fix(ci): use AUTO_SYNC_TOKEN for auto-sync main->staging (Class D)

Same shape as molecule-controlplane#29: per-job GITHUB_TOKEN
doesn't have the Gitea API permissions to open PRs / push branches
the auto-sync flow needs. AUTO_SYNC_TOKEN is the devops-engineer
persona PAT (per saved memory feedback_per_agent_gitea_identity_default).

Companion prod ops (already done):
- devops-engineer added as collaborator on molecule-core (write)
- devops-engineer added to staging branch protection push_whitelist
- AUTO_SYNC_TOKEN registered as Actions secret on molecule-core

.github/scripts/lint_secret_pattern_drift.py

@@ -37,7 +37,7 @@ CANONICAL_FILE = Path(".github/workflows/secret-scan.yml")
 CONSUMERS: list[tuple[str, str]] = [
     (
         "molecule-ai-workspace-runtime/molecule_runtime/scripts/pre-commit-checks.sh",
-        "https://raw.githubusercontent.com/Molecule-AI/molecule-ai-workspace-runtime/main/molecule_runtime/scripts/pre-commit-checks.sh",
+        "https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-runtime/raw/branch/main/molecule_runtime/scripts/pre-commit-checks.sh",
     ),
 ]
 

.github/workflows/auto-sync-main-to-staging.yml

@@ -103,7 +103,7 @@ jobs:
         with:
           fetch-depth: 0
           ref: staging
-          token: ${{ secrets.GITHUB_TOKEN }}
+          token: ${{ secrets.AUTO_SYNC_TOKEN }}
 
       - name: Configure git author
         run: |
@@ -174,7 +174,7 @@ jobs:
       - name: Open auto-sync PR + enable auto-merge
         if: steps.check.outputs.needs_sync == 'true'
         env:
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GH_TOKEN: ${{ secrets.AUTO_SYNC_TOKEN }}
           BRANCH: ${{ steps.check.outputs.branch }}
           MAIN_SHORT: ${{ steps.check.outputs.main_short }}
           DID_FF: ${{ steps.prep.outputs.did_ff }}

.github/workflows/block-internal-paths.yml

@@ -1,7 +1,7 @@
 name: Block internal-flavored paths
 
 # Hard CI gate. Internal content (positioning, competitive briefs, sales
-# playbooks, PMM/press drip, draft campaigns) lives in molecule-ai/internal —
+# playbooks, PMM/press drip, draft campaigns) lives in Molecule-AI/internal —
 # this public monorepo must never re-acquire those paths. CEO directive
 # 2026-04-23 after a fleet-wide audit found 79 internal files leaked here.
 #
@@ -135,7 +135,7 @@ jobs:
             echo "::error::Forbidden internal-flavored paths detected:"
             printf "$OFFENDING"
             echo ""
-            echo "These paths belong in molecule-ai/internal, not this public repo."
+            echo "These paths belong in Molecule-AI/internal, not this public repo."
             echo "See docs/internal-content-policy.md for canonical locations."
             echo ""
             echo "If your file is genuinely public-facing (e.g. a blog post"

.github/workflows/canary-verify.yml

@@ -108,7 +108,7 @@ jobs:
               echo
               echo "One or more canary secrets are unset (\`CANARY_TENANT_URLS\`, \`CANARY_ADMIN_TOKENS\`, \`CANARY_CP_SHARED_SECRET\`)."
               echo "Phase 2 canary fleet has not been stood up yet —"
-              echo "see [canary-tenants.md](https://github.com/molecule-ai/molecule-controlplane/blob/main/docs/canary-tenants.md)."
+              echo "see [canary-tenants.md](https://github.com/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"

