volName := ConfigVolumeName(workspaceID)
 	resp, err := p.cli.ContainerCreate(ctx, &container.Config{
 		Image: "alpine",
-		Cmd:   []string{"sh", "-c", "mkdir -p /vol && printf '%s' $TOKEN > /vol/.auth_token && chmod 0600 /vol/.auth_token"},
+		Cmd:   []string{"sh", "-c", writeAuthTokenVolumeCmd()},
 		Env:   []string{"TOKEN=" + token},
 	}, &container.HostConfig{
 		Binds: []string{volName + ":/vol"},

workspace-server/internal/provisioner/provisioner_test.go

@@ -1,7 +1,9 @@
 package provisioner
 
 import (
+	"archive/tar"
 	"errors"
+	"io"
 	"os"
 	"path/filepath"
 	"strings"
@@ -62,6 +64,72 @@ func TestValidateConfigSource_TemplateIsDirName(t *testing.T) {
 	}
 }
 
+func TestStartSeedsConfigsBeforeContainerStart(t *testing.T) {
+	src, err := os.ReadFile("provisioner.go")
+	if err != nil {
+		t.Fatalf("read provisioner.go: %v", err)
+	}
+	text := string(src)
+	copyTemplate := strings.Index(text, "p.CopyTemplateToContainer(ctx, resp.ID, cfg.TemplatePath)")
+	writeFiles := strings.Index(text, "p.WriteFilesToContainer(ctx, resp.ID, cfg.ConfigFiles)")
+	start := strings.Index(text, "p.cli.ContainerStart(ctx, resp.ID, container.StartOptions{})")
+
+	if copyTemplate < 0 || writeFiles < 0 || start < 0 {
+		t.Fatalf("expected Start to copy template, write config files, and start container")
+	}
+	if copyTemplate >= start || writeFiles >= start {
+		t.Fatalf("config seeding must happen before ContainerStart: copyTemplate=%d writeFiles=%d start=%d", copyTemplate, writeFiles, start)
+	}
+}
+
+func TestBuildTemplateTar_SkipsSymlinks(t *testing.T) {
+	dir := t.TempDir()
+	if err := os.WriteFile(filepath.Join(dir, "config.yaml"), []byte("name: safe\n"), 0644); err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+	outside := filepath.Join(t.TempDir(), "secret.txt")
+	if err := os.WriteFile(outside, []byte("do-not-copy\n"), 0644); err != nil {
+		t.Fatalf("write outside target: %v", err)
+	}
+	if err := os.Symlink(outside, filepath.Join(dir, "linked-secret.txt")); err != nil {
+		t.Fatalf("create symlink: %v", err)
+	}
+
+	buf, err := buildTemplateTar(dir)
+	if err != nil {
+		t.Fatalf("buildTemplateTar: %v", err)
+	}
+
+	names := map[string]string{}
+	tr := tar.NewReader(buf)
+	for {
+		hdr, err := tr.Next()
+		if errors.Is(err, io.EOF) {
+			break
+		}
+		if err != nil {
+			t.Fatalf("read tar: %v", err)
+		}
+		body, err := io.ReadAll(tr)
+		if err != nil {
+			t.Fatalf("read body for %s: %v", hdr.Name, err)
+		}
+		names[hdr.Name] = string(body)
+	}
+
+	if got := names["config.yaml"]; got != "name: safe\n" {
+		t.Fatalf("config.yaml body = %q, want safe config", got)
+	}
+	if _, ok := names["linked-secret.txt"]; ok {
+		t.Fatalf("symlink entry was copied into template tar: %#v", names)
+	}
+	for name, body := range names {
+		if strings.Contains(body, "do-not-copy") {
+			t.Fatalf("symlink target leaked through %s: %q", name, body)
+		}
+	}
+}
+
 // baseHostConfig returns a fresh HostConfig with typical pre-tier binds,
 // mimicking what Start() builds before calling ApplyTierConfig.
 func baseHostConfig(pluginsPath string) *container.HostConfig {
@@ -445,7 +513,10 @@ func TestWorkspaceConfig_ResetClaudeSessionFieldPresent(t *testing.T) {
 // we lose the "one bad publish doesn't break every workspace" guarantee.
 func TestSelectImage_PrefersExplicitImage(t *testing.T) {
 	pinned := "ghcr.io/molecule-ai/workspace-template-claude-code@sha256:3d6761a97ed07d7d33cfc19a8fbab81175d9d9179618d493dbc00c5f7ef076a3"
-	got := selectImage(WorkspaceConfig{Runtime: "claude-code", Image: pinned})
+	got, err := selectImage(WorkspaceConfig{Runtime: "claude-code", Image: pinned})
+	if err != nil {
+		t.Fatalf("selectImage with cfg.Image=pinned: unexpected error %v", err)
+	}
 	if got != pinned {
 		t.Errorf("selectImage with cfg.Image=pinned: got %q, want %q", got, pinned)
 	}
@@ -455,28 +526,46 @@ func TestSelectImage_PrefersExplicitImage(t *testing.T) {
 // pin lookup deliberately bypassed via WORKSPACE_IMAGE_LOCAL_OVERRIDE).
 // selectImage must use the legacy runtime→:latest map.
 func TestSelectImage_FallsBackToRuntimeMap(t *testing.T) {
-	got := selectImage(WorkspaceConfig{Runtime: "claude-code", Image: ""})
+	got, err := selectImage(WorkspaceConfig{Runtime: "claude-code", Image: ""})
+	if err != nil {
+		t.Fatalf("selectImage with empty Image: unexpected error %v", err)
+	}
 	want := RuntimeImages["claude-code"]
 	if got != want {
 		t.Errorf("selectImage with empty Image: got %q, want %q", got, want)
 	}
 }
 
