fix(workspace): resolve PR #475 test failures

OFFSEC-003 (commit 2add6333) wrapped tool_delegate_task results in [A2A_RESULT_FROM_PEER] boundary markers via sanitize_a2a_result(), but test_a2a_tools_impl.py was not updated. Fix: 1. test_success_returns_result_text: assert now expects boundary-wrapped result — send_a2a_message returns plain "Task completed!" which sanitize_a2a_result wraps before returning. 2. test_peer_name_cached_from_peer_names_dict: same — "done" is now wrapped. 3. test_peer_name_falls_back_to_id_prefix: same — "ok" is now wrapped. 4. Remove TestDelegateTaskDirect class (3 dead tests for a2a_tools.delegate_task which does not exist in the codebase — added in commit 93b7d9a8 when the function existed, removed in the a2a_tools_delegation.py extraction refactor). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
fix(workspace): default PLATFORM_URL to host.docker.internal in all modules
2026-05-11 14:42:21 +00:00 · 2026-05-11 12:43:26 +00:00 · 2026-05-11 12:37:37 +00:00 · 2026-05-11 12:35:58 +00:00 · 2026-05-11 12:25:54 +00:00 · 2026-05-11 12:25:46 +00:00
34 changed files with 1668 additions and 344 deletions
@@ -0,0 +1,172 @@
+#!/usr/bin/env bash
+# sop-tier-refire — re-evaluate sop-tier-check and POST status to PR head SHA.
+#
+# Invoked from `.gitea/workflows/sop-tier-refire.yml` when a repo
+# MEMBER/OWNER/COLLABORATOR comments `/refire-tier-check` on a PR.
+#
+# Behavior:
+#
+# 1. Resolve PR head SHA + author from PR_NUMBER.
+# 2. Rate-limit: if the sop-tier-check context has been POSTed in the
+#    last 30 seconds, skip (prevents comment-spam status thrash).
+# 3. Invoke `.gitea/scripts/sop-tier-check.sh` with the same env the
+#    canonical workflow provides. This is DRY: we re-use the exact AND-
+#    composition gate logic, not a watered-down approving-count check.
+# 4. POST the resulting status (success on exit 0, failure on non-zero)
+#    to `/repos/.../statuses/{HEAD_SHA}` with context
+#    "sop-tier-check / tier-check (pull_request)" — the same context name
+#    branch protection requires.
+#
+# Required env (set by sop-tier-refire.yml):
+#   GITEA_TOKEN    — org-level SOP_TIER_CHECK_TOKEN (read:org/user/issue/repo)
+#   GITEA_HOST     — e.g. git.moleculesai.app
+#   REPO           — owner/name
+#   PR_NUMBER      — PR number from issue_comment payload
+#   COMMENT_AUTHOR — login of the commenter (logged for audit)
+#
+# Optional:
+#   SOP_DEBUG=1                — verbose per-API-call diagnostics
+#   SOP_REFIRE_RATE_LIMIT_SEC  — override the 30s rate-limit (default 30)
+#   SOP_REFIRE_DISABLE_RATE_LIMIT=1 — for tests; skips the rate-limit check
+
+set -euo pipefail
+
+debug() {
+  if [ "${SOP_DEBUG:-}" = "1" ]; then
+    echo "  [debug] $*" >&2
+  fi
+}
+
+: "${GITEA_TOKEN:?GITEA_TOKEN required}"
+: "${GITEA_HOST:?GITEA_HOST required}"
+: "${REPO:?REPO required (owner/name)}"
+: "${PR_NUMBER:?PR_NUMBER required}"
+: "${COMMENT_AUTHOR:=unknown}"
+
+OWNER="${REPO%%/*}"
+NAME="${REPO##*/}"
+API="https://${GITEA_HOST}/api/v1"
+AUTH="Authorization: token ${GITEA_TOKEN}"
+CONTEXT="sop-tier-check / tier-check (pull_request)"
+RATE_LIMIT_SEC="${SOP_REFIRE_RATE_LIMIT_SEC:-30}"
+
+echo "::notice::sop-tier-refire start: repo=$OWNER/$NAME pr=$PR_NUMBER commenter=$COMMENT_AUTHOR"
+
+# 1. Fetch PR details — need head.sha and user.login.
+PR_FILE=$(mktemp)
+trap 'rm -f "$PR_FILE"' EXIT
+PR_HTTP=$(curl -sS -o "$PR_FILE" -w '%{http_code}' -H "$AUTH" \
+  "${API}/repos/${OWNER}/${NAME}/pulls/${PR_NUMBER}")
+if [ "$PR_HTTP" != "200" ]; then
+  echo "::error::GET /pulls/$PR_NUMBER returned HTTP $PR_HTTP (body $(head -c 200 "$PR_FILE"))"
+  exit 1
+fi
+HEAD_SHA=$(jq -r '.head.sha' <"$PR_FILE")
+PR_AUTHOR=$(jq -r '.user.login' <"$PR_FILE")
+PR_STATE=$(jq -r '.state' <"$PR_FILE")
+if [ -z "$HEAD_SHA" ] || [ "$HEAD_SHA" = "null" ]; then
+  echo "::error::Could not resolve head.sha from PR #$PR_NUMBER response"
+  exit 1
+fi
+debug "head_sha=$HEAD_SHA pr_author=$PR_AUTHOR state=$PR_STATE"
+
+if [ "$PR_STATE" != "open" ]; then
+  echo "::notice::PR #$PR_NUMBER state is $PR_STATE; refire is a no-op on closed PRs."
+  exit 0
+fi
+
+# 2. Rate-limit: skip if our context was updated in the last $RATE_LIMIT_SEC.
+# Gitea statuses endpoint returns latest first; we check the most recent
+# entry for our context name.
+if [ "${SOP_REFIRE_DISABLE_RATE_LIMIT:-}" != "1" ]; then
+  STATUSES_FILE=$(mktemp)
+  trap 'rm -f "$PR_FILE" "$STATUSES_FILE"' EXIT
+  ST_HTTP=$(curl -sS -o "$STATUSES_FILE" -w '%{http_code}' -H "$AUTH" \
+    "${API}/repos/${OWNER}/${NAME}/statuses/${HEAD_SHA}?limit=50&sort=newest")
+  debug "statuses-list HTTP=$ST_HTTP"
+  if [ "$ST_HTTP" = "200" ]; then
+    LAST_UPDATED=$(jq -r --arg c "$CONTEXT" \
+      '[.[] | select(.context == $c)] | first | .updated_at // ""' \
+      <"$STATUSES_FILE")
+    if [ -n "$LAST_UPDATED" ] && [ "$LAST_UPDATED" != "null" ]; then
+      # Parse RFC3339 → epoch. Use python -c for portability (date(1) -d
+      # differs between BSD/GNU; the Gitea runner is Ubuntu so GNU date
+      # works, but we keep python for future container variance).
+      LAST_EPOCH=$(python3 -c "import sys,datetime;print(int(datetime.datetime.fromisoformat(sys.argv[1].replace('Z','+00:00')).timestamp()))" "$LAST_UPDATED" 2>/dev/null || echo "0")
+      NOW_EPOCH=$(date -u +%s)
+      AGE=$((NOW_EPOCH - LAST_EPOCH))
+      debug "last status update: $LAST_UPDATED ($AGE seconds ago)"
+      if [ "$AGE" -lt "$RATE_LIMIT_SEC" ] && [ "$AGE" -ge 0 ]; then
+        echo "::notice::sop-tier-refire rate-limited — last status update was ${AGE}s ago (<${RATE_LIMIT_SEC}s window). Try again shortly."
+        exit 0
+      fi
+    fi
+  fi
+fi
+
+# 3. Invoke sop-tier-check.sh with the env it expects. Capture exit code.
+# The canonical script reads tier label, walks approving reviewers, and
+# evaluates the AND-composition expression — we want the SAME gate, not
+# a different gate.
+#
+# SOP_REFIRE_TIER_CHECK_SCRIPT env var lets tests substitute a mock —
+# sop-tier-check.sh uses bash 4+ associative arrays which trigger a known
+# bash 3.2 parser bug (`tier: unbound variable` from declare -A with
+# `set -u`). Linux Gitea runners ship bash 4/5 so production is fine;
+# the override exists so the bash 3.2 dev box can still exercise the
+# refire glue logic end-to-end.
+SCRIPT="${SOP_REFIRE_TIER_CHECK_SCRIPT:-$(dirname "$0")/sop-tier-check.sh}"
+if [ ! -f "$SCRIPT" ]; then
+  echo "::error::sop-tier-check.sh not found at $SCRIPT — refire requires the canonical script"
+  exit 1
+fi
+
+# Re-invoke. Pipe stdout/stderr through so the runner log shows the
+# tier-check decision inline.
+set +e
+GITEA_TOKEN="$GITEA_TOKEN" \
+  GITEA_HOST="$GITEA_HOST" \
+  REPO="$REPO" \
+  PR_NUMBER="$PR_NUMBER" \
+  PR_AUTHOR="$PR_AUTHOR" \
+  SOP_DEBUG="${SOP_DEBUG:-0}" \
+  SOP_LEGACY_CHECK="${SOP_LEGACY_CHECK:-0}" \
+  bash "$SCRIPT"
+TIER_EXIT=$?
+set -e
+debug "sop-tier-check.sh exit=$TIER_EXIT"
+
+# 4. POST the resulting status.
+if [ "$TIER_EXIT" -eq 0 ]; then
+  STATE="success"
+  DESCRIPTION="Refired via /refire-tier-check by $COMMENT_AUTHOR"
+else
+  STATE="failure"
+  DESCRIPTION="Refired via /refire-tier-check; tier-check failed (see workflow log)"
+fi
+
+# Status target_url points at the runner log so a curious reviewer can
+# follow it back. SERVER_URL + RUN_ID + JOB_ID isn't trivially constructible
+# from the bash env on Gitea 1.22.6, so we point at the PR itself.
+TARGET_URL="https://${GITEA_HOST}/${OWNER}/${NAME}/pulls/${PR_NUMBER}"
+
+POST_BODY=$(jq -nc \
+  --arg state "$STATE" \
+  --arg context "$CONTEXT" \
+  --arg description "$DESCRIPTION" \
+  --arg target_url "$TARGET_URL" \
+  '{state:$state, context:$context, description:$description, target_url:$target_url}')
+
+POST_FILE=$(mktemp)
+trap 'rm -f "$PR_FILE" "${STATUSES_FILE:-}" "$POST_FILE"' EXIT
+POST_HTTP=$(curl -sS -o "$POST_FILE" -w '%{http_code}' \
+  -X POST -H "$AUTH" -H "Content-Type: application/json" \
+  -d "$POST_BODY" \
+  "${API}/repos/${OWNER}/${NAME}/statuses/${HEAD_SHA}")
+if [ "$POST_HTTP" != "200" ] && [ "$POST_HTTP" != "201" ]; then
+  echo "::error::POST /statuses/$HEAD_SHA returned HTTP $POST_HTTP (body $(head -c 200 "$POST_FILE"))"
+  exit 1
+fi
+
+echo "::notice::sop-tier-refire posted state=$STATE for context=\"$CONTEXT\" on sha=$HEAD_SHA"
+exit "$TIER_EXIT"
@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Mock sop-tier-check.sh for sop-tier-refire tests.
+#
+# Exits 0 ("PASS") if $MOCK_TIER_RESULT == "pass", else exits 1.
+# This lets the refire tests cover the success + failure status-POST
+# paths without invoking the real sop-tier-check.sh (which uses bash 4+
+# associative arrays — known parser bug on macOS bash 3.2 dev box).
+
+set -euo pipefail
+
+case "${MOCK_TIER_RESULT:-pass}" in
+  pass)
+    echo "::notice::mock tier-check: PASS"
+    exit 0
+    ;;
+  fail_no_label)
+    echo "::error::mock tier-check: no tier label"
+    exit 1
+    ;;
+  fail_no_approvals)
+    echo "::error::mock tier-check: no approving reviews"
+    exit 1
+    ;;
+  *)
+    echo "::error::mock tier-check: unknown MOCK_TIER_RESULT=${MOCK_TIER_RESULT:-}"
+    exit 2
+    ;;
+esac
@@ -0,0 +1,208 @@
+#!/usr/bin/env python3
+"""Stub Gitea API for sop-tier-refire test scenarios.
+
+Reads $FIXTURE_STATE_DIR/scenario to decide what to return for each
+endpoint the sop-tier-refire.sh + sop-tier-check.sh scripts call.
+Captures every POST to /statuses/{sha} into posted_statuses.jsonl so
+the test can assert what the script tried to write.
+
+Scenarios:
+  T1_success         — tier:low + APPROVED by engineer → tier-check passes
+  T2_no_tier_label   — no tier label → tier-check exits 1 before POST
+  T3_no_approvals    — tier:low but zero approving reviews → exits 1
+  T4_closed          — PR state=closed → refire is a no-op
+  T5_rate_limited    — last status update 5 seconds ago → skip
+
+Usage:
+  FIXTURE_STATE_DIR=/tmp/x python3 _refire_fixture.py 8080
+"""
+
+import datetime
+import http.server
+import json
+import os
+import re
+import sys
+import urllib.parse
+
+
+STATE_DIR = os.environ["FIXTURE_STATE_DIR"]
+
+
+def scenario() -> str:
+    p = os.path.join(STATE_DIR, "scenario")
+    if not os.path.isfile(p):
+        return "T1_success"
+    with open(p) as f:
+        return f.read().strip()
+
+
+def now_iso() -> str:
+    return datetime.datetime.now(datetime.timezone.utc).isoformat()
+
+
+def append_post(body: dict) -> None:
+    with open(os.path.join(STATE_DIR, "posted_statuses.jsonl"), "a") as f:
+        f.write(json.dumps(body) + "\n")
+
+
+def pr_payload() -> dict:
+    sc = scenario()
+    state = "closed" if sc == "T4_closed" else "open"
+    return {
+        "number": 999,
+        "state": state,
+        "head": {"sha": "deadbeef0000111122223333444455556666"},
+        "user": {"login": "feature-author"},
+    }
+
+
+def labels_payload() -> list:
+    sc = scenario()
+    if sc == "T2_no_tier_label":
+        return [{"name": "bug"}]
+    # All other scenarios use tier:low
+    return [{"name": "tier:low"}, {"name": "ci"}]
+
+
+def reviews_payload() -> list:
+    sc = scenario()
+    if sc == "T3_no_approvals":
+        return []
+    # All other scenarios have one APPROVED review by an engineer
+    return [
+        {
+            "state": "APPROVED",
+            "user": {"login": "reviewer-engineer"},
+        }
+    ]
+
+
+def teams_payload() -> list:
+    # Mirror the real molecule-ai org teams referenced in TIER_EXPR
+    return [
+        {"id": 5, "name": "ceo"},
+        {"id": 2, "name": "engineers"},
+        {"id": 6, "name": "managers"},
+    ]
+
+
+def statuses_payload() -> list:
+    sc = scenario()
+    if sc == "T5_rate_limited":
+        recent = (
+            datetime.datetime.now(datetime.timezone.utc)
+            - datetime.timedelta(seconds=5)
+        ).isoformat()
+        return [
+            {
+                "context": "sop-tier-check / tier-check (pull_request)",
+                "state": "failure",
+                "updated_at": recent,
+            }
+        ]
+    return []
+
+
+def user_payload() -> dict:
+    # Mirrors the WHOAMI probe in sop-tier-check.sh
+    return {"login": "sop-tier-bot-fixture"}
+
+
+class Handler(http.server.BaseHTTPRequestHandler):
+    # Quiet — keep stdout for explicit logs only.
+    def log_message(self, *args, **kwargs):  # noqa: D401
+        pass
+
+    def _json(self, code: int, body) -> None:
+        payload = json.dumps(body).encode()
+        self.send_response(code)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(payload)))
+        self.end_headers()
+        self.wfile.write(payload)
+
+    def _empty(self, code: int) -> None:
+        self.send_response(code)
+        self.send_header("Content-Length", "0")
+        self.end_headers()
+
+    def do_GET(self):  # noqa: N802
+        u = urllib.parse.urlparse(self.path)
+        path = u.path
+
+        if path == "/_ping":
+            return self._json(200, {"ok": True})
+        if path == "/api/v1/user":
+            return self._json(200, user_payload())
+
+        # /api/v1/repos/{owner}/{name}/pulls/{n}
+        m = re.match(r"^/api/v1/repos/[^/]+/[^/]+/pulls/(\d+)$", path)
+        if m:
+            return self._json(200, pr_payload())
+
+        # /api/v1/repos/{owner}/{name}/issues/{n}/labels
+        if re.match(r"^/api/v1/repos/[^/]+/[^/]+/issues/\d+/labels$", path):
+            return self._json(200, labels_payload())
+
+        # /api/v1/repos/{owner}/{name}/pulls/{n}/reviews
+        if re.match(r"^/api/v1/repos/[^/]+/[^/]+/pulls/\d+/reviews$", path):
+            return self._json(200, reviews_payload())
+
+        # /api/v1/orgs/{owner}/teams
+        if re.match(r"^/api/v1/orgs/[^/]+/teams$", path):
+            return self._json(200, teams_payload())
+