.github/workflows/ci.yml

@@ -87,7 +87,7 @@ jobs:
         run: go mod download
       - if: needs.changes.outputs.platform == 'true'
         run: go build ./cmd/server
-      # CLI (molecli) moved to standalone repo: github.com/molecule-ai/molecule-cli
+      # CLI (molecli) moved to standalone repo: github.com/Molecule-AI/molecule-cli
       - if: needs.changes.outputs.platform == 'true'
         run: go vet ./... || true
       - if: needs.changes.outputs.platform == 'true'
@@ -165,7 +165,7 @@ jobs:
               # Strip the package-import prefix so we can match .coverage-allowlist.txt
               # entries written as paths relative to workspace-server/.
               # Handle both module paths: platform/workspace-server/... and platform/...
-              rel=$(echo "$file" | sed 's|^github.com/molecule-ai/molecule-monorepo/platform/workspace-server/||; s|^github.com/molecule-ai/molecule-monorepo/platform/||')
+              rel=$(echo "$file" | sed 's|^github.com/Molecule-AI/molecule-monorepo/platform/workspace-server/||; s|^github.com/Molecule-AI/molecule-monorepo/platform/||')
 
               if echo "$ALLOWLIST" | grep -qxF "$rel"; then
                 echo "::warning file=workspace-server/$rel::Critical file at ${pct}% coverage (allowlisted, #1823) — fix before expiry."
@@ -235,7 +235,7 @@ jobs:
         run: npx vitest run --coverage
       - name: Upload coverage summary as artifact
         if: needs.changes.outputs.canvas == 'true' && always()
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        uses: actions/upload-artifact@v3 # pinned to v3 for Gitea act_runner v0.6 compatibility (internal#46)
         with:
           name: canvas-coverage-${{ github.run_id }}
           path: canvas/coverage/
@@ -243,8 +243,8 @@ jobs:
           if-no-files-found: warn
 
   # MCP Server + SDK removed from CI — now in standalone repos:
-  # - github.com/molecule-ai/molecule-mcp-server (npm CI)
-  # - github.com/molecule-ai/molecule-sdk-python (PyPI CI)
+  # - github.com/Molecule-AI/molecule-mcp-server (npm CI)
+  # - github.com/Molecule-AI/molecule-sdk-python (PyPI CI)
 
   # e2e-api job moved to .github/workflows/e2e-api.yml (issue #458).
   # It now has workflow-level concurrency (cancel-in-progress: false) so
@@ -434,5 +434,5 @@ jobs:
           fi
 
       # SDK + plugin validation moved to standalone repo:
-      # github.com/molecule-ai/molecule-sdk-python
+      # github.com/Molecule-AI/molecule-sdk-python
 

.github/workflows/codeql.yml

@@ -43,9 +43,6 @@ permissions:
 jobs:
   analyze:
     name: Analyze (${{ matrix.language }})
-    # CodeQL set to advisory (non-blocking) on Gitea Actions — Hongming dec'''n 2026-05-07 (#156).
-    # Findings still emit as SARIF artifacts; failing CodeQL run does not block PR merge.
-    continue-on-error: true
     runs-on: ubuntu-latest
     timeout-minutes: 45
 

.github/workflows/e2e-staging-canvas.yml

@@ -139,7 +139,7 @@ jobs:
 
       - name: Upload Playwright report on failure
         if: failure() && needs.detect-changes.outputs.canvas == 'true'
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        uses: actions/upload-artifact@v3 # pinned to v3 for Gitea act_runner v0.6 compatibility (internal#46)
         with:
           name: playwright-report-staging
           path: canvas/playwright-report-staging/
@@ -147,7 +147,7 @@ jobs:
 
       - name: Upload screenshots on failure
         if: failure() && needs.detect-changes.outputs.canvas == 'true'
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        uses: actions/upload-artifact@v3 # pinned to v3 for Gitea act_runner v0.6 compatibility (internal#46)
         with:
           name: playwright-screenshots
           path: canvas/test-results/

.github/workflows/pr-guards.yml

@@ -19,4 +19,4 @@ permissions:
 
 jobs:
   disable-auto-merge-on-push:
-    uses: molecule-ai/molecule-ci/.github/workflows/disable-auto-merge-on-push.yml@main
+    uses: Molecule-AI/molecule-ci/.github/workflows/disable-auto-merge-on-push.yml@main

.github/workflows/publish-runtime.yml

@@ -25,7 +25,7 @@ name: publish-runtime
 #   3. Publishes to PyPI via the PyPA Trusted Publisher action (OIDC).
 #      No static API token is stored — PyPI verifies the workflow's
 #      OIDC claim against the trusted-publisher config registered for
-#      molecule-ai-workspace-runtime (molecule-ai/molecule-core,
+#      molecule-ai-workspace-runtime (Molecule-AI/molecule-core,
 #      publish-runtime.yml, environment pypi-publish).
 #
 # After publish: the 8 template repos pick up the new version on their
@@ -166,7 +166,7 @@ jobs:
 