-// TestSelectImage_UnknownRuntimeFallsBackToDefault preserves today's
-// behavior — an unrecognized runtime resolves to DefaultImage rather than
-// "" so ContainerCreate gets a usable arg and surfaces a meaningful
-// "No such image" error if the default itself is missing.
-func TestSelectImage_UnknownRuntimeFallsBackToDefault(t *testing.T) {
-	got := selectImage(WorkspaceConfig{Runtime: "no-such-runtime"})
-	if got != DefaultImage {
-		t.Errorf("selectImage with unknown runtime: got %q, want DefaultImage %q", got, DefaultImage)
+// TestSelectImage_NamedUnresolvableRuntimeRejects pins the fail-closed
+// contract (RFC internal#483 / security review 4269 /
+// feedback_platform_must_hardgate_base_contract): a NAMED runtime with no
+// resolvable image must reject with ErrUnresolvableRuntime, NOT silently
+// substitute DefaultImage. Pre-fix this returned langgraph — a user asking
+// for a removed runtime (crewai/deepagents/gemini-cli) silently got a
+// langgraph container. "crewai" is the concrete regression from the
+// security finding.
+func TestSelectImage_NamedUnresolvableRuntimeRejects(t *testing.T) {
+	for _, rt := range []string{"no-such-runtime", "crewai", "deepagents", "gemini-cli"} {
+		got, err := selectImage(WorkspaceConfig{Runtime: rt})
+		if !errors.Is(err, ErrUnresolvableRuntime) {
+			t.Errorf("selectImage(%q): got err %v, want ErrUnresolvableRuntime", rt, err)
+		}
+		if got != "" {
+			t.Errorf("selectImage(%q): got image %q, want \"\" on reject", rt, got)
+		}
+		if err != nil && !strings.Contains(err.Error(), rt) {
+			t.Errorf("selectImage(%q): error must name the offending runtime, got %v", rt, err)
+		}
 	}
 }
 