+        # /api/v1/teams/{id}/members/{login} → 204 if user is an engineer
+        m = re.match(r"^/api/v1/teams/(\d+)/members/([^/]+)$", path)
+        if m:
+            team_id, login = m.group(1), m.group(2)
+            # In our fixture reviewer-engineer ∈ engineers (id=2)
+            if team_id == "2" and login == "reviewer-engineer":
+                return self._empty(204)
+            return self._empty(404)
+
+        # /api/v1/orgs/{owner}/members/{login} — fallback path used when
+        # team-member probes all 403. We don't need it for these tests.
+        if re.match(r"^/api/v1/orgs/[^/]+/members/[^/]+$", path):
+            return self._empty(404)
+
+        # /api/v1/repos/{owner}/{name}/statuses/{sha}
+        if re.match(r"^/api/v1/repos/[^/]+/[^/]+/statuses/[^/]+$", path):
+            return self._json(200, statuses_payload())
+
+        return self._json(404, {"path": path, "msg": "fixture: no route"})
+
+    def do_POST(self):  # noqa: N802
+        u = urllib.parse.urlparse(self.path)
+        path = u.path
+        length = int(self.headers.get("Content-Length") or 0)
+        raw = self.rfile.read(length) if length else b""
+        try:
+            body = json.loads(raw) if raw else {}
+        except Exception:
+            body = {"_raw": raw.decode(errors="replace")}
+
+        if re.match(r"^/api/v1/repos/[^/]+/[^/]+/statuses/[^/]+$", path):
+            append_post(body)
+            # Echo back something status-shaped — script only checks HTTP code.
+            return self._json(
+                201,
+                {
+                    "context": body.get("context"),
+                    "state": body.get("state"),
+                    "created_at": now_iso(),
+                },
+            )
+
+        return self._json(404, {"path": path, "msg": "fixture: no route"})
+
+
+def main():
+    port = int(sys.argv[1])
+    srv = http.server.ThreadingHTTPServer(("127.0.0.1", port), Handler)
+    srv.serve_forever()
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,297 @@
+#!/usr/bin/env bash
+# Tests for sop-tier-refire.{yml,sh} — internal#292.
+#
+# Behavior matrix:
+#
+#   T1: PR open + APPROVED via tier:low → script invokes sop-tier-check
+#       and POSTs status=success.
+#   T2: PR open + missing tier label → sop-tier-check exits non-zero;
+#       refire POSTs status=failure (description mentions failure).
+#   T3: PR open + tier:low but NO approving reviews → sop-tier-check
+#       exits non-zero; refire POSTs status=failure.
+#   T4: PR CLOSED → refire exits 0 with no status POST (no-op on closed).
+#   T5: Rate-limit — recent status update within 30s → refire skips,
+#       no new POST.
+#   T6 (yaml-lint): workflow `if:` expression contains author_association
+#       gate + slash-command-trigger gate + PR-not-issue gate.
+#   T7 (yaml-lint): workflow file is parseable YAML.
+#
+# Tests T1-T5 run the real script against a local-fixture HTTP server
+# (python http.server with a stub handler — `tests/_refire_fixture.py`)
+# so the script's Gitea API calls hit the fixture, not the real Gitea.
+#
+# Tests T6/T7 are pure YAML checks against the workflow file.
+#
+# Hostile-self-review (per feedback_assert_exact_not_substring):
+# this test MUST FAIL if the workflow or script is absent. Verified by
+# running the test before the files exist (covered in the PR body).
+
+set -euo pipefail
+
+THIS_DIR="$(cd "$(dirname "$0")" && pwd)"
+SCRIPT_DIR="$(cd "$THIS_DIR/.." && pwd)"
+WORKFLOW_DIR="$(cd "$THIS_DIR/../../workflows" && pwd)"
+WORKFLOW="$WORKFLOW_DIR/sop-tier-refire.yml"
+SCRIPT="$SCRIPT_DIR/sop-tier-refire.sh"
+
+PASS=0
+FAIL=0
+FAILED_TESTS=""
+
+assert_eq() {
+  local label="$1"
+  local expected="$2"
+  local got="$3"
+  if [ "$expected" = "$got" ]; then
+    echo "  PASS  $label"
+    PASS=$((PASS + 1))
+  else
+    echo "  FAIL  $label"
+    echo "        expected: <$expected>"
+    echo "        got:      <$got>"
+    FAIL=$((FAIL + 1))
+    FAILED_TESTS="${FAILED_TESTS} ${label}"
+  fi
+}
+
+assert_contains() {
+  local label="$1"
+  local needle="$2"
+  local haystack="$3"
+  if printf '%s' "$haystack" | grep -qF "$needle"; then
+    echo "  PASS  $label"
+    PASS=$((PASS + 1))
+  else
+    echo "  FAIL  $label"
+    echo "        needle:    <$needle>"
+    echo "        haystack:  <$(printf '%s' "$haystack" | head -c 400)>"
+    FAIL=$((FAIL + 1))
+    FAILED_TESTS="${FAILED_TESTS} ${label}"
+  fi
+}
+
+assert_file_exists() {
+  local label="$1"
+  local path="$2"
+  if [ -f "$path" ]; then
+    echo "  PASS  $label"
+    PASS=$((PASS + 1))
+  else
+    echo "  FAIL  $label (not found: $path)"
+    FAIL=$((FAIL + 1))
+    FAILED_TESTS="${FAILED_TESTS} ${label}"
+  fi
+}
+
+# Existence (foundation — every other test depends on these)
+echo
+echo "== existence =="
+assert_file_exists "workflow file exists"  "$WORKFLOW"
+assert_file_exists "script file exists"    "$SCRIPT"
+if [ "$FAIL" -gt 0 ]; then
+  echo
+  echo "------"
+  echo "PASS=$PASS FAIL=$FAIL (existence)"
+  echo "Cannot proceed without these files."
+  exit 1
+fi
+
+# T6 / T7 — workflow YAML structure
+echo
+echo "== T6/T7 workflow yaml =="
+
+# YAML parseability
+PARSE_OUT=$(python3 -c 'import sys,yaml;yaml.safe_load(open(sys.argv[1]).read());print("ok")' "$WORKFLOW" 2>&1 || true)
+assert_eq "T7 workflow parses as YAML" "ok" "$PARSE_OUT"
+
+# Three required gates in the `if:` expression
+WORKFLOW_CONTENT=$(cat "$WORKFLOW")
+assert_contains "T6a workflow if: contains author_association gate" \
+  "github.event.comment.author_association" "$WORKFLOW_CONTENT"
+assert_contains "T6b workflow if: gates on MEMBER/OWNER/COLLABORATOR" \
+  '["MEMBER","OWNER","COLLABORATOR"]' "$WORKFLOW_CONTENT"
+assert_contains "T6c workflow if: contains slash-command trigger" \
+  "/refire-tier-check" "$WORKFLOW_CONTENT"
+assert_contains "T6d workflow if: gates on PR-not-issue" \
+  "github.event.issue.pull_request" "$WORKFLOW_CONTENT"
+assert_contains "T6e workflow listens on issue_comment" \
+  "issue_comment" "$WORKFLOW_CONTENT"
+assert_contains "T6f workflow requests statuses:write permission" \
+  "statuses: write" "$WORKFLOW_CONTENT"
+# Does NOT check out PR HEAD (security)
+if grep -q 'ref: \${{ github.event.pull_request.head' "$WORKFLOW"; then
+  echo "  FAIL  T6g workflow MUST NOT check out PR head (security)"
+  FAIL=$((FAIL + 1))
+  FAILED_TESTS="${FAILED_TESTS} T6g"
+else
+  echo "  PASS  T6g workflow does not check out PR head"
+  PASS=$((PASS + 1))
+fi
+
+# T1-T5 — script behavior against a local Gitea-fixture
+echo
+echo "== T1-T5 script behavior (vs local fixture) =="
+
+# Spin up the fixture HTTP server.
+FIXTURE_DIR=$(mktemp -d)
+trap 'rm -rf "$FIXTURE_DIR"; [ -n "${FIX_PID:-}" ] && kill "$FIX_PID" 2>/dev/null || true' EXIT
+FIXTURE_PY="$THIS_DIR/_refire_fixture.py"
+if [ ! -f "$FIXTURE_PY" ]; then
+  echo "::error::fixture server $FIXTURE_PY missing"
+  exit 1
+fi
+
+FIX_LOG="$FIXTURE_DIR/fixture.log"
+FIX_STATE_DIR="$FIXTURE_DIR/state"
+mkdir -p "$FIX_STATE_DIR"
+
+# Find an unused port.
+FIX_PORT=$(python3 -c 'import socket;s=socket.socket();s.bind(("127.0.0.1",0));print(s.getsockname()[1]);s.close()')
+
+FIXTURE_STATE_DIR="$FIX_STATE_DIR" python3 "$FIXTURE_PY" "$FIX_PORT" \
+  >"$FIX_LOG" 2>&1 &
+FIX_PID=$!
+
+# Wait for fixture readiness.
+for _ in $(seq 1 50); do
+  if curl -fsS "http://127.0.0.1:${FIX_PORT}/_ping" >/dev/null 2>&1; then
+    break
+  fi
+  sleep 0.1
+done
+if ! curl -fsS "http://127.0.0.1:${FIX_PORT}/_ping" >/dev/null 2>&1; then
+  echo "::error::fixture server failed to start. Log:"
+  cat "$FIX_LOG"
+  exit 1
+fi
+
+# Helper: set fixture state for a scenario, then run the script.
+# tier_result is one of: pass | fail_no_label | fail_no_approvals.
+# The refire script's tier-check invocation is mocked because the real
+# sop-tier-check.sh uses bash 4+ associative arrays — incompatible with
+# the macOS bash 3.2 dev shell. Linux Gitea runners use bash 4/5 so
+# production runs the real script. The mock exercises the success +
+# failure branches of refire's status-POST glue.
+run_scenario() {
+  local scenario="$1"
+  local tier_result="${2:-pass}"
+  echo "$scenario" >"$FIX_STATE_DIR/scenario"
+  : >"$FIX_STATE_DIR/posted_statuses.jsonl"  # clear status log
+
+  local out
+  set +e
+  out=$(
+    PATH="$FIXTURE_DIR/bin:$PATH" \
+    GITEA_TOKEN="fixture-token" \
+    GITEA_HOST="fixture.local" \
+    REPO="molecule-ai/molecule-core" \
+    PR_NUMBER="999" \
+    COMMENT_AUTHOR="test-runner" \
+    SOP_REFIRE_DISABLE_RATE_LIMIT="1" \
+    SOP_REFIRE_TIER_CHECK_SCRIPT="$THIS_DIR/_mock_tier_check.sh" \
+    MOCK_TIER_RESULT="$tier_result" \
+    FIXTURE_PORT="$FIX_PORT" \
+    bash "$SCRIPT" 2>&1
+  )
+  local rc=$?
+  set -e
+  echo "$out" >"$FIX_STATE_DIR/last_run.log"
+  echo "$rc" >"$FIX_STATE_DIR/last_rc"
+}
+
+# Install a curl shim that rewrites https://fixture.local → http://127.0.0.1:$PORT
+# Use bash prefix-strip (${var#prefix}) — it sidesteps the `/` delimiter
+# confusion of ${var/pattern/replacement}.
+mkdir -p "$FIXTURE_DIR/bin"
+cat >"$FIXTURE_DIR/bin/curl" <<SHIM
+#!/usr/bin/env bash
+# Test shim: rewrite https://fixture.local/* -> http://127.0.0.1:${FIX_PORT}/*
+# The fixture doesn't authenticate; -H Authorization passes through harmlessly.
+new_args=()
+for a in "\$@"; do
+  if [[ "\$a" == https://fixture.local/* ]]; then
+    rest="\${a#https://fixture.local}"
+    a="http://127.0.0.1:${FIX_PORT}\${rest}"
+  fi
+  new_args+=("\$a")
+done
+exec /usr/bin/curl "\${new_args[@]}"
+SHIM
+chmod +x "$FIXTURE_DIR/bin/curl"
+
+# T1: tier:low + 1 APPROVED + author is in engineers team → success
+run_scenario "T1_success" "pass"
+RC=$(cat "$FIX_STATE_DIR/last_rc")
+POSTED=$(cat "$FIX_STATE_DIR/posted_statuses.jsonl" 2>/dev/null || true)
+assert_eq "T1 exit code 0 (success)" "0" "$RC"
+assert_contains "T1 POSTed state=success" '"state": "success"' "$POSTED"
+assert_contains "T1 POST context is sop-tier-check / tier-check" \
+  '"context": "sop-tier-check / tier-check (pull_request)"' "$POSTED"
+assert_contains "T1 description names commenter" "test-runner" "$POSTED"
+
+# T2: missing tier label → tier-check fails → failure status POSTed
+run_scenario "T2_no_tier_label" "fail_no_label"
+RC=$(cat "$FIX_STATE_DIR/last_rc")
+POSTED=$(cat "$FIX_STATE_DIR/posted_statuses.jsonl" 2>/dev/null || true)
+# tier-check.sh exits 1; refire script forwards that exit, so RC != 0
+if [ "$RC" -ne 0 ]; then
+  echo "  PASS  T2 exit code non-zero (got $RC)"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL  T2 exit code should be non-zero, got 0"
+  FAIL=$((FAIL + 1))
+  FAILED_TESTS="${FAILED_TESTS} T2_rc"
+fi
+assert_contains "T2 POSTed state=failure" '"state": "failure"' "$POSTED"
+
+# T3: tier:low present but ZERO approving reviews → failure
+run_scenario "T3_no_approvals" "fail_no_approvals"
+RC=$(cat "$FIX_STATE_DIR/last_rc")
+POSTED=$(cat "$FIX_STATE_DIR/posted_statuses.jsonl" 2>/dev/null || true)
+if [ "$RC" -ne 0 ]; then
+  echo "  PASS  T3 exit code non-zero (got $RC)"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL  T3 exit code should be non-zero, got 0"
+  FAIL=$((FAIL + 1))
+  FAILED_TESTS="${FAILED_TESTS} T3_rc"
+fi
+assert_contains "T3 POSTed state=failure" '"state": "failure"' "$POSTED"
+
+# T4: closed PR — refire is a no-op (no POST, exit 0)
+run_scenario "T4_closed" "pass"
+RC=$(cat "$FIX_STATE_DIR/last_rc")
+POSTED=$(cat "$FIX_STATE_DIR/posted_statuses.jsonl" 2>/dev/null || true)
+assert_eq "T4 closed PR exits 0" "0" "$RC"
+assert_eq "T4 closed PR posts no status" "" "$POSTED"
+
+# T5: rate-limit — disable the env override and let scenario set a
+# recent statuses entry. Re-enable rate-limit for this scenario by NOT
+# passing SOP_REFIRE_DISABLE_RATE_LIMIT.
+echo "T5_rate_limited" >"$FIX_STATE_DIR/scenario"
+: >"$FIX_STATE_DIR/posted_statuses.jsonl"
+set +e
+T5_OUT=$(
+  PATH="$FIXTURE_DIR/bin:$PATH" \
+  GITEA_TOKEN="fixture-token" \
+  GITEA_HOST="fixture.local" \
+  REPO="molecule-ai/molecule-core" \
+  PR_NUMBER="999" \
+  COMMENT_AUTHOR="test-runner" \
+  FIXTURE_PORT="$FIX_PORT" \
+  bash "$SCRIPT" 2>&1
+)
+T5_RC=$?
+set -e
+POSTED=$(cat "$FIX_STATE_DIR/posted_statuses.jsonl" 2>/dev/null || true)
+assert_eq "T5 rate-limited exits 0" "0" "$T5_RC"
+assert_contains "T5 rate-limited log says skipped" "rate-limited" "$T5_OUT"
+assert_eq "T5 rate-limited posts no status" "" "$POSTED"
+
+echo
+echo "------"
+echo "PASS=$PASS FAIL=$FAIL"
+if [ "$FAIL" -gt 0 ]; then
+  echo "Failed:$FAILED_TESTS"
+fi
+[ "$FAIL" -eq 0 ]
@@ -56,7 +56,7 @@ on:
    #   2. Avoid colliding with the existing :15 sweep-cf-orphans
    #      and :45 sweep-cf-tunnels — both hit the CF API and we
    #      don't want to fight for rate-limit tokens.
-    #   3. Avoid the :30 heavy slot (canary-staging /30, sweep-aws-
+    #   3. Avoid the :30 heavy slot (staging-smoke /30, sweep-aws-
    #      secrets, sweep-stale-e2e-orgs every :15) — multiple
    #      overlapping cron registrations on the same minute is part
    #      of what GH drops under load.
@@ -124,7 +124,10 @@ jobs:
    env:
      CANVAS_E2E_STAGING: '1'
      MOLECULE_CP_URL: https://staging-api.moleculesai.app
-      MOLECULE_ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+      # 2026-05-11: secret canonicalised from MOLECULE_STAGING_ADMIN_TOKEN
+      # (dead in org secret store) to CP_STAGING_ADMIN_API_TOKEN per
+      # internal#322 — see this PR for the cross-workflow sweep.
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}

    defaults:
      run:
@@ -145,7 +148,7 @@ jobs:
        if: needs.detect-changes.outputs.canvas == 'true'
        run: |
          if [ -z "$MOLECULE_ADMIN_TOKEN" ]; then
-            echo "::error::Missing MOLECULE_STAGING_ADMIN_TOKEN"
+            echo "::error::Missing CP_STAGING_ADMIN_API_TOKEN"
            exit 2
          fi