       - name: Publish to PyPI (Trusted Publisher / OIDC)
         # PyPI side is configured: project molecule-ai-workspace-runtime →
-        # publisher molecule-ai/molecule-core, workflow publish-runtime.yml,
+        # publisher Molecule-AI/molecule-core, workflow publish-runtime.yml,
         # environment pypi-publish. The action mints a short-lived OIDC
         # token and exchanges it for a PyPI upload credential — no static
         # API token in this repo's secrets.
@@ -282,42 +282,33 @@ jobs:
           echo "::error::Refusing to fan out cascade against stale or corrupt PyPI surfaces."
           exit 1
 
-      - name: Fan out repository_dispatch
+      - name: Fan out via push to .runtime-version
         env:
-          # Fine-grained PAT with `actions:write` on the 8 template repos.
-          # GITHUB_TOKEN can't fire dispatches across repos — needs an explicit
-          # token. Stored as a repo secret; rotate per the standard schedule.
-          DISPATCH_TOKEN: ${{ secrets.TEMPLATE_DISPATCH_TOKEN }}
-          # Single source of truth: the publish job's output, which handles
-          # tag/manual-input/auto-bump uniformly. The previous fallback
-          # (`steps.version.outputs.version` from inside the cascade job)
-          # was a dead reference — different job, no shared step scope.
+          # Gitea PAT with write:repository scope on the 8 cascade-active
+          # template repos. Used here for `git push` (NOT for an API
+          # dispatch — Gitea 1.22.6 has no repository_dispatch endpoint;
+          # empirically verified across 6 candidate paths in molecule-
+          # core#20 issuecomment-913). The push trips each template's
+          # existing `on: push: branches: [main]` trigger on
+          # publish-image.yml, which then reads the updated
+          # .runtime-version via its resolve-version job.
+          DISPATCH_TOKEN: ${{ secrets.DISPATCH_TOKEN }}
           RUNTIME_VERSION: ${{ needs.publish.outputs.version }}
         run: |
           set +e   # don't abort on a single repo failure — collect them all
-          # Schedule-vs-dispatch behaviour split (hardened 2026-04-28
-          # after the sweep-cf-orphans soft-skip incident — same class
-          # of bug):
-          #
-          # The earlier "skipping cascade. templates will pick up the
-          # new version on their own next rebuild" message was wrong —
-          # templates only build on this dispatch trigger; without it
-          # they stay pinned to whatever runtime version they last saw.
-          # A silent skip here means "PyPI is current, templates are
-          # not" and the gap is invisible until someone notices a
-          # template still on the old version weeks later.
-          #
-          #   - push                → exit 1 (red CI surfaces the gap)
-          #   - workflow_dispatch   → exit 0 with a warning (operator
-          #                           ran this ad-hoc; let them rerun
-          #                           after fixing the secret)
+
+          # Soft-skip on workflow_dispatch when the token is missing
+          # (operator ad-hoc test); hard-fail on push so unattended
+          # publishes can't silently skip the cascade. Same shape as
+          # the original v1, intentional split per the schedule-vs-
+          # dispatch hardening 2026-04-28.
           if [ -z "$DISPATCH_TOKEN" ]; then
             if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
-              echo "::warning::TEMPLATE_DISPATCH_TOKEN secret not set — skipping cascade."
+              echo "::warning::DISPATCH_TOKEN secret not set — skipping cascade."
               echo "::warning::set it at Settings → Secrets and Variables → Actions, then rerun. Templates will stay on the prior runtime version until either this token is set or each template is rebuilt manually."
               exit 0
             fi
-            echo "::error::TEMPLATE_DISPATCH_TOKEN secret missing — cascade cannot fan out."