 // TestSelectImage_EmptyRuntimeFallsBackToDefault: same invariant for the
 // no-runtime-supplied path (legacy callers / older handler code).
 func TestSelectImage_EmptyRuntimeFallsBackToDefault(t *testing.T) {
-	got := selectImage(WorkspaceConfig{})
+	got, err := selectImage(WorkspaceConfig{})
+	if err != nil {
+		t.Fatalf("selectImage with zero cfg: unexpected error %v (empty runtime is a legitimate DefaultImage path)", err)
+	}
 	if got != DefaultImage {
 		t.Errorf("selectImage with zero cfg: got %q, want DefaultImage %q", got, DefaultImage)
 	}
@@ -868,7 +957,7 @@ func TestIsImageNotFoundErr(t *testing.T) {
 		{"nil", nil, false},
 		{"moby no such image", fmtErr(`Error response from daemon: No such image: workspace-template:openclaw`), true},
 		{"no such image lowercase", fmtErr(`error: no such image: foo:bar`), true},
-		{"image not found", fmtErr(`Error: image "workspace-template:crewai" not found`), true},
+		{"image not found", fmtErr(`Error: image "workspace-template:hermes" not found`), true},
 		{"generic not found without image", fmtErr(`container not found`), false},
 		{"unrelated error", fmtErr(`connection refused`), false},
 		{"permission denied", fmtErr(`permission denied`), false},

workspace-server/internal/provisioner/registry.go

@@ -21,9 +21,6 @@ var knownRuntimes = []string{
 	"autogen",
 	"claude-code",
 	"codex",
-	"crewai",
-	"deepagents",
-	"gemini-cli",
 	"hermes",
 	"langgraph",
 	"openclaw",

workspace-server/internal/provisioner/registry_test.go

@@ -53,8 +53,8 @@ func TestRuntimeImage_AllKnownRuntimes(t *testing.T) {
 		}
 	}
 	// Pin the count so adding a runtime requires explicit test acknowledgement.
-	if len(knownRuntimes) != 9 {
-		t.Errorf("knownRuntimes length = %d, want 9 (autogen, claude-code, codex, crewai, deepagents, gemini-cli, hermes, langgraph, openclaw)", len(knownRuntimes))
+	if len(knownRuntimes) != 6 {
+		t.Errorf("knownRuntimes length = %d, want 6 (autogen, claude-code, codex, hermes, langgraph, openclaw)", len(knownRuntimes))
 	}
 }
 