@@ -207,7 +210,7 @@ jobs:
      - name: Teardown safety net
        if: always() && needs.detect-changes.outputs.canvas == 'true'
        env:
-          ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
        run: |
          set +e
          STATE_FILE=".playwright-staging-state.json"
@@ -89,7 +89,10 @@ jobs:

    env:
      MOLECULE_CP_URL: https://staging-api.moleculesai.app
-      MOLECULE_ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+      # 2026-05-11: secret canonicalised from MOLECULE_STAGING_ADMIN_TOKEN
+      # (dead in org secret store) to CP_STAGING_ADMIN_API_TOKEN per
+      # internal#322 — see this PR for the cross-workflow sweep.
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
      E2E_RUN_ID: "${{ github.run_id }}-${{ github.run_attempt }}"
      E2E_KEEP_ORG: ${{ github.event.inputs.keep_org && '1' || '0' }}
      E2E_STALE_WAIT_SECS: ${{ github.event.inputs.stale_wait_secs || '180' }}
@@ -104,7 +107,7 @@ jobs:
            # missing — silent skip would mask infra rot. Manual dispatch
            # gets the same hard-fail; an operator running this on a fork
            # without secrets configured needs to know up-front.
-            echo "::error::MOLECULE_STAGING_ADMIN_TOKEN secret not set (Railway staging CP_ADMIN_API_TOKEN)"
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN secret not set (Railway staging CP_ADMIN_API_TOKEN)"
            exit 2
          fi
          echo "Admin token present ✓"
@@ -129,7 +132,7 @@ jobs:
      - name: Teardown safety net (runs on cancel/failure)
        if: always()
        env:
-          ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
        run: |
          set +e
          orgs=$(curl -sS "$MOLECULE_CP_URL/cp/admin/orgs" \
@@ -86,7 +86,10 @@ jobs:
      # Single admin-bearer secret drives provision + tenant-token
      # retrieval + teardown. Configure in
      # Settings → Secrets and variables → Actions → Repository secrets.
-      MOLECULE_ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+      # 2026-05-11: secret canonicalised from MOLECULE_STAGING_ADMIN_TOKEN
+      # (dead in org secret store) to CP_STAGING_ADMIN_API_TOKEN per
+      # internal#322 — see this PR for the cross-workflow sweep.
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
      # MiniMax is the PRIMARY LLM auth path post-2026-05-04. Switched
      # from hermes+OpenAI default after #2578 (the staging OpenAI key
      # account went over quota and stayed dead for 36+ hours, taking
@@ -95,7 +98,7 @@ jobs:
      # ANTHROPIC_BASE_URL to api.minimax.io/anthropic and reads
      # MINIMAX_API_KEY at boot — separate billing account so an
      # OpenAI quota collapse no longer wedges the gate. Mirrors the
-      # canary-staging.yml + continuous-synth-e2e.yml migrations.
+      # staging-smoke.yml + continuous-synth-e2e.yml migrations.
      E2E_MINIMAX_API_KEY: ${{ secrets.MOLECULE_STAGING_MINIMAX_API_KEY }}
      # Direct-Anthropic alternative for operators who don't want to
      # set up a MiniMax account (priority below MiniMax — first
@@ -122,7 +125,7 @@ jobs:
      - name: Verify admin token present
        run: |
          if [ -z "$MOLECULE_ADMIN_TOKEN" ]; then
-            echo "::error::MOLECULE_STAGING_ADMIN_TOKEN secret not set (Railway staging CP_ADMIN_API_TOKEN)"
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN secret not set (Railway staging CP_ADMIN_API_TOKEN)"
            exit 2
          fi
          echo "Admin token present ✓"
@@ -189,7 +192,7 @@ jobs:
      - name: Teardown safety net (runs on cancel/failure)
        if: always()
        env:
-          ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
        run: |
          # Best-effort: find any e2e-YYYYMMDD-* orgs matching this run and
          # nuke them. Catches the case where the script died before
@@ -11,11 +11,11 @@ name: E2E Staging Sanity (leak-detection self-check)
 #   - `continue-on-error: true` on the job (RFC §1 contract).
 #
 # Periodic assertion that the teardown safety nets in e2e-staging-saas
-# and canary-staging actually work. Runs the E2E harness with
-# E2E_INTENTIONAL_FAILURE=1, which poisons the tenant admin token after
-# the org is provisioned. The workspace-provision step then fails, the
-# script exits non-zero, and the EXIT trap + workflow always()-step
-# must still tear down cleanly.
+# and staging-smoke (formerly canary-staging) actually work. Runs the
+# E2E harness with E2E_INTENTIONAL_FAILURE=1, which poisons the tenant
+# admin token after the org is provisioned. The workspace-provision
+# step then fails, the script exits non-zero, and the EXIT trap +
+# workflow always()-step must still tear down cleanly.

 on:
  schedule:
@@ -42,8 +42,11 @@ jobs:

    env:
      MOLECULE_CP_URL: https://staging-api.moleculesai.app
-      MOLECULE_ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
-      E2E_MODE: canary
+      # 2026-05-11: secret canonicalised from MOLECULE_STAGING_ADMIN_TOKEN
+      # (dead in org secret store) to CP_STAGING_ADMIN_API_TOKEN per
+      # internal#322 — see this PR for the cross-workflow sweep.
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
+      E2E_MODE: smoke
      E2E_RUNTIME: hermes
      E2E_RUN_ID: "sanity-${{ github.run_id }}"
      E2E_INTENTIONAL_FAILURE: "1"
@@ -54,7 +57,7 @@ jobs:
      - name: Verify admin token present
        run: |
          if [ -z "$MOLECULE_ADMIN_TOKEN" ]; then
-            echo "::error::MOLECULE_STAGING_ADMIN_TOKEN not set"
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN not set"
            exit 2
          fi

@@ -118,7 +121,7 @@ jobs:
      - name: Teardown safety net
        if: always()
        env:
-          ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
        run: |
          set +e
          orgs=$(curl -sS "$MOLECULE_CP_URL/cp/admin/orgs" \
@@ -127,8 +130,14 @@ jobs:
          import json, sys
          d = json.load(sys.stdin)
          today = __import__('datetime').date.today().strftime('%Y%m%d')
+          # Match both the new e2e-smoke- prefix (post-2026-05-11 rename)
+          # and the legacy e2e-canary- prefix for one rollout cycle so
+          # any in-flight org provisioned under the old prefix on an
+          # older runner checkout still gets cleaned up. Remove the
+          # canary fallback after one week of no-old-prefix observations.
+          prefixes = (f'e2e-smoke-{today}-sanity-', f'e2e-canary-{today}-sanity-')
          candidates = [o['slug'] for o in d.get('orgs', [])
-                        if o.get('slug','').startswith(f'e2e-canary-{today}-sanity-')
+                        if any(o.get('slug','').startswith(p) for p in prefixes)
                        and o.get('status') not in ('purged',)]
          print('\n'.join(candidates))
          " 2>/dev/null)
@@ -68,7 +68,35 @@ jobs:
      run: ${{ steps.decide.outputs.run }}
    steps:
      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+      - name: Fetch base branch tip for diff
+        continue-on-error: true
+        run: |
+          # With the default fetch-depth: 1, actions/checkout only fetches the
+          # PR head commit. The base commit is NOT in the local history, so
+          # `git diff "$BASE" "$GITHUB_SHA"` fails. Fetch the base branch at
+          # depth 1 — the base commit is the immediate parent of the PR head
+          # on the base branch, so depth=1 is sufficient.
+          #
+          # Network: Gitea Actions runner (5.78.80.188) cannot reach the git
+          # remote over HTTPS (confirmed: git fetch times out at ~15s). The runner
+          # is on the same host as Gitea, but the container network namespace
+          # cannot reach the Gitea HTTPS endpoint.
+          #
+          # Fallback: if the base commit does not exist locally, skip the diff
+          # and set run=true (always run harness). This is safe: PRs where the
+          # base is unavailable still run the harness (correct), PRs where the
+          # base IS available get the correct path-based diff.
+          #
+          # Timeout: 20s. If the fetch completes, great. If it times out, the
+          # step exits non-zero and we fall through to run=true.
+          if timeout 20 git fetch origin "${{ github.event.pull_request.base.ref }}" --depth=1; then
+            echo "::notice::base branch fetched successfully"
+          else
+            echo "::warning::git fetch origin ${{ github.event.pull_request.base.ref }} --depth=1 timed out"
+            echo "::warning::Skipping diff — detect-changes will run the harness unconditionally."
+          fi
      - id: decide
+        continue-on-error: true
        run: |
          # workflow_dispatch: always run (manual trigger)
          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
@@ -95,7 +123,13 @@ jobs:
          fi

          # GitHub Actions and Gitea Actions both expose github.sha for HEAD.
-          DIFF=$(git diff --name-only "$BASE" "${{ github.sha }}" 2>/dev/null)
+          # git diff exits 1 when BASE is not in local history (e.g. shallow
+          # checkout where the base commit was never fetched). Capture and
+          # swallow that exit code — the empty diff means "run everything".
+          # The runner network cannot reach the git remote (confirmed: git fetch
+          # times out at ~15s), so a failed fetch is expected and we always fall
+          # through to the unconditional run=true below.
+          DIFF=$(git diff --name-only "$BASE" "${{ github.sha }}" 2>/dev/null) || true
          echo "debug=diff-base=$BASE diff-files=$DIFF" >> "$GITHUB_OUTPUT"

          if echo "$DIFF" | grep -qE '^workspace-server/|^canvas/|^tests/harness/|^.gitea/workflows/harness-replays\.yml$'; then
@@ -11,7 +11,7 @@ name: publish-canvas-image
 #   - `continue-on-error: true` on each job (RFC §1 contract).
 #   - **Open question for review**: this workflow pushes the canvas
 #     image to `ghcr.io`. GHCR was retired during the 2026-05-06
-#     Gitea migration in favor of ECR (per canary-verify.yml header
+#     Gitea migration in favor of ECR (per staging-verify.yml header
 #     notes). The image may not be consumable post-migration. Two
 #     options for follow-up: (a) retarget to
 #     `153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/canvas`,
@@ -43,10 +43,18 @@ jobs:
    steps:
      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
-          # Fetch full tag list so the bump logic can sanity-check against
-          # what's already in this repo (catches collision with prior
-          # manual tag pushes).
-          fetch-depth: 0
+          # Shallow clone — depth 1 is enough for the workspace-diff check.
+          # Tags needed for the collision check below are fetched explicitly
+          # in the next step, bypassing the runner-network timeout that
+          # full-history fetch triggers on Gitea Actions runners
+          # (runbooks/gitea-operational-quirks.md §runner-network-isolation).
+          fetch-depth: 1
+
+      - name: Fetch tags for collision check
+        # fetch-depth: 1 gets only the most recent commit's refs, not the
+        # tag that points at it. Do a targeted tag fetch so git tag --list
+        # below can detect collision with prior manual pushes.
+        run: git fetch origin --tags --depth=1

      - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
        with:
@@ -32,7 +32,7 @@ name: redeploy-tenants-on-main
 #
 # Registry: ECR (153263036946.dkr.ecr.us-east-2.amazonaws.com/
 # molecule-ai/platform-tenant). GHCR was retired 2026-05-07 during the
-# Gitea suspension migration. The canary-verify.yml promote step now
+# Gitea suspension migration. The staging-verify.yml promote step now
 # uses the same redeploy-fleet endpoint (fixes the silent-GHCR gap).
 #
 # Runtime ordering:
@@ -104,7 +104,7 @@ jobs:
        #      `staging-<sha>` to roll back to a known-good build.
        #   2. Default → `staging-<short_head_sha>`. The just-published
        #      digest. Bypasses the `:latest` retag path that's currently
-        #      dead (canary-verify soft-skips without canary fleet, so
+        #      dead (staging-verify soft-skips without canary fleet, so
        #      the only thing retagging `:latest` today is the manual
        #      promote-latest.yml — last run 2026-04-28). Auto-trigger
        #      from workflow_run uses workflow_run.head_sha; manual
@@ -359,7 +359,7 @@ jobs:

          # Belt-and-suspenders sanity floor: same logic as the staging
          # variant — see that file's comment for the full rationale.
-          # Floor only applies when fleet >= 4; below that, canary-verify
+          # Floor only applies when fleet >= 4; below that, staging-verify
          # is the actual gate.
          TOTAL_VERIFIED=${#SLUGS[@]}
          if [ $TOTAL_VERIFIED -ge 4 ] && [ $UNREACHABLE_COUNT -gt $((TOTAL_VERIFIED / 2)) ]; then
@@ -21,7 +21,7 @@ name: redeploy-tenants-on-staging
 #
 # Mirror of redeploy-tenants-on-main.yml, with the staging-CP host and
 # the :staging-latest tag. Sister workflow exists for prod (rolls
-# :latest after canary-verify). Both share the same shape — just
+# :latest after staging-verify). Both share the same shape — just
 # different CP_URL + target_tag + admin token secret.
 #
 # Why this workflow exists: publish-workspace-server-image now builds
@@ -336,7 +336,7 @@ jobs:
          # crashes on startup), not a teardown race. Hard-fail.
          #
          # Floor only applies when TOTAL_VERIFIED >= 4 — below that, the
-          # canary-verify step is the actual gate for "all tenants down"
+          # staging-verify step is the actual gate for "all tenants down"
          # detection (it runs against the canary first and aborts the
          # rollout if the canary fails to come up). Without the >=4 gate,
          # a 1-tenant fleet (e.g. a single ephemeral e2e-* tenant on a
@@ -0,0 +1,79 @@
+# sop-tier-refire — issue_comment-triggered refire of sop-tier-check.
+#
+# Closes internal#292. Gitea 1.22.6 doesn't refire workflows on the
+# `pull_request_review` event (go-gitea/gitea#33700); the `sop-tier-check`
+# workflow's review-event subscription is silently dead. The result:
+# PRs that get their approving review AFTER the tier-check ran on open/
+# synchronize keep their failing status check forever, and the only way
+# to merge is the admin force-merge path (audited via `audit-force-merge`
+# but the audit trail keeps growing; see `feedback_never_admin_merge_bypass`).
+#
+# Workaround pattern from `feedback_pull_request_review_no_refire`:
+# `issue_comment` events DO fire reliably on 1.22.6. When a repo
+# MEMBER/OWNER/COLLABORATOR comments `/refire-tier-check` on a PR, this
+# workflow re-runs the sop-tier-check logic and POSTs the resulting
+# status to the PR head SHA directly. No empty commit, no git history
+# bloat, no cascade re-fire of every other workflow on the PR.
+#
+# SECURITY MODEL:
+#
+# 1. `pull_request` exists on the issue (issue_comment fires on issues
+#    AND PRs; we only want PRs).
+# 2. `comment.author_association` must be MEMBER/OWNER/COLLABORATOR.
+#    Per the internal#292 core-security review (review#1066 ask): anyone
+#    can comment, but only repo collaborators+ can flip the status.
+#    Without this gate, a drive-by commenter on a public-issue-tracker
+#    surface could trigger a status flip.
+# 3. Comment body must contain `/refire-tier-check` — a slash-command-
+#    shaped trigger (not just any comment word). Prevents accidental
+#    triggering from prose like "we should refire tests" in a review.
+# 4. This workflow does NOT check out PR HEAD code. Like sop-tier-check,
+#    it only HTTP-calls the Gitea API. Trust boundary preserved.
+#
+# Note: `issue_comment` fires from the BASE branch's workflow file. There
+# is no `pull_request_target` equivalent to set; the trigger inherently
+# loads the workflow from the default branch.
+#
+# Rate-limit: a 1s pre-sleep + a "skip if status posted in last 30s"
+# guard prevents comment-spam from thrashing the status. See the script.
+
+name: sop-tier-check refire (issue_comment)
+
+on:
+  issue_comment:
+    types: [created]
+
+jobs:
+  refire:
+    # Three gates, all required:
+    #   - comment is on a PR (not a plain issue)
+    #   - commenter is MEMBER, OWNER, or COLLABORATOR
+    #   - comment body contains the slash-command trigger
+    if: |
+      github.event.issue.pull_request != null &&
+      contains(fromJson('["MEMBER","OWNER","COLLABORATOR"]'), github.event.comment.author_association) &&
+      contains(github.event.comment.body, '/refire-tier-check')
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: read
+      statuses: write
+    steps:
+      - name: Check out base branch (for the script)
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+        with:
+          # Load the script from the default branch (main), matching the
+          # sop-tier-check.yml security model.
+          ref: ${{ github.event.repository.default_branch }}
+      - name: Re-evaluate sop-tier-check and POST status
+        env:
+          # Same org-level secret sop-tier-check.yml + audit-force-merge.yml use.
+          # Fallback to GITHUB_TOKEN with a clear error if missing.
+          GITEA_TOKEN: ${{ secrets.SOP_TIER_CHECK_TOKEN || secrets.GITHUB_TOKEN }}
+          GITEA_HOST: git.moleculesai.app
+          REPO: ${{ github.repository }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          COMMENT_AUTHOR: ${{ github.event.comment.user.login }}
+          # Set to '1' for diagnostic per-API-call output. Off by default.
+          SOP_DEBUG: '0'
+        run: bash .gitea/scripts/sop-tier-refire.sh
@@ -1,6 +1,8 @@
-name: Canary — staging SaaS smoke (every 30 min)
+name: Staging SaaS smoke (every 30 min)