+            echo "::error::DISPATCH_TOKEN secret missing — cascade cannot fan out."
             echo "::error::PyPI was published, but the 8 template repos will NOT pick up the new version until this token is restored and a republish dispatches the cascade."
             echo "::error::set it at Settings → Secrets and Variables → Actions; then re-trigger publish-runtime via workflow_dispatch."
             exit 1
@@ -327,37 +318,119 @@ jobs:
             echo "::error::publish job did not expose a version output — cascade cannot fan out"
             exit 1
           fi
-          # All 9 active workspace template repos. The PR #2536 pruning
-          # ("deprecated, no shipping images") was empirically wrong:
-          # continuous-synth-e2e.yml defaults to langgraph as its primary
-          # canary (line 44), and every excluded template had successful
-          # publish-image runs as of 2026-05-03 — none were dormant.
-          # Symptom of the prune: today's a2a-sdk strict-mode fix
-          # (#2566 / commit e1628c4) cascaded to 4 templates but never
-          # reached langgraph, so the synth-E2E correctly canary'd a fix
-          # that had landed but not deployed. Re-added the 5 templates.
-          # Long-term: derive this list from manifest.json so cascade
-          # scope can't drift from E2E scope — tracked in RFC #388 as a
-          # Phase-1 invariant.
+
+          # All 9 workspace templates declared in manifest.json. The list
+          # MUST stay aligned with manifest.json's workspace_templates —
+          # cascade-list-drift-gate.yml enforces this in CI per the
+          # codex-stuck-on-stale-runtime invariant from PR #2556.
+          # Long-term goal: derive this list from manifest.json so it
+          # can't drift even on a manifest edit (RFC #388 Phase-1).
+          #
+          # Per-template publish-image.yml presence is checked at
+          # cascade-time below: codex doesn't ship one today, so the
+          # cascade soft-skips it with an informational message rather
+          # than dropping it from this list (which would re-introduce
+          # the drift the gate exists to catch).
+          GITEA_URL="${GITEA_URL:-https://git.moleculesai.app}"
           TEMPLATES="claude-code hermes openclaw codex langgraph crewai autogen deepagents gemini-cli"
           FAILED=""
+          SKIPPED=""
+
+          # Configure git identity once. The persona owning DISPATCH_TOKEN
+          # is the same identity that authored this commit on each
+          # template; using a generic "publish-runtime cascade" co-author
+          # trailer in the message keeps the audit trail honest about the
+          # workflow-driven origin.
+          git config --global user.name  "publish-runtime cascade"
+          git config --global user.email "publish-runtime@moleculesai.app"
+
+          WORKDIR="$(mktemp -d)"
           for tpl in $TEMPLATES; do
             REPO="molecule-ai/molecule-ai-workspace-template-$tpl"
-            STATUS=$(curl -sS -o /tmp/dispatch.out -w "%{http_code}" \
-              -X POST "https://api.github.com/repos/$REPO/dispatches" \
-              -H "Authorization: Bearer $DISPATCH_TOKEN" \
-              -H "Accept: application/vnd.github+json" \
-              -H "X-GitHub-Api-Version: 2022-11-28" \
-              -d "{\"event_type\":\"runtime-published\",\"client_payload\":{\"runtime_version\":\"$VERSION\"}}")
-            if [ "$STATUS" = "204" ]; then
-              echo "✓ dispatched $tpl ($VERSION)"
-            else
-              echo "::warning::✗ failed to dispatch $tpl: HTTP $STATUS — $(cat /tmp/dispatch.out)"
+            CLONE="$WORKDIR/$tpl"
+
+            # Pre-check: skip templates without a publish-image.yml.
+            # The cascade's job is to trip the template's on-push
+            # rebuild — if there's no rebuild workflow, pushing a
+            # .runtime-version commit is just noise on the target
+            # repo. Use the Gitea contents API (no clone required for
+            # the probe). 200 = present; 404 = absent.