workspace-server/internal/provisioner/token_ownership_test.go

@@ -0,0 +1,95 @@
+package provisioner
+
+import (
+	"archive/tar"
+	"errors"
+	"io"
+	"strings"
+	"testing"
+)
+
+// These tests pin the P0 fix for the fleet-wide list_peers 401 (Hermes and
+// every other template): the workspace-server token-injection paths wrote
+// /configs/.auth_token (and /configs/.platform_inbound_secret) as root:root
+// AFTER the template entrypoint's `chown -R agent:agent /configs` ran, so the
+// agent-uid (1000) MCP server (a2a_mcp_server, running via `gosu agent`) hit
+// `[Errno 13] Permission denied` reading the bearer → empty bearer → platform
+// 401 on /registry/{id}/peers (the literal tool_list_peers path).
+//
+// The agent uid is 1000:1000, verified from the templates:
+//   - workspace-configs-templates/claude-code-default/Dockerfile: `useradd -u 1000 ... agent`
+//   - workspace-configs-templates/hermes/Dockerfile:               `useradd -u 1000 ... agent`
+//   - workspace/entrypoint.sh / claude-code-default/entrypoint.sh:  `exec gosu agent` ("uid 1000")
+//
+// Both tests assert the real artifact (the tar headers Docker's CopyToContainer
+// honours for ownership, and the literal shell command the throwaway alpine
+// container runs), not a mock that bypasses ownership. They FAIL on pre-fix
+// code (no Uid/Gid in tar headers; no chown in the alpine command → root:root)
+// and PASS post-fix (agent-owned).
+
+// TestWriteFilesToContainerTar_FilesAreAgentOwned covers the issue #418
+// post-start re-injection path (WriteFilesToContainer): the tar it streams
+// into /configs via CopyToContainer must carry Uid/Gid = agent (1000) so the
+// extracted files land agent-readable, not root:root. This is the path that
+// (re)writes BOTH .auth_token and .platform_inbound_secret on a cadence.
+func TestWriteFilesToContainerTar_FilesAreAgentOwned(t *testing.T) {
+	files := map[string][]byte{
+		".auth_token":              []byte("tok-abc123"),
+		".platform_inbound_secret": []byte("inbound-secret-xyz"),
+		"nested/dir/file.txt":      []byte("data"),
+	}
+
+	buf, err := buildConfigFilesTar(files)
+	if err != nil {
+		t.Fatalf("buildConfigFilesTar: %v", err)
+	}
+
+	tr := tar.NewReader(buf)
+	seen := map[string]bool{}
+	for {
+		hdr, err := tr.Next()
+		if errors.Is(err, io.EOF) {
+			break
+		}
+		if err != nil {
+			t.Fatalf("read tar: %v", err)
+		}
+		if _, err := io.Copy(io.Discard, tr); err != nil {
+			t.Fatalf("drain %s: %v", hdr.Name, err)
+		}
+		seen[hdr.Name] = true
+		if hdr.Uid != AgentUID {
+			t.Fatalf("tar entry %q Uid = %d, want %d (agent) — root-owned injection causes the list_peers 401",
+				hdr.Name, hdr.Uid, AgentUID)
+		}
+		if hdr.Gid != AgentGID {
+			t.Fatalf("tar entry %q Gid = %d, want %d (agent)", hdr.Name, hdr.Gid, AgentGID)
+		}
+	}
+
+	for _, want := range []string{".auth_token", ".platform_inbound_secret"} {
+		if !seen[want] {
+			t.Fatalf("tar missing %q (seen: %v)", want, seen)
+		}
+	}
+}
+
+// TestWriteAuthTokenVolumeCmd_ChownsToAgent covers the issue #1877 pre-start
+// volume-write path (WriteAuthTokenToVolume): the throwaway alpine container
+// writes /vol/.auth_token then chmod 0600 but, pre-fix, never chowns it, so it
+// stays root:root (alpine runs the command as root). The literal command must
+// chown the file to the agent uid:gid so the agent-uid MCP server can read it.
+func TestWriteAuthTokenVolumeCmd_ChownsToAgent(t *testing.T) {
+	cmd := writeAuthTokenVolumeCmd()
+
+	if !strings.Contains(cmd, "chmod 0600 /vol/.auth_token") {
+		t.Fatalf("alpine cmd lost the 0600 chmod (regression): %q", cmd)
+	}
+
+	wantChown := "chown 1000:1000 /vol/.auth_token"
+	if !strings.Contains(cmd, wantChown) {
+		t.Fatalf("alpine cmd = %q, missing %q — without it .auth_token stays root:root "+
+			"and the agent-uid MCP server gets EACCES → empty bearer → list_peers 401",
+			cmd, wantChown)
+	}
+}

workspace-server/internal/registry/access_test.go

@@ -14,8 +14,9 @@ func setupMockDB(t *testing.T) sqlmock.Sqlmock {
 	if err != nil {
 		t.Fatalf("sqlmock: %v", err)
 	}
+	prevDB := db.DB
 	db.DB = mockDB
-	t.Cleanup(func() { mockDB.Close() })
+	t.Cleanup(func() { mockDB.Close(); db.DB = prevDB })
 	return mock
 }
 

workspace-server/internal/registry/healthsweep_test.go

@@ -31,8 +31,9 @@ func setupTestDB(t *testing.T) sqlmock.Sqlmock {
 	if err != nil {
 		t.Fatalf("failed to create sqlmock: %v", err)
 	}
+	prevDB := db.DB
 	db.DB = mockDB
-	t.Cleanup(func() { mockDB.Close() })
+	t.Cleanup(func() { mockDB.Close(); db.DB = prevDB })
 	return mock
 }
 

workspace-server/internal/registry/hibernation_test.go

@@ -17,8 +17,9 @@ func setupHibernationMock(t *testing.T) sqlmock.Sqlmock {
 	if err != nil {
 		t.Fatalf("sqlmock.New: %v", err)
 	}
+	prevDB := db.DB
 	db.DB = mockDB
-	t.Cleanup(func() { mockDB.Close() })
+	t.Cleanup(func() { mockDB.Close(); db.DB = prevDB })
 	return mock
 }
 