-# Ported from .github/workflows/canary-staging.yml on 2026-05-11 per RFC
+# Renamed from canary-staging.yml on 2026-05-11 per Hongming directive
+# ("canary naming changed to staging for all"). Originally ported from
+# .github/workflows/canary-staging.yml on 2026-05-11 per RFC
 # internal#219 §1 sweep. Differences from the GitHub version:
 #   - Dropped `workflow_dispatch.inputs` (Gitea 1.22.6 parser rejects them
 #     per feedback_gitea_workflow_dispatch_inputs_unsupported).
@@ -21,21 +23,21 @@ name: Canary — staging SaaS smoke (every 30 min)
 # catches drift in the 30-min window between those runs (AMI health, CF
 # cert rotation, WorkOS session stability, etc.).
 #
-# Lean mode: E2E_MODE=canary skips the child workspace + HMA memory +
+# Lean mode: E2E_MODE=smoke skips the child workspace + HMA memory +
 # peers/activity checks. One parent workspace + one A2A turn is enough
 # to signal "SaaS stack end-to-end is alive."

 on:
  schedule:
    # Every 30 min. Cron on GitHub-hosted runners has a known drift of
-    # a few minutes under load — that's fine for a canary.
+    # a few minutes under load — that's fine for a smoke check.
    - cron: '*/30 * * * *'
 # Serialise with the full-SaaS workflow so they don't contend for the
 # same org-create quota on staging. Different group key from
-# e2e-staging-saas since we don't mind queueing canaries behind one
-# full run, but two canaries SHOULD queue against each other.
+# e2e-staging-saas since we don't mind queueing smoke runs behind one
+# full run, but two smoke runs SHOULD queue against each other.
 concurrency:
-  group: canary-staging
+  group: staging-smoke
  cancel-in-progress: false

 permissions:
@@ -47,32 +49,47 @@ env:
  GITHUB_SERVER_URL: https://git.moleculesai.app

 jobs:
-  canary:
-    name: Canary smoke
+  smoke:
+    name: Staging SaaS smoke
    runs-on: ubuntu-latest
-    # Phase 3 (RFC #219 §1): surface broken workflows without blocking.
-    continue-on-error: true
+    # NOTE: Phase 3 (RFC #219 §1) `continue-on-error: true` removed
+    # 2026-05-11. The "surface broken workflows without blocking"
+    # rationale was correctly applied to advisory/lint workflows but
+    # wrong for this smoke — it is the 30-min canary cadence for the
+    # entire staging SaaS stack, and silent failure here masks the
+    # exact regressions the smoke exists to surface (AMI rot, CF cert
+    # drift, WorkOS session breakage, secret rotations). Same class of
+    # failure as PR#461 (`sweep-stale-e2e-orgs`) where Phase-3 silent
+    # failure leaked EC2. The four other `e2e-staging-*` workflows
+    # KEEP `continue-on-error: true` per RFC #219 §1 — they are
+    # advisory and matrix-style; this one is the canary. A follow-up
+    # `notify-failure` step below also surfaces breakage to ops even
+    # if branch-protection wiring is adjusted to keep this off the
+    # required-checks list.
    # 25 min headroom over the 15-min TLS-readiness deadline in
    # tests/e2e/test_staging_full_saas.sh (#2107). Without the buffer
    # the job is killed at the wall-clock 15:00 mark BEFORE the bash
    # `fail` + diagnostic burst can fire, leaving every cancellation
-    # silent. Sibling staging E2E jobs run at 20-45 min — keeping
-    # canary tighter than them so a true wedge still surfaces here
+    # silent. Sibling staging E2E jobs run at 20-45 min — keeping the
+    # smoke tighter than them so a true wedge still surfaces here
    # first.
    timeout-minutes: 25

    env:
      MOLECULE_CP_URL: https://staging-api.moleculesai.app
-      MOLECULE_ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
-      # MiniMax is the canary's PRIMARY LLM auth path post-2026-05-04.
+      # 2026-05-11: secret canonicalised from MOLECULE_STAGING_ADMIN_TOKEN
+      # (dead in org secret store) to CP_STAGING_ADMIN_API_TOKEN per
+      # internal#322 — see this PR for the cross-workflow sweep.
+      MOLECULE_ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
+      # MiniMax is the smoke's PRIMARY LLM auth path post-2026-05-04.
      # Switched from hermes+OpenAI after #2578 (the staging OpenAI key
      # account went over quota and stayed dead for 36+ hours, taking
-      # the canary red the entire time). claude-code template's
+      # the smoke red the entire time). claude-code template's
      # `minimax` provider routes ANTHROPIC_BASE_URL to
      # api.minimax.io/anthropic and reads MINIMAX_API_KEY at boot —
      # ~5-10x cheaper per token than gpt-4.1-mini AND on a separate
      # billing account, so OpenAI quota collapse no longer wedges the
-      # canary. Mirrors the migration continuous-synth-e2e.yml made on
+      # smoke. Mirrors the migration continuous-synth-e2e.yml made on
      # 2026-05-03 (#265) for the same reason. tests/e2e/test_staging_
      # full_saas.sh branches SECRETS_JSON on which key is present —
      # MiniMax wins when set.
@@ -86,16 +103,16 @@ jobs:
      # E2E_RUNTIME=hermes overridden via workflow_dispatch can still
      # exercise the OpenAI path without re-editing the workflow.
      E2E_OPENAI_API_KEY: ${{ secrets.MOLECULE_STAGING_OPENAI_API_KEY }}
-      E2E_MODE: canary
+      E2E_MODE: smoke
      E2E_RUNTIME: claude-code
-      # Pin the canary to a specific MiniMax model rather than relying
+      # Pin the smoke to a specific MiniMax model rather than relying
      # on the per-runtime default (which could resolve to "sonnet" →
      # direct Anthropic and defeat the cost saving). M2.7-highspeed
      # is "Token Plan only" but cheap-per-token and fast.
      E2E_MODEL_SLUG: MiniMax-M2.7-highspeed
-      E2E_RUN_ID: "canary-${{ github.run_id }}"
+      E2E_RUN_ID: "smoke-${{ github.run_id }}"
      # Debug-only: when an operator dispatches with keep_on_failure=true,
-      # the canary script's E2E_KEEP_ORG=1 path skips teardown so the
+      # the smoke script's E2E_KEEP_ORG=1 path skips teardown so the
      # tenant org + EC2 stay alive for SSM-based log capture. Cron runs
      # never set this (the input only exists on workflow_dispatch) so
      # unattended cron always tears down. See molecule-core#129
@@ -109,7 +126,7 @@ jobs:
      - name: Verify admin token present
        run: |
          if [ -z "$MOLECULE_ADMIN_TOKEN" ]; then
-            echo "::error::MOLECULE_STAGING_ADMIN_TOKEN not set"
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN not set"
            exit 2
          fi

@@ -119,7 +136,7 @@ jobs:
          # langgraph (operator-dispatched only) use OpenAI. Hard-fail
          # rather than soft-skip per the lesson from synth E2E #2578:
          # an empty key silently falls through to the wrong
-          # SECRETS_JSON branch and the canary fails 5 min later with
+          # SECRETS_JSON branch and the smoke fails 5 min later with
          # a confusing auth error instead of the clean "secret
          # missing" message at the top.
          case "${E2E_RUNTIME}" in
@@ -155,8 +172,8 @@ jobs:
          fi
          echo "LLM key present ✓ (runtime=${E2E_RUNTIME}, key=${required_secret_name}, len=${#required_secret_value})"

-      - name: Canary run
-        id: canary
+      - name: Smoke run
+        id: smoke
        run: bash tests/e2e/test_staging_full_saas.sh

      # Alerting: open a sticky issue on the FIRST failure; comment on
@@ -184,6 +201,9 @@ jobs:
        run: |
          set -euo pipefail
          API="${SERVER_URL%/}/api/v1"
+          # Title kept stable across the canary-staging.yml → staging-smoke.yml
+          # rename (2026-05-11) so any open alert issue from the old name
+          # still title-matches and auto-closes on the next green run.
          TITLE="Canary failing: staging SaaS smoke"
          RUN_URL="${SERVER_URL}/${REPO}/actions/runs/${RUN_ID}"

@@ -194,18 +214,18 @@ jobs:
          if [ -n "$EXISTING" ]; then
            curl -fsS -X POST -H "Authorization: token $GITEA_TOKEN" -H "Content-Type: application/json" \
              "${API}/repos/${REPO}/issues/${EXISTING}/comments" \
-              -d "$(jq -nc --arg run "$RUN_URL" '{body: ("Canary still failing. " + $run)}')" >/dev/null
+              -d "$(jq -nc --arg run "$RUN_URL" '{body: ("Smoke still failing. " + $run)}')" >/dev/null
            echo "Commented on existing issue #${EXISTING}"
          else
            NOW=$(date -u +%Y-%m-%dT%H:%M:%SZ)
            BODY=$(jq -nc --arg t "$TITLE" --arg now "$NOW" --arg run "$RUN_URL" \
-              '{title: $t, body: ("Canary run failed at " + $now + ".\n\nRun: " + $run + "\n\nThis issue auto-closes on the next green canary run. Consecutive failures add a comment here rather than a new issue.")}')
+              '{title: $t, body: ("Smoke run failed at " + $now + ".\n\nRun: " + $run + "\n\nThis issue auto-closes on the next green smoke run. Consecutive failures add a comment here rather than a new issue.")}')
            curl -fsS -X POST -H "Authorization: token $GITEA_TOKEN" -H "Content-Type: application/json" \
              "${API}/repos/${REPO}/issues" -d "$BODY" >/dev/null
-            echo "Opened canary failure issue (first red)"
+            echo "Opened smoke failure issue (first red)"
          fi

-      - name: Auto-close canary issue on success (Gitea API)
+      - name: Auto-close smoke issue on success (Gitea API)
        if: success()
        env:
          GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -215,6 +235,8 @@ jobs:
        run: |
          set -euo pipefail
          API="${SERVER_URL%/}/api/v1"
+          # Title kept stable across the canary-staging.yml → staging-smoke.yml
+          # rename so open alert issues from the old name still match.
          TITLE="Canary failing: staging SaaS smoke"

          NUMS=$(curl -fsS -H "Authorization: token $GITEA_TOKEN" \
@@ -225,37 +247,36 @@ jobs:
          for N in $NUMS; do
            curl -fsS -X POST -H "Authorization: token $GITEA_TOKEN" -H "Content-Type: application/json" \
              "${API}/repos/${REPO}/issues/${N}/comments" \
-              -d "$(jq -nc --arg now "$NOW" '{body: ("Canary recovered at " + $now + ". Closing.")}')" >/dev/null
+              -d "$(jq -nc --arg now "$NOW" '{body: ("Smoke recovered at " + $now + ". Closing.")}')" >/dev/null
            curl -fsS -X PATCH -H "Authorization: token $GITEA_TOKEN" -H "Content-Type: application/json" \
              "${API}/repos/${REPO}/issues/${N}" -d '{"state":"closed"}' >/dev/null
-            echo "Closed recovered canary issue #${N}"
+            echo "Closed recovered smoke issue #${N}"
          done

      - name: Teardown safety net
        if: always()
        env:
-          ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+          ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
        run: |
          set +e
          # Slug prefix matches what test_staging_full_saas.sh emits
-          # in canary mode:
-          #   SLUG="e2e-canary-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
-          # Earlier this was `e2e-{today}-canary-` — that was the
-          # full-mode pattern (date FIRST, mode SECOND); canary slugs
-          # have mode FIRST, date SECOND. The mismatch silently
-          # never matched, leaving every cancelled-canary EC2 alive
-          # until the once-an-hour sweep eventually caught it
-          # (incident 2026-04-26 21:03Z: 1h25m EC2 leak before manual
-          # cleanup; same gap on three earlier cancellations today).
+          # in smoke mode:
+          #   SLUG="e2e-smoke-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
+          # Earlier (pre-2026-05-11 canary→staging rename) the prefix was
+          # `e2e-canary-`; both prefixes are matched here for one
+          # release cycle so cleanup still catches any in-flight org
+          # provisioned under the old prefix on an older runner that
+          # hasn't picked up the renamed script. Remove the canary
+          # fallback after one week of no-old-prefix observations.
          orgs=$(curl -sS "$MOLECULE_CP_URL/cp/admin/orgs" \
            -H "Authorization: Bearer $ADMIN_TOKEN" 2>/dev/null \
            | python3 -c "
          import json, sys, os, datetime
          run_id = os.environ.get('GITHUB_RUN_ID', '')
          d = json.load(sys.stdin)
-          # Scope to slugs from THIS canary run when GITHUB_RUN_ID is
-          # available; the canary workflow sets E2E_RUN_ID='canary-\${run_id}'
-          # so the slug suffix is '-canary-\${run_id}-...'. Mirrors the
+          # Scope to slugs from THIS smoke run when GITHUB_RUN_ID is
+          # available; the smoke workflow sets E2E_RUN_ID='smoke-\${run_id}'
+          # so the slug suffix is '-smoke-\${run_id}-...'. Mirrors the
          # full-mode safety net's per-run scoping (e2e-staging-saas.yml)
          # added after the 2026-04-21 cross-run cleanup incident.
          # Sweep both today AND yesterday's UTC dates so a run that
@@ -265,9 +286,11 @@ jobs:
          yesterday = today - datetime.timedelta(days=1)
          dates = (today.strftime('%Y%m%d'), yesterday.strftime('%Y%m%d'))
          if run_id:
-              prefixes = tuple(f'e2e-canary-{d}-canary-{run_id}' for d in dates)
+              prefixes = tuple(f'e2e-smoke-{d}-smoke-{run_id}' for d in dates) \
+                       + tuple(f'e2e-canary-{d}-canary-{run_id}' for d in dates)
          else:
-              prefixes = tuple(f'e2e-canary-{d}-' for d in dates)
+              prefixes = tuple(f'e2e-smoke-{d}-' for d in dates) \
+                       + tuple(f'e2e-canary-{d}-' for d in dates)
          candidates = [o['slug'] for o in d.get('orgs', [])
                        if any(o.get('slug','').startswith(p) for p in prefixes)
                        and o.get('status') not in ('purged',)]
@@ -280,8 +303,8 @@ jobs:
          # stale sweep caught it (up to 2h later). Now we capture the
          # response code and surface non-2xx as a workflow warning, so
          # the run page shows which slug leaked. We still don't `exit 1`
-          # on cleanup failure — a single-canary cleanup miss shouldn't
-          # fail-flag the canary itself when the actual smoke check
+          # on cleanup failure — a single-smoke cleanup miss shouldn't
+          # fail-flag the smoke itself when the actual smoke check
          # passed. The sweep-stale-e2e-orgs cron (now every 15 min,
          # 30-min threshold) is the safety net for whatever slips past.
          # See molecule-controlplane#420.
@@ -290,21 +313,34 @@ jobs:
            # Tempfile-routed -w + set +e/-e prevents curl-exit-code
            # pollution of the captured status (lint-curl-status-capture.yml).
            set +e
-            curl -sS -o /tmp/canary-cleanup.out -w "%{http_code}" \
+            curl -sS -o /tmp/smoke-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/canary-cleanup.code
+              -d "{\"confirm\":\"$slug\"}" >/tmp/smoke-cleanup.code
            set -e
-            code=$(cat /tmp/canary-cleanup.code 2>/dev/null || echo "000")
+            code=$(cat /tmp/smoke-cleanup.code 2>/dev/null || echo "000")
            if [ "$code" = "200" ] || [ "$code" = "204" ]; then
              echo "[teardown] deleted $slug (HTTP $code)"
            else
-              echo "::warning::canary teardown for $slug returned HTTP $code — sweep-stale-e2e-orgs will catch it within ~45 min. Body: $(head -c 300 /tmp/canary-cleanup.out 2>/dev/null)"
+              echo "::warning::smoke teardown for $slug returned HTTP $code — sweep-stale-e2e-orgs will catch it within ~45 min. Body: $(head -c 300 /tmp/smoke-cleanup.out 2>/dev/null)"
              leaks+=("$slug")
            fi
          done
          if [ ${#leaks[@]} -gt 0 ]; then
-            echo "::warning::canary teardown left ${#leaks[@]} leak(s): ${leaks[*]}"
+            echo "::warning::smoke teardown left ${#leaks[@]} leak(s): ${leaks[*]}"
          fi
          exit 0
+
+      - name: Notify on smoke failure
+        # Fail-loud companion to dropping `continue-on-error: true`.
+        # The Open-issue-on-failure step above handles the human-facing
+        # alert; this step emits a clearly-tagged ::error:: line that
+        # log-tail consumers (Loki SOPRefireRule, orchestrator triage
+        # loop) can grep on. Mirrors PR#461's sweep-stale-e2e-orgs
+        # pattern. Runs AFTER the teardown safety net (which is
+        # if: always()) so failures don't suppress cleanup.
+        if: failure()
+        run: |
+          echo "::error::staging-smoke FAILED — staging SaaS canary is red. See prior step logs + the auto-filed alert issue. Common causes: (a) CP_STAGING_ADMIN_API_TOKEN secret missing/rotated, (b) staging-api.moleculesai.app 5xx, (c) MiniMax/Anthropic LLM key dead, (d) AMI/CF/WorkOS drift. The 30-min cron will retry, but a chronic red here indicates the staging SaaS stack is broken end-to-end."
+          exit 1
@@ -1,6 +1,8 @@
-name: canary-verify
+name: Staging verify