+            HTTP=$(curl -sS -o /dev/null -w "%{http_code}" \
+              -H "Authorization: token $DISPATCH_TOKEN" \
+              "$GITEA_URL/api/v1/repos/$REPO/contents/.github/workflows/publish-image.yml")
+            if [ "$HTTP" = "404" ]; then
+              echo "↷ $tpl has no publish-image.yml — soft-skip (informational; manifest still tracks it)"
+              SKIPPED="$SKIPPED $tpl"
+              continue
+            fi
+            if [ "$HTTP" != "200" ]; then
+              echo "::warning::$tpl publish-image.yml probe returned HTTP $HTTP — proceeding anyway, push will surface the real failure if any"
+            fi
+
+            # Use a per-template attempt loop so a transient race (e.g.
+            # human pushing to the same template at the same instant)
+            # doesn't lose the cascade. Bounded retries (3) — beyond
+            # that we surface the failure and let the operator retry.
+            attempt=0
+            success=false
+            while [ $attempt -lt 3 ]; do
+              attempt=$((attempt + 1))
+              rm -rf "$CLONE"
+              if ! git clone --depth=1 \
+                  "https://x-access-token:${DISPATCH_TOKEN}@${GITEA_URL#https://}/$REPO.git" \
+                  "$CLONE" >/tmp/clone.log 2>&1; then
+                echo "::warning::clone $tpl attempt $attempt failed: $(tail -n3 /tmp/clone.log)"
+                sleep 2
+                continue
+              fi
+
+              cd "$CLONE"
+              echo "$VERSION" > .runtime-version
+
+              # Idempotency guard: if the file already matches, this
+              # publish is a re-run for a version already cascaded.
+              # Don't push a no-op commit (would spuriously re-trip the
+              # template's on-push and rebuild for nothing).
+              if git diff --quiet -- .runtime-version; then
+                echo "✓ $tpl already at $VERSION — no commit needed (idempotent)"
+                success=true
+                cd - >/dev/null
+                break
+              fi
+
+              git add .runtime-version
+              git commit -m "chore: pin runtime to $VERSION (publish-runtime cascade)" \
+                -m "Co-Authored-By: publish-runtime cascade <publish-runtime@moleculesai.app>" \
+                >/dev/null
+
+              if git push origin HEAD:main >/tmp/push.log 2>&1; then
+                echo "✓ $tpl pushed $VERSION on attempt $attempt"
+                success=true
+                cd - >/dev/null
+                break
+              fi
+
+              # Likely a non-fast-forward — pull-rebase and retry.
+              # Don't force-push: that would silently overwrite a racing
+              # human/cascade commit.
+              echo "::warning::push $tpl attempt $attempt failed, pull-rebasing: $(tail -n3 /tmp/push.log)"
+              git pull --rebase origin main >/tmp/rebase.log 2>&1 || true
+              cd - >/dev/null
+            done
+
+            if [ "$success" != "true" ]; then
               FAILED="$FAILED $tpl"
             fi
           done
+          rm -rf "$WORKDIR"
+
           if [ -n "$FAILED" ]; then
-            echo "::warning::Cascade incomplete. Failed templates:$FAILED"
-            # Don't fail the whole job — PyPI publish already succeeded;
-            # operators can retry the failed templates manually.
+            echo "::error::Cascade incomplete after 3 retries each. Failed templates:$FAILED"
+            echo "::error::PyPI publish succeeded; failed templates lag the new version. Re-run this workflow_dispatch with the same version to retry only the laggers (idempotent — already-cascaded templates skip)."
+            exit 1
+          fi
+          if [ -n "$SKIPPED" ]; then
+            echo "Cascade complete: pinned $VERSION on cascade-active templates. Soft-skipped (no publish-image.yml):$SKIPPED"
+          else
+            echo "Cascade complete: $VERSION pinned across all manifest workspace_templates."
           fi