workspace-server/internal/registry/liveness_test.go

@@ -18,8 +18,9 @@ func setupLivenessTestDB(t *testing.T) sqlmock.Sqlmock {
 	if err != nil {
 		t.Fatalf("failed to create sqlmock: %v", err)
 	}
+	prevDB := db.DB
 	db.DB = mockDB
-	t.Cleanup(func() { mockDB.Close() })
+	t.Cleanup(func() { mockDB.Close(); db.DB = prevDB })
 	return mock
 }
 

workspace-server/internal/scheduler/scheduler_test.go

@@ -24,8 +24,9 @@ func setupTestDB(t *testing.T) sqlmock.Sqlmock {
 	if err != nil {
 		t.Fatalf("failed to create sqlmock: %v", err)
 	}
+	prevDB := db.DB
 	db.DB = mockDB
-	t.Cleanup(func() { mockDB.Close() })
+	t.Cleanup(func() { mockDB.Close(); db.DB = prevDB })
 	return mock
 }
 

workspace/a2a_mcp_server.py

@@ -698,8 +698,8 @@ def _format_channel_content(
 # --- MCP Server (JSON-RPC over stdio) ---
 
 
-def _warn_if_stdio_not_pipe(stdin_fd: int = 0, stdout_fd: int = 1) -> None:
-    """Warn when stdio isn't a pipe — but continue anyway.
+def _assert_stdio_is_pipe_compatible(stdin_fd: int = 0, stdout_fd: int = 1) -> None:
+    """Assert that stdio fds are pipe/socket/char-device compatible.
 
     The legacy asyncio.connect_read_pipe / connect_write_pipe transport
     rejected regular files, PTYs, and sockets with:
@@ -723,6 +723,10 @@ def _warn_if_stdio_not_pipe(stdin_fd: int = 0, stdout_fd: int = 1) -> None:
             )
 
 
+# Deprecated alias — the canonical name is _assert_stdio_is_pipe_compatible.
+_warn_if_stdio_not_pipe = _assert_stdio_is_pipe_compatible
+
+
 async def main():  # pragma: no cover
     """Run MCP server on stdio — reads JSON-RPC requests, writes responses.
 
@@ -979,7 +983,7 @@ def cli_main(transport: str = "stdio", port: int = 9100) -> None:  # pragma: no
     if transport == "http":
         asyncio.run(_run_http_server(port))
     else:
-        _warn_if_stdio_not_pipe()
+        _assert_stdio_is_pipe_compatible()
         asyncio.run(main())
 
 

workspace/inbox.py

@@ -431,6 +431,43 @@ def _is_self_notify_row(row: dict[str, Any]) -> bool:
     return source_id is None or source_id == ""
 
 
+def _is_self_echo_row(row: dict[str, Any], workspace_id: str) -> bool:
+    """Return True if ``row`` is a self-originated a2a_receive row.
+
+    Internal #469: when a workspace delegates to a target that never picks
+    up the task, ``tool_delegate_task`` calls ``report_activity`` which
+    POSTs to the platform with source_id set to the *sender's* workspace
+    UUID (mandated by spoof-defense in workspace-server's a2a_proxy). The
+    activity API exposes that row under type=a2a_receive, so the inbox
+    poller re-fetches it. Without this guard the row is surfaced as
+    kind='peer_agent' with the workspace's own identity as peer_id —
+    the workspace sees its own delegation-failure echoed back as if a
+    peer had delegated to it.
+
+    The guard mirrors the existing _is_self_notify_row pattern: both
+    skip rows that would otherwise create spurious inbound signal. The
+    long-term fix (making the platform write a distinct activity_type
+    for agent-outbound rows) is tracked separately; this guard stays
+    because it only excludes rows the agent never wants.
+
+    ``workspace_id`` must be non-empty — an empty-string workspace_id
+    (single-workspace legacy path) can never match a UUID source_id, so
+    the predicate is always False there, which is safe.
+
+    RFC #2829 PR-2 note: rows with method="delegate_result" are excluded
+    from the self-echo guard even when source_id matches our workspace_id.
+    The platform may write a delegation-result row with source_id set to
+    our workspace_id (e.g. a self-delegation or edge case in the platform's
+    result-writing path). Such rows must reach the inbox so that
+    message_from_activity can surface them as peer_agent inbound and the
+    runtime receives the delegation result. Silently filtering them as
+    self-echo would break delegation result delivery.
+    """
+    if not workspace_id:
+        return False
+    return row.get("source_id") == workspace_id and row.get("method") != "delegate_result"
+
+
 def message_from_activity(row: dict[str, Any]) -> InboxMessage:
     """Convert one /activity row into an InboxMessage.
 