-# Ported from .github/workflows/canary-verify.yml on 2026-05-11 per RFC
+# Renamed from canary-verify.yml on 2026-05-11 per Hongming directive
+# ("canary naming changed to staging for all"). Originally ported from
+# .github/workflows/canary-verify.yml on 2026-05-11 per RFC
 # internal#219 §1 sweep. Differences from the GitHub version:
 #   - Dropped `workflow_dispatch.inputs` (Gitea 1.22.6 parser rejects them
 #     per feedback_gitea_workflow_dispatch_inputs_unsupported).
@@ -23,13 +25,22 @@ name: canary-verify
 # digest. On red, :latest stays on the prior known-good digest and
 # prod is untouched.
 #
+# Terminology note (2026-05-11): The deployment STRATEGY here is still
+# called "canary release" (a small subset of tenants gets the new image
+# first, the rest follow on green). The "canary" word stays for the
+# pre-fan-out cohort concept (see docs/architecture/canary-release.md
+# and CANARY_SLUG in redeploy-tenants-on-*.yml). What changed is the
+# FILE NAME and the SECRETS feeding this workflow — both are renamed
+# to drop the redundant "canary-" prefix that conflated workflow
+# identity with deployment strategy.
+#
 # Registry note (2026-05-10): This workflow previously used GHCR
 # (ghcr.io/molecule-ai/platform-tenant) — that registry was retired
 # during the 2026-05-06 Gitea suspension migration when publish-
 # workspace-server-image.yml switched to the operator's ECR org
 # (153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/
 # platform-tenant). The GHCR → ECR migration was never applied to
-# this file, so canary-verify was silently smoke-testing the stale
+# this file, so this workflow was silently smoke-testing the stale
 # GHCR image while the actual staging/prod tenants ran the ECR image.
 # Result: smoke tests could not catch a broken ECR build. Fix:
 #   - Wait step: reads SHA from running canary /health (tenant-
@@ -43,8 +54,9 @@ name: canary-verify
 #     to ECR on staging and main merges.
 #   - Canary tenants are configured to pull :staging-<sha> from ECR
 #     (TENANT_IMAGE env set to the ECR :staging-<sha> tag).
-#   - Repo secrets CANARY_TENANT_URLS / CANARY_ADMIN_TOKENS /
-#     CANARY_CP_SHARED_SECRET are populated.
+#   - Repo secrets MOLECULE_STAGING_TENANT_URLS /
+#     MOLECULE_STAGING_ADMIN_TOKENS / MOLECULE_STAGING_CP_SHARED_SECRET
+#     are populated.

 on:
  workflow_run:
@@ -65,7 +77,7 @@ env:
  GITHUB_SERVER_URL: https://git.moleculesai.app

 jobs:
-  canary-smoke:
+  staging-smoke:
    # Skip when the upstream workflow failed — no image to test against.
    # workflow_dispatch trigger dropped in this Gitea port; only the
    # workflow_run path remains.
@@ -97,15 +109,15 @@ jobs:
        # other registry — the canary is telling us what it's actually
        # running, which is the ground truth for smoke testing.
        env:
-          CANARY_TENANT_URLS: ${{ secrets.CANARY_TENANT_URLS }}
+          MOLECULE_STAGING_TENANT_URLS: ${{ secrets.MOLECULE_STAGING_TENANT_URLS }}
          EXPECTED_SHA: ${{ steps.compute.outputs.sha }}
        run: |
-          if [ -z "$CANARY_TENANT_URLS" ]; then
+          if [ -z "$MOLECULE_STAGING_TENANT_URLS" ]; then
            echo "No canary URLs configured — falling back to 60s wait"
            sleep 60
            exit 0
          fi
-          IFS=',' read -ra URLS <<< "$CANARY_TENANT_URLS"
+          IFS=',' read -ra URLS <<< "$MOLECULE_STAGING_TENANT_URLS"
          MAX_WAIT=420  # 7 minutes
          INTERVAL=30
          ELAPSED=0
@@ -129,7 +141,7 @@ jobs:
          done
          echo "Timeout after ${MAX_WAIT}s — proceeding anyway (smoke suite will validate)"

-      - name: Run canary smoke suite
+      - name: Run staging smoke suite
        id: smoke
        # Graceful-skip when no canary fleet is configured (Phase 2 not yet
        # stood up — see molecule-controlplane/docs/canary-tenants.md).
@@ -138,29 +150,29 @@ jobs:
        # promote-latest.yml is the release gate while canary is absent.
        # Once the fleet is real: delete the early-exit branch.
        env:
-          CANARY_TENANT_URLS: ${{ secrets.CANARY_TENANT_URLS }}
-          CANARY_ADMIN_TOKENS: ${{ secrets.CANARY_ADMIN_TOKENS }}
-          CANARY_CP_BASE_URL: https://staging-api.moleculesai.app
-          CANARY_CP_SHARED_SECRET: ${{ secrets.CANARY_CP_SHARED_SECRET }}
+          MOLECULE_STAGING_TENANT_URLS: ${{ secrets.MOLECULE_STAGING_TENANT_URLS }}
+          MOLECULE_STAGING_ADMIN_TOKENS: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKENS }}
+          MOLECULE_STAGING_CP_BASE_URL: https://staging-api.moleculesai.app
+          MOLECULE_STAGING_CP_SHARED_SECRET: ${{ secrets.MOLECULE_STAGING_CP_SHARED_SECRET }}
        run: |
          set -euo pipefail
-          if [ -z "${CANARY_TENANT_URLS:-}" ] \
-            || [ -z "${CANARY_ADMIN_TOKENS:-}" ] \
-            || [ -z "${CANARY_CP_SHARED_SECRET:-}" ]; then
+          if [ -z "${MOLECULE_STAGING_TENANT_URLS:-}" ] \
+            || [ -z "${MOLECULE_STAGING_ADMIN_TOKENS:-}" ] \
+            || [ -z "${MOLECULE_STAGING_CP_SHARED_SECRET:-}" ]; then
            {
-              echo "## ⚠️ canary-verify skipped"
+              echo "## ⚠️ staging-verify skipped"
              echo
-              echo "One or more canary secrets are unset (\`CANARY_TENANT_URLS\`, \`CANARY_ADMIN_TOKENS\`, \`CANARY_CP_SHARED_SECRET\`)."
+              echo "One or more canary secrets are unset (\`MOLECULE_STAGING_TENANT_URLS\`, \`MOLECULE_STAGING_ADMIN_TOKENS\`, \`MOLECULE_STAGING_CP_SHARED_SECRET\`)."
              echo "Phase 2 canary fleet has not been stood up yet —"
              echo "see [canary-tenants.md](https://git.moleculesai.app/molecule-ai/molecule-controlplane/blob/main/docs/canary-tenants.md)."
              echo
              echo "**Skipped — promote-to-latest will NOT auto-fire.** Dispatch \`promote-latest.yml\` manually when ready."
            } >> "$GITHUB_STEP_SUMMARY"
            echo "ran=false" >> "$GITHUB_OUTPUT"
-            echo "::notice::canary-verify: skipped — no canary fleet configured"
+            echo "::notice::staging-verify: skipped — no canary fleet configured"
            exit 0
          fi
-          bash scripts/canary-smoke.sh
+          bash scripts/staging-smoke.sh
          echo "ran=true" >> "$GITHUB_OUTPUT"

      - name: Summary on failure
@@ -173,7 +185,7 @@ jobs:
            echo ":latest stays pinned to the prior good digest — prod is untouched."
            echo
            echo "Fix forward and merge again, or investigate the specific failed"
-            echo "assertions in the canary-smoke step log above."
+            echo "assertions in the staging-smoke step log above."
          } >> "$GITHUB_STEP_SUMMARY"

  promote-to-latest:
@@ -188,13 +200,13 @@ jobs:
    # silently promoting a stale GHCR image while actual prod tenants
    # pulled from ECR. Canary smoke tests were GHCR-targeted and could
    # not catch a broken ECR build.
-    needs: canary-smoke
-    if: ${{ needs.canary-smoke.result == 'success' && needs.canary-smoke.outputs.smoke_ran == 'true' }}
+    needs: staging-smoke
+    if: ${{ needs.staging-smoke.result == 'success' && needs.staging-smoke.outputs.smoke_ran == 'true' }}
    runs-on: ubuntu-latest
    # Phase 3 (RFC #219 §1): surface broken workflows without blocking.
    continue-on-error: true
    env:
-      SHA: ${{ needs.canary-smoke.outputs.sha }}
+      SHA: ${{ needs.staging-smoke.outputs.sha }}
      CP_URL: ${{ vars.CP_URL || 'https://staging-api.moleculesai.app' }}
      # CP_ADMIN_API_TOKEN gates write access to the redeploy endpoint.
      # Stored at the repo level so all workflows pick it up automatically.
@@ -264,9 +276,9 @@ jobs:
      - name: Summary
        run: |
          {
-            echo "## Canary verified — :latest promoted via CP redeploy-fleet"
+            echo "## Staging verified — :latest promoted via CP redeploy-fleet"
            echo ""
-            echo "- **Target tag:** \`staging-${{ needs.canary-smoke.outputs.sha }}\`"
+            echo "- **Target tag:** \`staging-${{ needs.staging-smoke.outputs.sha }}\`"
            echo "- **Registry:** ECR (\`${TENANT_IMAGE_NAME}\`)"
            echo "- **Canary slug:** \`${CANARY_SLUG:-<none>}\` (soak ${SOAK_SECONDS}s)"
            echo "- **Batch size:** ${BATCH_SIZE:-3}"
@@ -63,12 +63,21 @@ jobs:
  sweep:
    name: Sweep e2e orgs
    runs-on: ubuntu-latest
-    # Phase 3 (RFC #219 §1): surface broken workflows without blocking.
-    continue-on-error: true
+    # NOTE: Phase 3 (RFC #219 §1) `continue-on-error: true` removed
+    # 2026-05-11. The "surface broken workflows without blocking"
+    # rationale was correctly applied to advisory/lint workflows but
+    # wrong for this janitor — silent failure here masks real-money
+    # tenant leaks. Hongming observed 15 leaked EC2 in molecule-canary
+    # (004947743811) us-east-2 at 11:05Z 2026-05-11 because the sweep
+    # had been exiting 2 every tick and the failure was swallowed.
+    # See `feedback_strict_root_only_after_class_a` — critical janitors
+    # must fail loud. A follow-up `notify-failure` step below also
+    # surfaces breakage to ops even if branch-protection wiring is
+    # adjusted to keep this off the required-checks list.
    timeout-minutes: 15
    env:
      MOLECULE_CP_URL: https://staging-api.moleculesai.app
-      ADMIN_TOKEN: ${{ secrets.MOLECULE_STAGING_ADMIN_TOKEN }}
+      ADMIN_TOKEN: ${{ secrets.CP_STAGING_ADMIN_API_TOKEN }}
      MAX_AGE_MINUTES: ${{ github.event.inputs.max_age_minutes || '30' }}
      DRY_RUN: ${{ github.event.inputs.dry_run || 'false' }}
      # Refuse to delete more than this many orgs in one tick. If the
@@ -81,7 +90,7 @@ jobs:
      - name: Verify admin token present
        run: |
          if [ -z "$ADMIN_TOKEN" ]; then
-            echo "::error::MOLECULE_STAGING_ADMIN_TOKEN not set"
+            echo "::error::CP_STAGING_ADMIN_API_TOKEN not set"
            exit 2
          fi
          echo "Admin token present ✓"
@@ -99,7 +108,8 @@ jobs:

          # Filter:
          #   1. slug starts with one of the ephemeral test prefixes:
-          #        - 'e2e-'    — covers e2e-canary-, e2e-canvas-*, etc.
+          #        - 'e2e-'    — covers e2e-smoke- (formerly e2e-canary-),
+          #                      e2e-canvas-*, etc.
          #        - 'rt-e2e-' — runtime-test harness fixtures (RFC #2251);
          #                      missing this prefix left two such tenants
          #                      orphaned 8h on staging (2026-05-03), then
@@ -241,3 +251,17 @@ jobs:
        if: env.DRY_RUN == 'true'
        run: |
          echo "DRY RUN — would have deleted ${{ steps.identify.outputs.count }} org(s) AND triggered orphan-tunnels cleanup. Re-run with dry_run=false to actually delete."
+
+      - name: Notify on sweep failure
+        # Fail-loud companion to dropping `continue-on-error: true`.
+        # If any prior step failed (missing token, CP 5xx, safety-cap
+        # tripped, etc.) emit a clearly-tagged ::error:: line so the
+        # Gitea runs UI + any log-tail consumer (Loki SOPRefireRule)
+        # flags this. Without this step, an early `exit 2` shows as a
+        # red run but the message can scroll past in busy log windows;
+        # the explicit tag here is greppable from the orchestrator
+        # triage loop.
+        if: failure()
+        run: |
+          echo "::error::sweep-stale-e2e-orgs FAILED — staging tenants are LEAKING. See prior step logs. Common causes: (a) CP_STAGING_ADMIN_API_TOKEN secret missing/rotated, (b) staging-api.moleculesai.app 5xx, (c) safety-cap tripped (CP admin API returning malformed orgs). Manual cleanup of leaked EC2 + DNS may be required while this is broken."
+          exit 1
@@ -41,9 +41,10 @@ const pendingApproval = (id = "a1", workspaceId = "ws-1"): {
  created_at: "2026-05-10T10:00:00Z",
 });

-// Shared spy reference so individual tests can call mockGet.mockRestore()
-// without needing to pass it through beforeEach → it scope chain.
+// Shared spy references so individual tests can reset or reject the POST mock
+// without needing to call spyOn again (which would create a duplicate spy).
 let mockGet: ReturnType<typeof vi.spyOn>;
+let mockPost: ReturnType<typeof vi.spyOn>;

 // ─── Tests ────────────────────────────────────────────────────────────────────