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

@@ -37,7 +37,6 @@ on:
       - 'workspace-server/**'
       - 'canvas/**'
       - 'manifest.json'
-      - 'scripts/**'
       - '.github/workflows/publish-workspace-server-image.yml'
   workflow_dispatch:
 
@@ -182,4 +181,3 @@ jobs:
             org.opencontainers.image.source=https://github.com/${{ github.repository }}
             org.opencontainers.image.revision=${{ github.sha }}
             org.opencontainers.image.description=Molecule AI tenant platform + canvas — pending canary verify
-

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

@@ -9,7 +9,7 @@ name: redeploy-tenants-on-main
 #
 # This workflow closes the gap by calling the control-plane admin
 # endpoint that performs a canary-first, batched, health-gated rolling
-# redeploy across every live tenant. Implemented in molecule-ai/
+# redeploy across every live tenant. Implemented in Molecule-AI/
 # molecule-controlplane as POST /cp/admin/tenants/redeploy-fleet
 # (feat/tenant-auto-redeploy, landing alongside this workflow).
 #
@@ -146,7 +146,7 @@ jobs:
 
       - name: Call CP redeploy-fleet
         # CP_ADMIN_API_TOKEN must be set as a repo/org secret on
-        # molecule-ai/molecule-core, matching the staging/prod CP's
+        # Molecule-AI/molecule-core, matching the staging/prod CP's
         # CP_ADMIN_API_TOKEN env. Stored in Railway, mirrored to this
         # repo's secrets for CI.
         env:

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

@@ -97,7 +97,7 @@ jobs:
 
       - name: Call staging-CP redeploy-fleet
         # CP_STAGING_ADMIN_API_TOKEN must be set as a repo/org secret
-        # on molecule-ai/molecule-core, matching staging-CP's
+        # on Molecule-AI/molecule-core, matching staging-CP's
         # CP_ADMIN_API_TOKEN env var (visible in Railway controlplane
         # / staging environment). Stored separately from the prod
         # CP_ADMIN_API_TOKEN so a leak of one doesn't auth the other.

.github/workflows/retarget-main-to-staging.yml

@@ -96,7 +96,7 @@ jobs:
             --body "$(cat <<'BODY'
           [retarget-bot] This PR was opened against `main` and has been retargeted to `staging` automatically.
 
-          **Why:** per [SHARED_RULES rule 8](https://github.com/molecule-ai/molecule-ai-org-template-molecule-dev/blob/main/SHARED_RULES.md), all feature work targets `staging` first; the CEO promotes `staging → main` separately.
+          **Why:** per [SHARED_RULES rule 8](https://github.com/Molecule-AI/molecule-ai-org-template-molecule-dev/blob/main/SHARED_RULES.md), all feature work targets `staging` first; the CEO promotes `staging → main` separately.
 
           **What changed:** just the base branch — no code change. CI will re-run against `staging`. If you get merge conflicts, rebase on `staging`.
 

.github/workflows/secret-scan.yml

@@ -12,7 +12,7 @@ name: Secret scan
 #
 #   jobs:
 #     secret-scan:
-#       uses: molecule-ai/molecule-core/.github/workflows/secret-scan.yml@staging
+#       uses: Molecule-AI/molecule-core/.github/workflows/secret-scan.yml@staging
 #
 # Pin to @staging not @main — staging is the active default branch,
 # main lags via the staging-promotion workflow. Updates ride along