@@ -623,6 +660,16 @@ def _poll_once(
             # the same self-notify on every iteration.
             last_id = str(row.get("id", "")) or last_id
             continue
+        if _is_self_echo_row(row, workspace_id):
+            # Internal #469: tool_delegate_task writes its own a2a_receive
+            # row with source_id = this workspace's UUID (spoof-defense).
+            # The poll fetches it back as kind='peer_agent', making the
+            # workspace echo its own delegation-failure as an inbound from
+            # a phantom peer. Skip it — the real delegation-result path
+            # (delegate_result push) is separate and unaffected. Cursor
+            # still advances so the next poll doesn't re-seen this row.
+            last_id = str(row.get("id", "")) or last_id
+            continue
         message = message_from_activity(row)
         if not message.activity_id:
             continue

workspace/tests/test_a2a_mcp_server.py

@@ -1826,8 +1826,8 @@ def test_inbox_bridge_swallows_closed_loop_runtime_error():
 
 
 class TestStdioPipeAssertion:
-    """Pin _warn_if_stdio_not_pipe — the diagnostic warning that replaces
-    the old fatal _assert_stdio_is_pipe_compatible guard.
+    """Pin _assert_stdio_is_pipe_compatible — the canonical function name.
+    _warn_if_stdio_not_pipe is a deprecated alias.
 
     The universal stdio transport now works with ANY file descriptor
     (pipes, regular files, PTYs, sockets), so the old exit-2 behavior
@@ -1838,12 +1838,12 @@ class TestStdioPipeAssertion:
 
     def test_pipe_pair_passes_silently(self, caplog):
         """Happy path — both fds are pipes. No warning emitted."""
-        from a2a_mcp_server import _warn_if_stdio_not_pipe
+        from a2a_mcp_server import _assert_stdio_is_pipe_compatible
 
         r, w = os.pipe()
         try:
             with caplog.at_level("WARNING"):
-                _warn_if_stdio_not_pipe(stdin_fd=r, stdout_fd=w)
+                _assert_stdio_is_pipe_compatible(stdin_fd=r, stdout_fd=w)
             assert "not a pipe" not in caplog.text
         finally:
             os.close(r)
@@ -1852,14 +1852,14 @@ class TestStdioPipeAssertion:
     def test_regular_file_stdout_warns(self, tmp_path, caplog):
         """Reproducer for runtime#61: stdout redirected to a regular file.
         Now emits a warning instead of exiting."""
-        from a2a_mcp_server import _warn_if_stdio_not_pipe
+        from a2a_mcp_server import _assert_stdio_is_pipe_compatible
 
         r, _w = os.pipe()
         regular = tmp_path / "captured.log"
         f = open(regular, "wb")
         try:
             with caplog.at_level("WARNING"):
-                _warn_if_stdio_not_pipe(stdin_fd=r, stdout_fd=f.fileno())
+                _assert_stdio_is_pipe_compatible(stdin_fd=r, stdout_fd=f.fileno())
             assert "stdout" in caplog.text
             assert "not a pipe" in caplog.text
         finally:
@@ -1868,7 +1868,7 @@ class TestStdioPipeAssertion:
 