@@ -139,8 +140,8 @@ describe("ApprovalBanner — renders approval cards", () => {
 describe("ApprovalBanner — decisions", () => {
  beforeEach(() => {
    vi.useFakeTimers();
-    vi.spyOn(api, "get").mockResolvedValueOnce([pendingApproval("a1")]);
-    vi.spyOn(api, "post").mockResolvedValue({});
+    mockGet = vi.spyOn(api, "get").mockResolvedValueOnce([pendingApproval("a1")]);
+    mockPost = vi.spyOn(api, "post").mockResolvedValue({});
  });

  afterEach(() => {
@@ -196,7 +197,7 @@ describe("ApprovalBanner — decisions", () => {
  });

  it("shows an error toast when POST fails", async () => {
-    vi.mocked(api.post).mockRejectedValueOnce(new Error("Network error"));
+    mockPost.mockReset().mockRejectedValue(new Error("Network error"));
    render(<ApprovalBanner />);
    await act(async () => { await vi.runOnlyPendingTimersAsync(); });
    fireEvent.click(screen.getAllByRole("button", { name: /approve/i })[0]);
@@ -208,8 +209,9 @@ describe("ApprovalBanner — decisions", () => {
  });

  it("keeps the card visible when the POST fails", async () => {
-    // Use mockRejectedValueOnce on the same spy as beforeEach (don't call spyOn again)
-    vi.mocked(api.post).mockRejectedValueOnce(new Error("Network error"));
+    // Reset the post mock before rejecting so the beforeEach's resolved value
+    // is gone and we get a clean rejection instead of a resolved→rejected queue.
+    mockPost.mockReset().mockRejectedValue(new Error("Network error"));
    render(<ApprovalBanner />);
    await act(async () => { await vi.runOnlyPendingTimersAsync(); });
    fireEvent.click(screen.getAllByRole("button", { name: /approve/i })[0]);
@@ -0,0 +1,291 @@
+// @vitest-environment jsdom
+/**
+ * Toolbar tests.
+ *
+ * Covers:
+ *   - Renders with 0 workspaces
+ *   - Shows online/offline/failed/provisioning status pills when nodes exist
+ *   - WebSocket status pill: connected → "Live"
+ *   - WebSocket status pill: connecting → "Reconnecting"
+ *   - WebSocket status pill: disconnected → "Offline"
+ *   - Stop All button visible when activeTasks > 0
+ *   - Restart Pending button visible when needsRestart nodes exist
+ *   - Help button opens the help popover
+ *   - Help popover closes on Escape or pointer-outside
+ *   - KeyboardShortcutsDialog opens via ? shortcut (when not in input)
+ */
+import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
+import { render, screen, fireEvent, cleanup } from "@testing-library/react";
+import React from "react";
+
+afterEach(() => {
+  cleanup();
+  vi.clearAllMocks();
+});
+
+// Reset store state between tests so mutations don't leak.
+beforeEach(() => {
+  defaultStore.nodes = [];
+  defaultStore.wsStatus = "connected";
+  defaultStore.showA2AEdges = false;
+  defaultStore.selectedNodeId = null;
+  mockSetShowA2AEdges.mockClear();
+  mockSetPanelTab.mockClear();
+  mockSetSearchOpen.mockClear();
+  mockUpdateNodeData.mockClear();
+});
+
+// ── Mock targets ───────────────────────────────────────────────────────────────
+
+vi.mock("@/components/Toaster", () => ({
+  showToast: vi.fn(),
+}));
+
+vi.mock("@/components/ConfirmDialog", () => ({
+  ConfirmDialog: () => null,
+}));
+
+vi.mock("@/components/settings/SettingsButton", () => ({
+  SettingsButton: () => null,
+}));
+
+vi.mock("@/components/settings/SettingsPanel", () => ({
+  settingsGearRef: { current: null },
+}));
+
+vi.mock("@/components/ThemeToggle", () => ({
+  ThemeToggle: () => null,
+}));
+
+vi.mock("@/components/KeyboardShortcutsDialog", () => ({
+  KeyboardShortcutsDialog: ({ open }: { open: boolean; onClose: () => void }) =>
+    open ? <div role="dialog" data-testid="shortcuts-dialog">Shortcuts</div> : null,
+}));
+
+vi.mock("@/lib/design-tokens", () => ({
+  statusDotClass: (status: string) => {
+    const map: Record<string, string> = {
+      online: "bg-emerald-400",
+      offline: "bg-zinc-500",
+      paused: "bg-indigo-400",
+      degraded: "bg-amber-400",
+      failed: "bg-red-400",
+      provisioning: "bg-sky-400",
+    };
+    return map[status] ?? "bg-zinc-500";
+  },
+}));
+
+vi.mock("@/lib/api", () => ({
+  api: {
+    post: vi.fn(() => Promise.resolve()),
+  },
+}));
+
+// ── Store mocks ────────────────────────────────────────────────────────────────
+
+const mockSetShowA2AEdges = vi.fn();
+const mockSetPanelTab = vi.fn();
+const mockSetSearchOpen = vi.fn();
+const mockUpdateNodeData = vi.fn();
+
+const makeNodes = (
+  statuses: Array<"online" | "offline" | "failed" | "provisioning">,
+  activeTasks: number[] = [],
+  needsRestart: boolean[] = [],
+  parentIds: (string | null)[] = []
+) => {
+  return statuses.map((status, i) => ({
+    id: `ws-${i}`,
+    data: {
+      name: `Workspace ${i}`,
+      role: "agent",
+      tier: 1,
+      status,
+      parentId: parentIds[i] ?? null,
+      activeTasks: activeTasks[i] ?? 0,
+      needsRestart: needsRestart[i] ?? false,
+    },
+  }));
+};
+
+// Nodes must be React Flow nodes (id + data), but Toolbar only reads data fields.
+// makeNodes returns { id, data: { activeTasks, needsRestart, ... } }.
+const toStoreNodes = (nodes: ReturnType<typeof makeNodes>) =>
+  nodes.map((n) => ({ id: n.id, data: n.data }));
+
+const defaultStore = {
+  nodes: [] as ReturnType<typeof makeNodes>,
+  wsStatus: "connected" as "connected" | "connecting" | "disconnected",
+  showA2AEdges: false,
+  selectedNodeId: null as string | null,
+  sidePanelWidth: 480,
+  setShowA2AEdges: mockSetShowA2AEdges,
+  setPanelTab: mockSetPanelTab,
+  setSearchOpen: mockSetSearchOpen,
+  updateNodeData: mockUpdateNodeData,
+  selectedNodeIds: new Set<string>(),
+  clearSelection: vi.fn(),
+  batchRestart: vi.fn(() => Promise.resolve()),
+  batchPause: vi.fn(() => Promise.resolve()),
+  batchDelete: vi.fn(() => Promise.resolve()),
+};
+
+vi.mock("@/store/canvas", () => ({
+  useCanvasStore: vi.fn((selector: (s: typeof defaultStore) => unknown) =>
+    selector(defaultStore)
+  ),
+}));
+
+// ── Component under test ───────────────────────────────────────────────────────
+import { Toolbar } from "../Toolbar";
+
+// ── Tests ─────────────────────────────────────────────────────────────────────
+
+describe("Toolbar — workspace count display", () => {
+  it("shows '0 workspaces' when the canvas is empty", () => {
+    render(<Toolbar />);
+    expect(screen.getByText(/0 workspaces?/)).toBeTruthy();
+  });
+
+  it("shows 'N workspaces' when nodes exist", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online", "online"]));
+    render(<Toolbar />);
+    expect(screen.getByText(/2 workspaces?/)).toBeTruthy();
+  });
+});
+
+describe("Toolbar — status pills", () => {
+  it("shows the online pill when nodes are online", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online"]));
+    render(<Toolbar />);
+    // StatusPill uses aria-label
+    expect(screen.getByLabelText(/1 online/i)).toBeTruthy();
+  });
+
+  it("shows the offline pill only when offline nodes exist", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["offline"]));
+    render(<Toolbar />);
+    expect(screen.getByLabelText(/1 offline/i)).toBeTruthy();
+  });
+
+  it("shows the failed pill when failed nodes exist", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["failed"]));
+    render(<Toolbar />);
+    expect(screen.getByLabelText(/1 failed/i)).toBeTruthy();
+  });
+
+  it("shows the provisioning pill when provisioning nodes exist", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["provisioning"]));
+    render(<Toolbar />);
+    expect(screen.getByLabelText(/1 starting/i)).toBeTruthy();
+  });
+
+  it("suppresses offline pill when no offline nodes", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online", "online"]));
+    render(<Toolbar />);
+    expect(screen.queryByLabelText(/offline/i)).toBeNull();
+  });
+});
+
+describe("Toolbar — WebSocket status pill", () => {
+  it('shows "Live" when connected', () => {
+    defaultStore.wsStatus = "connected";
+    render(<Toolbar />);
+    expect(screen.getByText("Live")).toBeTruthy();
+  });
+
+  it('shows "Reconnecting" when connecting', () => {
+    defaultStore.wsStatus = "connecting";
+    render(<Toolbar />);
+    expect(screen.getByText("Reconnecting")).toBeTruthy();
+  });
+
+  it('shows "Offline" when disconnected', () => {
+    defaultStore.wsStatus = "disconnected";
+    render(<Toolbar />);
+    expect(screen.getByText("Offline")).toBeTruthy();
+  });
+});
+
+describe("Toolbar — Stop All", () => {
+  it("is hidden when no active tasks", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online"], [0]));
+    render(<Toolbar />);
+    expect(screen.queryByRole("button", { name: /Stop All/i })).toBeNull();
+  });
+
+  it("is visible when active tasks > 0", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online", "online"], [2, 2]));
+    render(<Toolbar />);
+    // aria-label: "Stop all running tasks (2)"
+    expect(screen.getByRole("button", { name: /stop all running tasks/i })).toBeTruthy();
+  });
+});
+
+describe("Toolbar — Restart Pending", () => {
+  it("is hidden when no nodes need restart", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online"], [], [false]));
+    render(<Toolbar />);
+    expect(screen.queryByRole("button", { name: /Restart Pending/i })).toBeNull();
+  });
+
+  it("is visible when nodes need restart", () => {
+    defaultStore.nodes = toStoreNodes(makeNodes(["online"], [], [true]));
+    render(<Toolbar />);
+    // aria-label: "Restart 1 workspaces pending config or secret changes"
+    expect(screen.getByRole("button", { name: /restart 1 workspace/i })).toBeTruthy();
+  });
+});
+
+describe("Toolbar — Help popover", () => {
+  it("opens when help button is clicked", () => {
+    render(<Toolbar />);
+    const helpBtn = screen.getByRole("button", { name: /open shortcuts and tips/i });
+    fireEvent.click(helpBtn);
+    expect(screen.getByRole("dialog")).toBeTruthy();
+  });
+
+  it("closes when close button is clicked", () => {
+    render(<Toolbar />);
+    const helpBtn = screen.getByRole("button", { name: /open shortcuts and tips/i });
+    fireEvent.click(helpBtn);
+    expect(screen.getByRole("dialog")).toBeTruthy();
+    const closeBtn = screen.getByRole("button", { name: /close help dialog/i });
+    fireEvent.click(closeBtn);
+    expect(screen.queryByRole("dialog")).toBeNull();
+  });
+});
+
+describe("Toolbar — A2A edges toggle", () => {
+  it("calls setShowA2AEdges on click", () => {
+    defaultStore.showA2AEdges = false;
+    render(<Toolbar />);
+    const toggle = screen.getByRole("button", { name: /show a2a edges/i });
+    fireEvent.click(toggle);
+    expect(mockSetShowA2AEdges).toHaveBeenCalledWith(true);
+  });
+});
+
+describe("Toolbar — ? shortcut opens shortcuts dialog", () => {
+  it("opens KeyboardShortcutsDialog when ? is pressed outside an input", () => {
+    render(<Toolbar />);
+    expect(screen.queryByTestId("shortcuts-dialog")).toBeNull();
+    fireEvent.keyDown(window, { key: "?" });
+    expect(screen.getByTestId("shortcuts-dialog")).toBeTruthy();
+  });
+
+  it("does not fire ? shortcut when focus is in an input", () => {
+    render(
+      <div>
+        <input data-testid="test-input" type="text" />
+        <Toolbar />
+      </div>
+    );
+    const input = screen.getByTestId("test-input");
+    fireEvent.focus(input);
+    // Fire on the input element so e.target.tagName === "INPUT" is true
+    fireEvent.keyDown(input, { key: "?" });
+    expect(screen.queryByTestId("shortcuts-dialog")).toBeNull();
+  });
+});
@@ -55,10 +55,10 @@ describe("statusDotClass", () => {

 describe("TIER_CONFIG", () => {
  it("has entries for all four tier levels", () => {
-    expect(TIER_CONFIG).toHaveProperty(1);
-    expect(TIER_CONFIG).toHaveProperty(2);
-    expect(TIER_CONFIG).toHaveProperty(3);
-    expect(TIER_CONFIG).toHaveProperty(4);
+    expect(TIER_CONFIG).toHaveProperty("1");
+    expect(TIER_CONFIG).toHaveProperty("2");
+    expect(TIER_CONFIG).toHaveProperty("3");
+    expect(TIER_CONFIG).toHaveProperty("4");
  });

  it("each tier has label, color, and border fields", () => {
@@ -2,7 +2,7 @@

 How a workspace-server code change reaches the prod tenant fleet — and how to stop it if something's wrong.

-> **⚠️ State note (2026-04-22):** this doc describes the **intended design**. As of this write, the canary fleet described below is **not actually running** — no canary tenants are provisioned, `CANARY_TENANT_URLS` / `CANARY_ADMIN_TOKENS` / `CANARY_CP_SHARED_SECRET` are empty in repo secrets, and `canary-verify.yml` fails every run.
+> **⚠️ State note (2026-04-22, secret names refreshed 2026-05-11):** this doc describes the **intended design**. As of this write, the canary fleet described below is **not actually running** — no canary tenants are provisioned, `MOLECULE_STAGING_TENANT_URLS` / `MOLECULE_STAGING_ADMIN_TOKENS` / `MOLECULE_STAGING_CP_SHARED_SECRET` are empty in repo secrets, and `staging-verify.yml` (formerly `canary-verify.yml`) fails every run.
 >
 > Current merges gate on manual `promote-latest.yml` dispatches, not canary. See [molecule-controlplane/docs/canary-tenants.md](https://git.moleculesai.app/molecule-ai/molecule-controlplane/src/branch/main/docs/canary-tenants.md) for the Phase 1 code work that's already shipped + the Phase 2 plan for actually standing up the fleet + a "should we even do this now?" decision framework.
 >
@@ -22,7 +22,7 @@ publish-workspace-server-image.yml   ← pushes :staging-<sha> ONLY
 Canary tenants auto-update to :staging-<sha>
      │   (5-min auto-updater cycle on each canary EC2)
      ▼
-canary-verify.yml waits 6 min, runs scripts/canary-smoke.sh
+staging-verify.yml waits 6 min, runs scripts/staging-smoke.sh
      │
      ├─► GREEN → crane tag :staging-<sha> → :latest
      │                                       │
@@ -42,7 +42,7 @@ Canary tenants are configured to pull `:staging-<sha>` (not `:latest`) via `TENA

 ## Smoke suite

-`scripts/canary-smoke.sh` hits each canary tenant (URL + ADMIN_TOKEN pair) and asserts:
+`scripts/staging-smoke.sh` hits each canary tenant (URL + ADMIN_TOKEN pair) and asserts:

 - `/admin/liveness` returns a subsystems map (tenant booted, AdminAuth reachable)
 - `/workspaces` returns a JSON array (wsAuth + DB healthy)
@@ -59,8 +59,8 @@ Expand by editing the script — each `check "name" "expected" "$response"` call
 3. Re-trigger provision (or delete + recreate if the org was already provisioned into staging) — the fresh EC2 lands in the canary AWS account (see internal runbook for the specific ID)

 Then set repo secrets:
- `CANARY_TENANT_URLS` — append the new tenant's URL
- `CANARY_ADMIN_TOKENS` — append its ADMIN_TOKEN in the same position
+- `MOLECULE_STAGING_TENANT_URLS` — append the new tenant's URL
+- `MOLECULE_STAGING_ADMIN_TOKENS` — append its ADMIN_TOKEN in the same position

 ## Rolling back `:latest`

@@ -50,7 +50,7 @@ pipeline.
 | `check-merge-group-trigger.yml` | The workflow's own header (lines 18-23) documents that it's vacuously satisfied on Gitea — Gitea has no merge queue, no `merge_group:` event type, no `gh-readonly-queue/...` refs. Nothing to lint. |
 | `codeql.yml` | The workflow's own header (lines 3-67) documents that `github/codeql-action/init@v4` hits api.github.com bundle endpoints not implemented by Gitea (observed: `::error::404 page not found` in Initialize CodeQL step). Per Hongming decision 2026-05-07 (task #156): CodeQL is ADVISORY/non-blocking until a Gitea-compatible SAST pipeline lands. Replacement options (Semgrep self-host, Sonatype, GitHub-mirror-for-SAST) tracked in #156. |
 | `pr-guards.yml` | The workflow's own header documents that Gitea has no `gh pr merge --auto` primitive — the guard is a structural no-op on Gitea. Branch protection on `main` does NOT reference any `pr-guards` check name; deletion is safe. |
-| `promote-latest.yml` | Uses `imjasonh/setup-crane` against `ghcr.io/molecule-ai/platform` — the GHCR registry was retired during the 2026-05-06 Gitea migration (per `canary-verify.yml` header notes, the canonical tenant image moved to ECR `153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/platform-tenant`). The workflow can no longer find any image to retag. Follow-up issue suggested if an ECR-based retag promote is desired. |
+| `promote-latest.yml` | Uses `imjasonh/setup-crane` against `ghcr.io/molecule-ai/platform` — the GHCR registry was retired during the 2026-05-06 Gitea migration (per `staging-verify.yml` header notes — file was renamed from `canary-verify.yml` on 2026-05-11; the canonical tenant image moved to ECR `153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/platform-tenant`). The workflow can no longer find any image to retag. Follow-up issue suggested if an ECR-based retag promote is desired. |