CONTRIBUTING.md

@@ -22,7 +22,7 @@ development workflow, conventions, and how to get your changes merged.
 
 ```bash
 # Clone the repo
-git clone https://github.com/Molecule-AI/molecule-core.git
+git clone https://git.moleculesai.app/molecule-ai/molecule-core.git
 cd molecule-core
 
 # Install git hooks
@@ -57,7 +57,7 @@ See `CLAUDE.md` for a full list of environment variables and their purposes.
 
 This repo is scoped to **code** (canvas, workspace, workspace-server, related
 infra). Public content (blog posts, marketing copy, OG images, SEO briefs,
-DevRel demos) lives in [`Molecule-AI/docs`](https://github.com/Molecule-AI/docs).
+DevRel demos) lives in [`Molecule-AI/docs`](https://git.moleculesai.app/molecule-ai/docs).
 The `Block forbidden paths` CI gate fails any PR that writes to `marketing/`
 or other removed paths — open against `Molecule-AI/docs` instead.
 
@@ -110,7 +110,7 @@ causing a render loop when any node position changed.
 
 1. **Repo-wide:** "Automatically delete head branches" is on. Once a PR merges, the branch is deleted server-side. Any subsequent `git push` to that branch fails with `remote rejected — no such branch`.
 
-2. **CI:** the `pr-guards` workflow (calling [molecule-ci `disable-auto-merge-on-push`](https://github.com/Molecule-AI/molecule-ci/blob/main/.github/workflows/disable-auto-merge-on-push.yml)) fires on every push to an open PR. If auto-merge was already enabled, it's disabled and a comment is posted. You must explicitly re-enable after verifying the new commit.
+2. **CI:** the `pr-guards` workflow (calling [molecule-ci `disable-auto-merge-on-push`](https://git.moleculesai.app/molecule-ai/molecule-ci/src/branch/main/.github/workflows/disable-auto-merge-on-push.yml)) fires on every push to an open PR. If auto-merge was already enabled, it's disabled and a comment is posted. You must explicitly re-enable after verifying the new commit.
 
 **Workflow rules that follow from the guards:**
 - Push **all** commits before running `gh pr merge --auto`.
@@ -180,9 +180,9 @@ and run CI manually.
 Code in this repo lands in molecule-core. Some related runtime artifacts
 live in their own repos:
 
-- [`Molecule-AI/molecule-ai-workspace-runtime`](https://github.com/Molecule-AI/molecule-ai-workspace-runtime) — Python adapter SDK (`molecule_runtime`) that runs inside containerized Molecule workspaces. Bridges Claude Code SDK / hermes / langgraph / etc. → A2A queue.
-- [`Molecule-AI/molecule-sdk-python`](https://github.com/Molecule-AI/molecule-sdk-python) — `A2AServer` + `RemoteAgentClient` for external agents that register over the public `/registry/register` flow.
-- [`Molecule-AI/molecule-mcp-claude-channel`](https://github.com/Molecule-AI/molecule-mcp-claude-channel) — Claude Code channel plugin. Bridges A2A traffic into a running Claude Code session via MCP `notifications/claude/channel`. Polling-based (no tunnel required); install with `claude --channels plugin:molecule@Molecule-AI/molecule-mcp-claude-channel`.
+- [`Molecule-AI/molecule-ai-workspace-runtime`](https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-runtime) — Python adapter SDK (`molecule_runtime`) that runs inside containerized Molecule workspaces. Bridges Claude Code SDK / hermes / langgraph / etc. → A2A queue.
+- [`Molecule-AI/molecule-sdk-python`](https://git.moleculesai.app/molecule-ai/molecule-sdk-python) — `A2AServer` + `RemoteAgentClient` for external agents that register over the public `/registry/register` flow.
+- [`Molecule-AI/molecule-mcp-claude-channel`](https://git.moleculesai.app/molecule-ai/molecule-mcp-claude-channel) — Claude Code channel plugin. Bridges A2A traffic into a running Claude Code session via MCP `notifications/claude/channel`. Polling-based (no tunnel required); install with `claude --channels plugin:molecule@Molecule-AI/molecule-mcp-claude-channel`.
 
 When extending the **A2A surface** in molecule-core (`workspace-server/internal/handlers/a2a_proxy.go` etc.), consider whether the change has a downstream impact on the runtime SDK or the channel plugin — they're versioned independently but share the wire shape.
 

README.md

@@ -1,7 +1,7 @@
 <div align="center">
 
 <p>
-  <img src="./docs/assets/branding/molecule-icon.png" alt="Molecule AI Icon Logo" width="160" />
+  <img src="./docs/assets/branding/molecule-icon.svg" alt="Molecule AI" width="160" />
 </p>
 
 <p>
@@ -39,8 +39,8 @@
   <a href="./docs/agent-runtime/workspace-runtime.md"><strong>Workspace Runtime</strong></a>
 </p>
 
-[![Deploy on Railway](https://railway.app/button.svg)](https://railway.app/new/template?template=https://github.com/Molecule-AI/molecule-monorepo)