     def test_regular_file_stdin_warns(self, tmp_path, caplog):
         """Symmetric case — stdin redirected from a regular file."""
-        from a2a_mcp_server import _warn_if_stdio_not_pipe
+        from a2a_mcp_server import _assert_stdio_is_pipe_compatible
 
         regular = tmp_path / "input.json"
         regular.write_bytes(b'{"jsonrpc":"2.0","id":1,"method":"initialize"}\n')
@@ -1876,7 +1876,7 @@ class TestStdioPipeAssertion:
         _r, w = os.pipe()
         try:
             with caplog.at_level("WARNING"):
-                _warn_if_stdio_not_pipe(stdin_fd=f.fileno(), stdout_fd=w)
+                _assert_stdio_is_pipe_compatible(stdin_fd=f.fileno(), stdout_fd=w)
             assert "stdin" in caplog.text
             assert "not a pipe" in caplog.text
         finally:
@@ -1886,13 +1886,13 @@ class TestStdioPipeAssertion:
     def test_closed_fd_warns_about_stat_error(self, caplog):
         """If stdio is closed, os.fstat raises OSError. Warning is
         skipped silently (can't stat the fd)."""
-        from a2a_mcp_server import _warn_if_stdio_not_pipe
+        from a2a_mcp_server import _assert_stdio_is_pipe_compatible
 
         r, w = os.pipe()
         os.close(w)  # Now `w` is a stale fd — fstat will fail.
         try:
             with caplog.at_level("WARNING"):
-                _warn_if_stdio_not_pipe(stdin_fd=r, stdout_fd=w)
+                _assert_stdio_is_pipe_compatible(stdin_fd=r, stdout_fd=w)
             # No warning emitted because fstat failed before the check
             assert "not a pipe" not in caplog.text
         finally:

workspace/tests/test_inbox.py

@@ -495,6 +495,151 @@ def test_poll_once_skips_self_notify_rows(state: inbox.InboxState):
     assert [m.activity_id for m in queue] == ["act-real"]
 