 ## Category C — ported to .gitea/

@@ -0,0 +1,150 @@
+# Gitea Actions operational quirks (molecule-core)
+
+Documents persistent operational findings about Gitea Actions runner behaviour
+that differ from GitHub Actions and require workarounds in workflow YAML or
+runbooks.
+
+> Last updated: 2026-05-11 (core-devops-agent)
+
+---
+
+## Gitea 1.22.6 runner network isolation
+
+### Finding
+
+The Gitea Actions runner (container on host `5.78.80.188`) cannot reach the
+git remote (`https://git.moleculesai.app`) over HTTPS from inside the runner
+container. Any `git fetch`, `git clone`, or `git push` command that contacts
+the remote times out at 12–15 s.
+
+This is **not a Gitea Actions bug** — it is an operator-level network policy
+where the runner container's network namespace is restricted from reaching the
+Gitea host HTTPS endpoint. The runner can reach external hosts (GitHub,
+Docker Hub, PyPI) normally.
+
+### Impact
+
+Workflows that rely on `git fetch origin <ref>` or `actions/checkout` with
+`fetch-depth: 0` (full history) will hang or time out.
+
+Specifically:
+- `actions/checkout@v*` with `fetch-depth: 0` hangs (fetching full repo
+  history takes >30 s before hitting the timeout).
+- `git fetch origin main --depth=1` times out at ~15 s.
+- `git clone <url>` times out at ~15 s.
+
+### Affected workflows
+
+| Workflow | Issue | Workaround |
+|---|---|---|
+| `harness-replays.yml` detect-changes job | `git fetch origin main --depth=1` times out | Added `timeout 20` + graceful fallback to `run=true` (always run harness) per PR #441 |
+| `publish-workspace-server-image.yml` | In-image `git clone` of workspace templates | Pre-clone manifest deps before compose build (Task #173 pattern) |
+| Any workflow using `fetch-depth: 0` | Full history fetch times out | Use `fetch-depth: 1` + explicit `git fetch` for needed refs |
+
+### How to diagnose
+
+```bash
+# From inside the runner (add as a debug step):
+timeout 20 git fetch origin main --depth=1
+# If this times out: runner cannot reach git remote
+```
+
+### Verification
+
+Confirmed 2026-05-11 by running `timeout 20 git fetch origin main --depth=1`
+in the `detect-changes` job of `harness-replays.yml` — consistently times
+out at 15 s. Runner can reach `https://api.github.com` and `https://pypi.org`
+without issue.
+
+### References
+
+- PR #441: fix for `harness-replays.yml` detect-changes
+- Task #173: pre-clone manifest deps pattern for compose build
+- internal#102: tracking customer-private + marketplace third-party repos
+- `feedback_oss_first_repo_visibility_default`: 5 workspace-template repos
+  flipped public to allow pre-clone without auth
+
+---
+
+## `continue-on-error` only works at step level, not job level
+
+### Finding
+
+Gitea Actions (1.22.6) does not honour `continue-on-error: true` at the **job**
+level the way GitHub Actions does. A job with `continue-on-error: true` that
+fails still reports `status: failure` in the commit status API.
+
+Only `continue-on-error: true` at the **step** level works as expected.
+
+### Impact
+
+If you want a job to always "pass" in the status API (so dependent jobs can
+run and the overall CI does not show `failure`), you must add
+`continue-on-error: true` to every step that can fail, AND ensure each step
+exits with code 0 (e.g., append `|| true` to commands that might fail).
+
+### Affected workflows
+
+| Workflow | Fix |
+|---|---|
+| `harness-replays.yml` detect-changes | Added `continue-on-error: true` to fetch step + decide step; added `|| true` to `DIFF=$(git diff ...)` per PR #441 |
+
+### How to diagnose
+
+```yaml
+# WRONG — job reports as failure despite flag
+jobs:
+  my-job:
+    continue-on-error: true   # ← ignored by Gitea
+    steps:
+      - run: git diff ...    # ← if this fails, job = failure
+        # job-level flag does not help
+
+# RIGHT — step-level flag prevents step from failing
+jobs:
+  my-job:
+    steps:
+      - run: git diff ... || true  # ← step exits 0
+        continue-on-error: true     # ← belt and suspenders
+```
+
+### References
+
+- Gitea Actions quirk #10 (from migration checklist)
+- PR #441: fix applied to `harness-replays.yml`
+
+---
+
+## `workflow_dispatch.inputs` not supported
+
+Gitea 1.22.6 parser rejects `workflow_dispatch.inputs`. Drop from all workflow
+YAML files ported from GitHub Actions. Manual triggers should use
+`workflow_dispatch` without `inputs:`.
+
+**Reference**: `feedback_gitea_workflow_dispatch_inputs_unsupported`
+
+---
+
+## `merge_group` not supported
+
+Gitea has no merge queue concept. Drop `merge_group:` triggers from all
+workflow YAML files.
+
+---
+
+## `environment:` blocks not supported
+
+Gitea has no environments concept. Drop `environment:` from all workflow YAML
+files. Secrets and variables are repo-level.
+
+---
+
+## `fetch-depth: 0` on `actions/checkout` times out
+
+`actions/checkout` with `fetch-depth: 0` triggers a full repo history fetch
+which exceeds the runner's network timeout to the git remote (~15 s).
+
+**Workaround**: Use `fetch-depth: 1` (default) and add explicit
+`git fetch origin <ref> --depth=1` for any additional refs needed.
+
+**Reference**: PR #441 detect-changes fetch step.
@@ -43,7 +43,7 @@ endpoint handler for the supported range.
 - `cleanup-rogue-workspaces.sh` — emergency teardown for leaked
  workspaces. Prompts for confirmation. Pair with the harnesses if a
  cleanup trap fails (see `cleanup_*_failed` events).
- `canary-smoke.sh` — quick smoke test for canary releases.
+- `staging-smoke.sh` — quick smoke test for the staging canary fleet (formerly `canary-smoke.sh`).
 - `dev-start.sh` — local-dev platform bring-up.

 The rest are self-documenting in their header comments.
@@ -1,29 +1,40 @@
 #!/bin/bash
-# canary-smoke.sh — runs the post-deploy smoke suite against the
-# staging canary tenant fleet. Called by the canary-verify.yml GitHub
+# staging-smoke.sh — runs the post-deploy smoke suite against the
+# staging canary tenant fleet. Called by the staging-verify.yml Gitea
 # Actions workflow after a new workspace-server image lands in ECR;
 # exits non-zero on any failure so the workflow can block the
 # redeploy-fleet promotion that would otherwise release broken code
 # to the prod tenant fleet.
 #
+# Naming note (2026-05-11): The script (and its input env vars) were
+# renamed from canary-smoke.sh / CANARY_* to staging-smoke.sh /
+# MOLECULE_STAGING_* per Hongming directive. The tested COHORT is still
+# called the "canary fleet" (a small subset of staging tenants that
+# ingest :staging-<sha> before the rest of the fleet); that strategy
+# concept is unchanged.
+#
 # Registry note: GHCR was retired 2026-05-06. Images are now pushed
 # to the operator's ECR org (153263036946.dkr.ecr.us-east-2.amazonaws.com/
 # molecule-ai/platform-tenant). The registry URL is a runtime concern for
 # the CI push step; this script tests the running tenant directly.
 #
 # Environment:
-#   CANARY_TENANT_URLS       space-sep list of canary tenant base URLs
-#                            (e.g. "https://canary-pm.staging.moleculesai.app
-#                                   https://canary-mcp.staging.moleculesai.app")
-#   CANARY_ADMIN_TOKENS      space-sep list of ADMIN_TOKENs, positionally
-#                            matched to CANARY_TENANT_URLS. Canary tenants
-#                            are provisioned with known ADMIN_TOKENs so CI
-#                            can hit their admin-gated endpoints.
-#   CANARY_CP_BASE_URL       CP base URL the canaries call back to
-#                            (https://staging-api.moleculesai.app)
-#   CANARY_CP_SHARED_SECRET  matches CP's PROVISION_SHARED_SECRET so this
-#                            script can also exercise /cp/workspaces/* via
-#                            the canary's own CPProvisioner identity.
+#   MOLECULE_STAGING_TENANT_URLS       space-sep list of canary tenant base
+#                                       URLs (e.g. "https://canary-pm.staging.
+#                                       moleculesai.app https://canary-mcp.
+#                                       staging.moleculesai.app")
+#   MOLECULE_STAGING_ADMIN_TOKENS      space-sep list of ADMIN_TOKENs,
+#                                       positionally matched to
+#                                       MOLECULE_STAGING_TENANT_URLS.
+#                                       Canary tenants are provisioned with
+#                                       known ADMIN_TOKENs so CI can hit
+#                                       their admin-gated endpoints.
+#   MOLECULE_STAGING_CP_BASE_URL       CP base URL the canaries call back to
+#                                       (https://staging-api.moleculesai.app)
+#   MOLECULE_STAGING_CP_SHARED_SECRET  matches CP's PROVISION_SHARED_SECRET
+#                                       so this script can also exercise
+#                                       /cp/workspaces/* via the canary's
+#                                       own CPProvisioner identity.
 #
 # Exit codes: 0 = all green, 1 = assertion failure, 2 = setup/env problem.

@@ -31,12 +42,12 @@ set -euo pipefail

 # ── Setup ────────────────────────────────────────────────────────────────

-: "${CANARY_TENANT_URLS:?space-sep list of canary base URLs required}"
-: "${CANARY_ADMIN_TOKENS:?space-sep list of ADMIN_TOKENs required, same order as URLs}"
-: "${CANARY_CP_BASE_URL:?CP base URL required}"
+: "${MOLECULE_STAGING_TENANT_URLS:?space-sep list of canary base URLs required}"
+: "${MOLECULE_STAGING_ADMIN_TOKENS:?space-sep list of ADMIN_TOKENs required, same order as URLs}"
+: "${MOLECULE_STAGING_CP_BASE_URL:?CP base URL required}"

-read -r -a URLS <<< "$CANARY_TENANT_URLS"
-read -r -a TOKENS <<< "$CANARY_ADMIN_TOKENS"
+read -r -a URLS <<< "$MOLECULE_STAGING_TENANT_URLS"
+read -r -a TOKENS <<< "$MOLECULE_STAGING_ADMIN_TOKENS"

 if [ "${#URLS[@]}" -ne "${#TOKENS[@]}" ]; then
  echo "ERROR: URLS(${#URLS[@]}) and TOKENS(${#TOKENS[@]}) length mismatch" >&2
@@ -69,7 +80,7 @@ check() {
 # tenant never gets the wrong token.
 acurl() {
  local base="$1" token="$2"; shift 2
-  curl -sS --max-time 20 -H "Authorization: Bearer $token" "$@" -- "$base${CANARY_ACURL_PATH:-}"
+  curl -sS --max-time 20 -H "Authorization: Bearer $token" "$@" -- "$base${ACURL_PATH:-}"
 }

 # ── Checks (run per canary tenant) ───────────────────────────────────────
@@ -80,7 +91,7 @@ for i in "${!URLS[@]}"; do
  printf "\n── %s ──\n" "$base"

  # 1. Liveness — the tenant is up and responding to admin auth.
-  CANARY_ACURL_PATH="/admin/liveness" resp=$(acurl "$base" "$token" || true)
+  ACURL_PATH="/admin/liveness" resp=$(acurl "$base" "$token" || true)
  check "liveness returns a subsystems map" '"subsystems"' "$resp"

  # 2. CP env refresh — the workspace-server fetched MOLECULE_CP_SHARED_SECRET
@@ -89,25 +100,25 @@ for i in "${!URLS[@]}"; do
  # booted without crashing on the refresh call. A startup failure in
  # refreshEnvFromCP logs but still boots (best-effort semantics), so
  # this is a sanity check, not a proof.
-  CANARY_ACURL_PATH="/workspaces" resp=$(acurl "$base" "$token" || true)
+  ACURL_PATH="/workspaces" resp=$(acurl "$base" "$token" || true)
  check "workspace list is JSON array" "[" "$resp"

  # 3. Memory commit round-trip — scope=LOCAL so test data stays on this
  # tenant. Verifies encryption + scrubber + retrieval end-to-end.
  probe_id="canary-smoke-$(date +%s)-$i"
  body=$(printf '{"scope":"LOCAL","namespace":"canary-smoke","content":"probe-%s"}' "$probe_id")
-  CANARY_ACURL_PATH="/memories/commit" resp=$(curl -sS --max-time 20 \
+  ACURL_PATH="/memories/commit" resp=$(curl -sS --max-time 20 \
    -X POST -H "Content-Type: application/json" -H "Authorization: Bearer $token" \
    --data "$body" "$base/memories/commit" || true)
  check "memory commit accepted" '"id"' "$resp"

-  CANARY_ACURL_PATH="/memories/search?query=probe-${probe_id}" \
+  ACURL_PATH="/memories/search?query=probe-${probe_id}" \
    resp=$(curl -sS --max-time 20 -H "Authorization: Bearer $token" \
    "$base/memories/search?query=probe-${probe_id}" || true)
  check "memory search finds the probe" "probe-${probe_id}" "$resp"

  # 4. Events admin read — AdminAuth path (C4 fail-closed proof on SaaS).
-  CANARY_ACURL_PATH="/events" resp=$(acurl "$base" "$token" || true)
+  ACURL_PATH="/events" resp=$(acurl "$base" "$token" || true)
  check "events endpoint returns JSON" "[" "$resp"

  # 5. Negative: unauth'd admin call must 401 (C4 regression gate).
@@ -117,7 +128,7 @@ for i in "${!URLS[@]}"; do
  # 6. POST /org/import unauth → 401. Proves the route is compiled in
  # and AdminAuth is enforced. A missing route returns 404 (the failure
  # mode caught by issue #213). Regression guard for the silent-GHCR-
-  # migration gap: canary-verify was testing a stale GHCR image while
+  # migration gap: staging-verify (formerly canary-verify) was testing a stale GHCR image while
  # actual tenants ran ECR — this test would have caught a missing-route
  # binary before it reached prod.
  unauth_code=$(curl -sS -o /dev/null -w '%{http_code}' \
@@ -7,11 +7,11 @@ Four workflows + a shared bash harness that together cover the SaaS stack end to
 | Workflow | Cadence | Wall time | Scope |
 |---|---|---|---|
 | `e2e-staging-saas.yml` | push + nightly 07:00 UTC | ~20 min | Full API: org → tenant → 2 workspaces → A2A → HMA → delegation → leak check |
-| `canary-staging.yml` | every 30 min | ~8 min | Minimum smoke + self-managed alert issue |
+| `staging-smoke.yml` | every 30 min | ~8 min | Minimum smoke + self-managed alert issue |
 | `e2e-staging-canvas.yml` | push + weekly Sunday 08:00 | ~25 min | All 13 canvas workspace-panel tabs via Playwright |
 | `e2e-staging-sanity.yml` | weekly Monday 06:00 | ~10 min | Intentional-failure: teardown safety-net self-check |

-`tests/e2e/test_staging_full_saas.sh` is the shared harness all workflows invoke (with `E2E_MODE={full|canary}` and `E2E_INTENTIONAL_FAILURE={0|1}` toggles).
+`tests/e2e/test_staging_full_saas.sh` is the shared harness all workflows invoke (with `E2E_MODE={full|smoke}` and `E2E_INTENTIONAL_FAILURE={0|1}` toggles).

 ### Full-SaaS checklist (sections)

@@ -49,7 +49,15 @@ Runs the harness with `E2E_INTENTIONAL_FAILURE=1`, which poisons the tenant admi

 Set in **Settings → Secrets and variables → Actions → Repository secrets**:

-### `MOLECULE_STAGING_ADMIN_TOKEN`
+### `CP_STAGING_ADMIN_API_TOKEN`
+
+> **Historical-rename note (2026-05-11):** previously named
+> `MOLECULE_STAGING_ADMIN_TOKEN`. Canonicalised to
+> `CP_STAGING_ADMIN_API_TOKEN` per internal#322 (the Railway staging
+> service exposes it as `CP_ADMIN_API_TOKEN`; the `CP_*` repo-secret
+> prefix matches the upstream env name + makes the service it talks
+> to obvious in workflow YAMLs). See the original PR for the
+> cross-workflow sweep.

 The `CP_ADMIN_API_TOKEN` env currently set on the Railway staging molecule-platform → controlplane service.

@@ -82,7 +90,7 @@ bash tests/e2e/test_staging_full_saas.sh
 ## Cost

 - Full run: ~20 min, ~$0.007
- Canary (48/day): ~$0.06/day
+- Smoke (48/day): ~$0.06/day
 - Canvas (few/week): ~$0.01/day
 - Sanity (weekly): ~$0.002/week
 - **Total staging burn: < $0.15/day** at expected CI load
@@ -27,7 +27,11 @@
 #   E2E_PROVISION_TIMEOUT_SECS   default 900 (15 min cold EC2 budget)
 #   E2E_KEEP_ORG                 1 → skip teardown (debugging only)
 #   E2E_RUN_ID                   Slug suffix; CI: ${GITHUB_RUN_ID}
-#   E2E_MODE                     full (default) | canary
+#   E2E_MODE                     full (default) | smoke
+#                                (legacy alias `canary` still accepted —
+#                                 mapped to `smoke` for back-compat with
+#                                 any in-flight runner picking up an older
+#                                 workflow checkout)
 #   E2E_INTENTIONAL_FAILURE      1 → poison tenant token mid-run so the
 #                                script fails; the EXIT trap MUST still
 #                                tear down cleanly (and exit 4 on leak).
@@ -49,15 +53,23 @@ RUNTIME="${E2E_RUNTIME:-hermes}"
 PROVISION_TIMEOUT_SECS="${E2E_PROVISION_TIMEOUT_SECS:-900}"
 RUN_ID_SUFFIX="${E2E_RUN_ID:-$(date +%H%M%S)-$$}"
 MODE="${E2E_MODE:-full}"
+# `canary` is a legacy alias for `smoke` retained for back-compat with
+# any in-flight runner picking up an older workflow checkout during the
+# 2026-05-11 canary→staging rename rollout. Both map to the same slug
+# prefix below. Remove the `canary` alias after one week of no-old-mode
+# observations.
+if [ "$MODE" = "canary" ]; then
+  MODE="smoke"
+fi
 case "$MODE" in
-  full|canary) ;;
-  *) echo "E2E_MODE must be 'full' or 'canary' (got: $MODE)" >&2; exit 2 ;;
+  full|smoke) ;;
+  *) echo "E2E_MODE must be 'full' or 'smoke' (got: $MODE)" >&2; exit 2 ;;
 esac

-# Canary runs get a distinct prefix so their safety-net sweeper only
+# Smoke runs get a distinct slug prefix so their safety-net sweeper only
 # touches their own runs, not in-flight full runs.
-if [ "$MODE" = "canary" ]; then
-  SLUG="e2e-canary-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
+if [ "$MODE" = "smoke" ]; then
+  SLUG="e2e-smoke-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
 else
  SLUG="e2e-$(date +%Y%m%d)-${RUN_ID_SUFFIX}"
 fi
@@ -25,10 +25,10 @@ _WORKSPACE_ID_raw = os.environ.get("WORKSPACE_ID")
 if not _WORKSPACE_ID_raw:
    raise RuntimeError("WORKSPACE_ID environment variable is required but not set")
 WORKSPACE_ID = _WORKSPACE_ID_raw
-if os.path.exists("/.dockerenv") or os.environ.get("DOCKER_VERSION"):
-    PLATFORM_URL = os.environ.get("PLATFORM_URL", "http://host.docker.internal:8080")
-else:
-    PLATFORM_URL = os.environ.get("PLATFORM_URL", "http://localhost:8080")
+# Platform URL: always host.docker.internal inside containers. The platform API
+# is only reachable via the Docker network mesh from inside a workspace
+# container regardless of the runtime environment (Docker/host).
+PLATFORM_URL = os.environ.get("PLATFORM_URL", "http://host.docker.internal:8080")


 async def discover(target_id: str) -> dict | None:
@@ -26,10 +26,10 @@ _WORKSPACE_ID_raw = os.environ.get("WORKSPACE_ID")
 if not _WORKSPACE_ID_raw:
    raise RuntimeError("WORKSPACE_ID environment variable is required but not set")
 WORKSPACE_ID = _WORKSPACE_ID_raw
-if os.path.exists("/.dockerenv") or os.environ.get("DOCKER_VERSION"):
-    PLATFORM_URL = os.environ.get("PLATFORM_URL", "http://host.docker.internal:8080")
-else:
-    PLATFORM_URL = os.environ.get("PLATFORM_URL", "http://localhost:8080")
+# Platform URL: always host.docker.internal inside containers. The platform API
+# is only reachable via the Docker network mesh from inside a workspace
+# container regardless of the runtime environment (Docker/host).
+PLATFORM_URL = os.environ.get("PLATFORM_URL", "http://host.docker.internal:8080")

 # Cache workspace ID → name mappings (populated by list_peers calls)
 _peer_names: dict[str, str] = {}
@@ -54,6 +54,18 @@ import httpx

 logger = logging.getLogger(__name__)

+
+def _platform_url() -> str:
+    """Return the platform URL, defaulting to host.docker.internal.
+
+    The workspace runtime always runs inside a Docker container, so
+    ``localhost`` refers to the container itself, not the platform host.
+    The platform API is only reachable via ``host.docker.internal`` from
+    within a workspace container, regardless of how the container was started.
+    """
+    return os.environ.get("PLATFORM_URL", "http://host.docker.internal:8080")
+
+
 # ─────────────────────────────────────────────────────────────────────────────
 # Constants
 # ─────────────────────────────────────────────────────────────────────────────
@@ -79,12 +91,12 @@ async def _fetch_latest_checkpoint(workspace_id: str) -> Optional[dict]:
        workspace_id: The workspace to query.

    Reads:
-        PLATFORM_URL  Platform base URL (default ``http://localhost:8080``).
+        PLATFORM_URL  Platform base URL (default ``http://host.docker.internal:8080``).
    """
    try:
        from platform_auth import auth_headers as _auth_headers  # type: ignore[import]

-        platform_url = os.environ.get("PLATFORM_URL", "http://localhost:8080")
+        platform_url = _platform_url()
        url = f"{platform_url}/workspaces/{workspace_id}/checkpoints/latest"
        async with httpx.AsyncClient(timeout=5.0) as client:
            resp = await client.get(url, headers=_auth_headers())
@@ -125,12 +137,12 @@ async def _save_checkpoint(
        payload:       Optional JSON-serialisable dict stored as JSONB.

    Reads:
-        PLATFORM_URL   Platform base URL (default ``http://localhost:8080``).
+        PLATFORM_URL   Platform base URL (default ``http://host.docker.internal:8080``).
    """
    try:
        from platform_auth import auth_headers as _auth_headers  # type: ignore[import]

-        platform_url = os.environ.get("PLATFORM_URL", "http://localhost:8080")
+        platform_url = _platform_url()
        url = f"{platform_url}/workspaces/{workspace_id}/checkpoints"
        body: dict = {
            "workflow_id": workflow_id,
@@ -48,6 +48,27 @@ def get_machine_ip() -> str:  # pragma: no cover
        return "127.0.0.1"


+def _check_delegation_results_pending() -> bool:
+    """Check if there are unconsumed delegation results waiting.
+
+    Reads ``DELEGATION_RESULTS_FILE``.  Returns ``True`` if the file
+    exists and contains non-whitespace content (after stripping) — meaning
+    the idle loop should skip this tick.  Returns ``False`` if the file is
+    absent, empty, or contains only whitespace.
+
+    The extracted form lets unit tests call this directly rather than mirroring
+    the logic (anti-pattern flagged as #401).
+    """
+    from heartbeat import DELEGATION_RESULTS_FILE
+
+    try:
+        with open(DELEGATION_RESULTS_FILE) as rf:
+            rf.seek(0)
+            return bool(rf.read().strip())
+    except FileNotFoundError:
+        return False
+
+
 # Re-exported from transcript_auth for the inline /transcript handler.
 # Separate module keeps the security-critical gate import-light + unit-testable.
 from transcript_auth import transcript_authorized as _transcript_authorized
@@ -678,20 +699,15 @@ async def main():  # pragma: no cover
                # heartbeat's own self-message wake the agent after results are
                # written. The agent then sees the results in _prepare_prompt()
                # and processes them before composing.
-                from heartbeat import DELEGATION_RESULTS_FILE as _DRF
-                try:
-                    with open(_DRF) as _rf:
-                        _rf.seek(0)
-                        _content = _rf.read().strip()
-                    if _content:
-                        print(
-                            f"Idle loop: skipping — {len(_content)} bytes of unconsumed "
-                            f"delegation results pending (heartbeat will notify agent)",
-                            flush=True,
-                        )
-                        continue
-                except FileNotFoundError:
-                    pass  # No results file — normal, proceed with idle prompt
+                # Guard logic extracted to _check_delegation_results_pending() for
+                # direct unit-testing (#401 follow-up).
+                if _check_delegation_results_pending():
+                    print(
+                        "Idle loop: skipping — unconsumed delegation results pending "
+                        "(heartbeat will notify agent)",
+                        flush=True,
+                    )
+                    continue

                # Self-post the idle prompt via the platform A2A proxy (same
                # path as initial_prompt). The agent's own concurrency control
@@ -279,7 +279,7 @@ class TestToolDelegateTask:
             patch("a2a_tools.report_activity", new=AsyncMock()):
            result = await a2a_tools.tool_delegate_task("ws-1", "do something")

-        assert result == "Task completed!"
+        assert result == "[A2A_RESULT_FROM_PEER]\nTask completed!\n[/A2A_RESULT_FROM_PEER]"

    async def test_error_response_returns_delegation_failed_message(self):
        """When send_a2a_message returns _A2A_ERROR_PREFIX text, delegation fails."""
@@ -307,7 +307,7 @@ class TestToolDelegateTask:
             patch("a2a_tools.report_activity", new=AsyncMock()):
            result = await a2a_tools.tool_delegate_task("ws-cached", "task")

-        assert result == "done"
+        assert result == "[A2A_RESULT_FROM_PEER]\ndone\n[/A2A_RESULT_FROM_PEER]"

    async def test_peer_name_falls_back_to_id_prefix(self):
        """When peer has no name and cache is empty, name = first 8 chars of workspace_id."""
@@ -321,110 +321,11 @@ class TestToolDelegateTask:
             patch("a2a_tools.report_activity", new=AsyncMock()):
            result = await a2a_tools.tool_delegate_task("ws-nona000", "task")

-        assert result == "ok"
+        assert result == "[A2A_RESULT_FROM_PEER]\nok\n[/A2A_RESULT_FROM_PEER]"
        # Cache should now have been set
        assert a2a_tools._peer_names.get("ws-nona000") is not None


-# ---------------------------------------------------------------------------
-# delegate_task (non-tool, direct httpx path — used by adapter templates)
-# ---------------------------------------------------------------------------
-
-class TestDelegateTaskDirect:
-
-    async def test_string_form_error_returns_error_message(self):
-        """The A2A proxy can return {"error": "plain string"}. Must not raise
-        AttributeError: 'str' object has no attribute 'get'."""
-        import a2a_tools
-
-        # Mock: discover succeeds, A2A POST returns a string-form error
-        mc = AsyncMock()
-        mc.__aenter__ = AsyncMock(return_value=mc)
-        mc.__aexit__ = AsyncMock(return_value=False)
-
-        async def fake_post(url, **kwargs):
-            r = MagicMock()
-            r.status_code = 200
-            r.json = MagicMock(return_value={"error": "peer workspace unreachable"})
-            return r
-
-        async def fake_get(url, **kwargs):
-            r = MagicMock()
-            r.status_code = 200
-            r.json = MagicMock(return_value={"url": "http://peer.svc/a2a"})
-            return r
-
-        mc.post = fake_post
-        mc.get = fake_get
-
-        with patch("a2a_tools.httpx.AsyncClient", return_value=mc):
-            result = await a2a_tools.delegate_task("ws-peer-123", "do a thing")
-
-        assert "Error" in result
-        assert "peer workspace unreachable" in result
-
-    async def test_dict_form_error_returns_error_message(self):
-        """{"error": {"message": "...", "code": ...}} — the pre-existing path."""
-        import a2a_tools
-
-        mc = AsyncMock()
-        mc.__aenter__ = AsyncMock(return_value=mc)
-        mc.__aexit__ = AsyncMock(return_value=False)
-
-        async def fake_post(url, **kwargs):
-            r = MagicMock()
-            r.status_code = 200
-            r.json = MagicMock(return_value={"error": {"message": "internal server error", "code": 500}})
-            return r
-
-        async def fake_get(url, **kwargs):
-            r = MagicMock()
-            r.status_code = 200
-            r.json = MagicMock(return_value={"url": "http://peer.svc/a2a"})
-            return r
-
-        mc.post = fake_post
-        mc.get = fake_get
-
-        with patch("a2a_tools.httpx.AsyncClient", return_value=mc):
-            result = await a2a_tools.delegate_task("ws-peer-456", "do a thing")
-
-        assert "Error" in result
-        assert "internal server error" in result
-
-    async def test_success_returns_result_text(self):
-        """Happy path: result with parts returns the first text part."""
-        import a2a_tools
-
-        mc = AsyncMock()
-        mc.__aenter__ = AsyncMock(return_value=mc)
-        mc.__aexit__ = AsyncMock(return_value=False)
-
-        async def fake_post(url, **kwargs):
-            r = MagicMock()
-            r.status_code = 200
-            r.json = MagicMock(return_value={
-                "result": {
-                    "parts": [{"kind": "text", "text": "Task done!"}]
-                }
-            })
-            return r
-
-        async def fake_get(url, **kwargs):
-            r = MagicMock()
-            r.status_code = 200
-            r.json = MagicMock(return_value={"url": "http://peer.svc/a2a"})
-            return r
-
-        mc.post = fake_post
-        mc.get = fake_get
-
-        with patch("a2a_tools.httpx.AsyncClient", return_value=mc):
-            result = await a2a_tools.delegate_task("ws-peer-789", "do a thing")
-
-        assert result == "Task done!"
-
-
 # ---------------------------------------------------------------------------
 # tool_delegate_task_async
 # ---------------------------------------------------------------------------
@@ -4,77 +4,82 @@ The idle loop skips sending the idle prompt when DELEGATION_RESULTS_FILE
 contains unconsumed results, preventing the agent from composing a stale tick
 before processing pending delegation notifications from the heartbeat.

-Source: workspace/main.py:_run_idle_loop() pending-results guard.
+Source: ``workspace/main.py:_check_delegation_results_pending()`` (extracted from
+``_run_idle_loop()`` guard; see PR #432 follow-up).
+
+The guard is extracted into a module-level function so unit tests call the
+real production logic directly — not a mirror copy.  This avoids the
+test-mirror anti-pattern (issue #401) where a copied implementation
+drifts from the production code it is supposed to test.
 """
 from __future__ import annotations

+import io
 import json
+from unittest.mock import patch

-import pytest
-
-
-def check_results_pending(file_path: str) -> bool:
-    """Mirror the guard logic from workspace/main.py:_run_idle_loop().
-
-    Returns True if the results file exists and is non-empty,
-    meaning the idle loop should skip this tick.
-    """
-    try:
-        with open(file_path) as rf:
-            rf.seek(0)
-            content = rf.read().strip()
-        return bool(content)
-    except FileNotFoundError:
-        return False
+from main import _check_delegation_results_pending


 class TestIdleLoopPendingCheck:
-    """Tests for the idle-loop pending-delegation-results guard."""
+    """Tests for the idle-loop pending-delegation-results guard.

-    def test_no_file_means_proceed(self, tmp_path):
+    Each test patches ``builtins.open`` so ``_check_delegation_results_pending``
+    reads the controlled payload instead of the real DELEGATION_RESULTS_FILE.
+    No filesystem side-effects.
+    """
+
+    def _patch_open(self, payload: str | None):
+        """Patch builtins.open for _check_delegation_results_pending.
+
+        Args:
+            payload: file contents to return. None → FileNotFoundError.
+        """
+        if payload is None:
+            return patch("builtins.open", side_effect=FileNotFoundError)
+        else:
+            fake_file = io.StringIO(payload)
+            return patch("builtins.open", return_value=fake_file)
+
+    def test_no_file_means_proceed(self):
        """No delegation results file → idle loop fires normally."""
-        results_file = tmp_path / "delegation_results.jsonl"
-        assert not check_results_pending(str(results_file))
+        with self._patch_open(None):
+            assert _check_delegation_results_pending() is False

-    def test_empty_file_means_proceed(self, tmp_path):
+    def test_empty_file_means_proceed(self):
        """Empty file → no pending results → idle loop fires."""
-        results_file = tmp_path / "delegation_results.jsonl"
-        results_file.write_text("", encoding="utf-8")
-        assert not check_results_pending(str(results_file))
+        with self._patch_open(""):
+            assert _check_delegation_results_pending() is False

-    def test_whitespace_only_file_means_proceed(self, tmp_path):
+    def test_whitespace_only_file_means_proceed(self):
        """File with only whitespace → treated as empty → idle loop fires."""
-        results_file = tmp_path / "delegation_results.jsonl"
-        results_file.write_text("  \n  ", encoding="utf-8")
-        assert not check_results_pending(str(results_file))
+        with self._patch_open("  \n  "):
+            assert _check_delegation_results_pending() is False

-    def test_single_result_means_skip(self, tmp_path):
+    def test_single_result_means_skip(self):
        """File with one delegation result → skip idle tick."""
-        results_file = tmp_path / "delegation_results.jsonl"
-        results_file.write_text(
+        payload = (
            json.dumps({
                "status": "completed",
                "delegation_id": "del-abc",
                "summary": "Done",
-            }) + "\n",
-            encoding="utf-8",
+            }) + "\n"
        )
-        assert check_results_pending(str(results_file))
+        with self._patch_open(payload):
+            assert _check_delegation_results_pending() is True

-    def test_multiple_results_means_skip(self, tmp_path):
+    def test_multiple_results_means_skip(self):
        """File with multiple delegation results → skip idle tick."""
-        results_file = tmp_path / "delegation_results.jsonl"
-        results_file.write_text(
+        payload = (
            json.dumps({"status": "completed", "delegation_id": "del-1", "summary": "A"})
            + "\n"
            + json.dumps({"status": "failed", "delegation_id": "del-2", "summary": "B"})
-            + "\n",
-            encoding="utf-8",
+            + "\n"
        )
-        assert check_results_pending(str(results_file))
+        with self._patch_open(payload):
+            assert _check_delegation_results_pending() is True

-    def test_file_with_only_newline_means_proceed(self, tmp_path):
+    def test_file_with_only_newline_means_proceed(self):
        """File with only a newline character → stripped to empty → fires."""
-        results_file = tmp_path / "delegation_results.jsonl"
-        results_file.write_text("\n", encoding="utf-8")
-        assert not check_results_pending(str(results_file))
+        with self._patch_open("\n"):
+            assert _check_delegation_results_pending() is False