 
+# ---------------------------------------------------------------------------
+# _is_self_echo_row — internal #469 fix
+# ---------------------------------------------------------------------------
+#
+# When a workspace delegates to a target that never picks up the task,
+# tool_delegate_task calls report_activity("a2a_receive", ...) which POSTs
+# to the platform with source_id set to the *sender's* workspace UUID
+# (spoof-defense). The activity API returns that row under type=a2a_receive
+# on the next poll, so message_from_activity sets peer_id = workspace's own
+# UUID — the workspace sees its own delegation-failure as an inbound from
+# a phantom peer. _is_self_echo_row guards against this.
+#
+# Internal #469 was live-reproduced on hongming.moleculesai.app 2026-05-16.
+
+
+def test_is_self_echo_row_true_when_source_id_matches_workspace():
+    row = {"source_id": "ws-abc123", "method": "a2a_receive"}
+    assert inbox._is_self_echo_row(row, "ws-abc123") is True
+
+
+def test_is_self_echo_row_false_when_source_id_differs():
+    """A real peer agent (different workspace_id) must NOT be filtered."""
+    row = {"source_id": "ws-peer", "method": "a2a_receive"}
+    assert inbox._is_self_echo_row(row, "ws-1") is False
+
+
+def test_is_self_echo_row_false_when_source_id_is_none():
+    """Canvas-user inbound has no source_id — never an echo."""
+    row = {"source_id": None, "method": "a2a_receive"}
+    assert inbox._is_self_echo_row(row, "ws-1") is False
+
+
+def test_is_self_echo_row_false_when_workspace_id_is_empty():
+    """Single-workspace legacy path with empty workspace_id cannot
+    match a UUID source_id — predicate is always False, which is safe."""
+    row = {"source_id": "ws-abc123", "method": "a2a_receive"}
+    assert inbox._is_self_echo_row(row, "") is False
+
+
+def test_is_self_echo_row_false_when_source_id_key_absent():
+    row = {"method": "a2a_receive"}
+    assert inbox._is_self_echo_row(row, "ws-1") is False
+
+
+def test_is_self_echo_row_false_for_delegate_result():
+    """RFC #2829 PR-2 regression pin: a row with source_id matching our
+    workspace_id but method=delegate_result must NOT be filtered as a
+    self-echo. The platform may write a delegation-result row with our
+    workspace_id as source_id; such rows must reach the inbox so the
+    runtime receives the delegation result. Silently filtering them would
+    break delegate_result delivery."""
+    row = {"source_id": "ws-1", "method": "delegate_result"}
+    assert inbox._is_self_echo_row(row, "ws-1") is False
+
+
+def test_poll_once_skips_self_echo_rows(state: inbox.InboxState):
+    """Internal #469 regression pin: a row with source_id matching our
+    workspace_id must NOT land in the inbox queue — it is our own
+    delegation-report echoing back, not a real peer inbound."""
+    rows = [
+        {
+            "id": "act-real-peer",
+            "source_id": "ws-peer",
+            "method": "a2a_receive",
+            "summary": None,
+            "request_body": {"parts": [{"type": "text", "text": "real peer inbound"}]},
+            "created_at": "2026-04-30T22:00:00Z",
+        },
+        {
+            "id": "act-self-echo",
+            "source_id": "ws-1",
+            "method": "a2a_receive",
+            "summary": "task result: target timed out",
+            "request_body": None,
+            "created_at": "2026-04-30T22:00:01Z",
+        },
+    ]
+    resp = _make_response(200, rows)
+    p, _ = _patch_httpx(resp)
+    with p:
+        n = inbox._poll_once(state, "http://platform", "ws-1", {})
+
+    # Only the real peer inbound counted; self-echo silently dropped.
+    assert n == 1
+    queue = state.peek(10)
+    assert [m.activity_id for m in queue] == ["act-real-peer"]
+    assert queue[0].peer_id == "ws-peer"
+
+
+def test_poll_once_advances_cursor_past_self_echo(state: inbox.InboxState):
+    """Cursor must advance past self-echo rows even though we don't
+    enqueue them. Otherwise the next poll re-fetches the same self-echo
+    on every iteration, wasting requests and blocking real inbound."""
+    state.save_cursor("act-old")
+    rows = [
+        {
+            "id": "act-self-echo",
+            "source_id": "ws-1",
+            "method": "a2a_receive",
+            "summary": "task result: timeout",
+            "request_body": None,
+            "created_at": "2026-04-30T22:00:00Z",
+        },
+    ]
+    resp = _make_response(200, rows)
+    p, _ = _patch_httpx(resp)
+    with p:
+        n = inbox._poll_once(state, "http://platform", "ws-1", {})
+
+    assert n == 0
+    assert state.peek(10) == []
+    # Cursor must move past the skipped row so we don't re-poll it.
+    assert state.load_cursor() == "act-self-echo"
+
+
+def test_poll_once_self_echo_does_not_fire_notification(state: inbox.InboxState):
+    """The notification callback (channel push to Claude Code etc.)
+    must not fire for self-echo rows. Same rationale as self-notify:
+    push-capable hosts would see the echo loop on the push channel."""
+    rows = [
+        {
+            "id": "act-self-echo",
+            "source_id": "ws-1",
+            "method": "a2a_receive",
+            "summary": "task result: timeout",
+            "request_body": None,
+            "created_at": "2026-04-30T22:00:00Z",
+        },
+    ]
+    received: list[dict] = []
+    inbox.set_notification_callback(received.append)
+    try:
+        resp = _make_response(200, rows)
+        p, _ = _patch_httpx(resp)
+        with p:
+            inbox._poll_once(state, "http://platform", "ws-1", {})
+    finally:
+        inbox.set_notification_callback(None)
+
+    assert received == [], (
+        "self-echo rows must not surface as MCP notifications — "
+        "doing so re-creates the echo loop on push-capable hosts"
+    )
+
+
 def test_poll_once_advances_cursor_past_self_notify(state: inbox.InboxState):
     """Cursor must advance past self-notify rows even though we don't
     enqueue them. Otherwise the next poll re-fetches the same self-