-[![Deploy to Render](https://render.com/images/deploy-to-render-button.svg)](https://render.com/deploy?repo=https://github.com/Molecule-AI/molecule-monorepo)
+[![Deploy on Railway](https://railway.app/button.svg)](https://railway.app/new/template?template=https://git.moleculesai.app/molecule-ai/molecule-core)
+[![Deploy to Render](https://render.com/images/deploy-to-render-button.svg)](https://render.com/deploy?repo=https://git.moleculesai.app/molecule-ai/molecule-core)
 
 </div>
 
@@ -53,8 +53,8 @@ Molecule AI is the most powerful way to govern an AI agent organization in produ
 It combines the parts that are usually scattered across demos, internal glue code, and framework-specific tooling into one product:
 
 - one org-native control plane for teams, roles, hierarchy, and lifecycle
-- one runtime layer that lets LangGraph, DeepAgents, Claude Code, CrewAI, AutoGen, and OpenClaw run side by side
-- one memory model that keeps recall, sharing, and skill evolution aligned with organizational boundaries
+- one runtime layer that lets **eight** agent runtimes — LangGraph, DeepAgents, Claude Code, CrewAI, AutoGen, **Hermes**, **Gemini CLI**, and OpenClaw — run side by side behind one workspace contract
+- one memory model that keeps recall, sharing, and skill evolution aligned with organizational boundaries (Memory v2 backed by pgvector for semantic recall)
 - one operational surface for observing, pausing, restarting, inspecting, and improving live workspaces
 
 Most teams can build a workflow, a strong single agent, a coding agent, or a custom multi-agent graph.
@@ -75,7 +75,7 @@ You do not wire collaboration paths by hand. Hierarchy defines the default commu
 
 ### 3. Runtime choice stops being a dead-end decision
 
-LangGraph, DeepAgents, Claude Code, CrewAI, AutoGen, and OpenClaw can all plug into the same workspace abstraction. Teams can standardize governance without forcing every group onto one runtime.
+LangGraph, DeepAgents, Claude Code, CrewAI, AutoGen, Hermes, Gemini CLI, and OpenClaw can all plug into the same workspace abstraction. Teams can standardize governance without forcing every group onto one runtime.
 
 ### 4. Memory is treated like infrastructure
 
@@ -117,6 +117,8 @@ Molecule AI is not trying to replace the frameworks below. It is the system that
 | **Claude Code** | Shipping on `main` | Real coding workflows, CLI-native continuity | Secure workspace abstraction, A2A delegation, org boundaries, shared control plane |
 | **CrewAI** | Shipping on `main` | Role-based crews | Persistent workspace identity, policy consistency, shared canvas and registry |
 | **AutoGen** | Shipping on `main` | Assistant/tool orchestration | Standardized deployment, hierarchy-aware collaboration, shared ops plane |
+| **Hermes 4** | Shipping on `main` | Hybrid reasoning, native tools, json_schema (NousResearch/hermes-agent) | Option B upstream hook, A2A bridge to OpenAI-compat API, multi-provider provider derivation |
+| **Gemini CLI** | Shipping on `main` | Google Gemini CLI continuity | Workspace lifecycle, A2A, hierarchy-aware collaboration, shared ops plane |
 | **OpenClaw** | Shipping on `main` | CLI-native runtime with its own session model | Workspace lifecycle, templates, activity logs, topology-aware collaboration |
 | **NemoClaw** | WIP on `feat/nemoclaw-t4-docker` | NVIDIA-oriented runtime path | Planned to join the same abstraction once merged; not yet part of `main` |
 
@@ -182,9 +184,10 @@ The result is not just “an agent that learns.” It is **an organization that
 
 ## What Ships In `main`
 
-### Canvas
+### Canvas (v4)
 
 - Next.js 15 + React Flow + Zustand
+- **warm-paper theme system** — light / dark / follow-system, SSR cookie + nonce'd boot script + ThemeProvider; terminal + code surfaces stay dark unconditionally
 - drag-to-nest team building
 - empty-state deployment + onboarding wizard
 - template palette
@@ -193,8 +196,9 @@ The result is not just “an agent that learns.” It is **an organization that
 
 ### Platform
 
-- Go/Gin control plane
-- workspace CRUD and provisioning
+- Go 1.25 / Gin control plane (80+ HTTP endpoints + Gorilla WebSocket fanout)
+- workspace CRUD and provisioning (pluggable Provisioner — Docker locally, EC2 + SSM in production)
+- **A2A response path is a typed discriminated union (RFC #2967)** — frozen dataclasses + total parser; 100% unit + adversarial fuzz coverage
 - registry and heartbeats
 - browser-safe A2A proxy
 - team expansion/collapse
@@ -204,10 +208,10 @@ The result is not just “an agent that learns.” It is **an organization that
 
 ### Runtime
 
-- unified `workspace/` image
-- adapter-driven execution
+- unified `workspace/` image; thin AMI in production (us-east-2)
+- adapter-driven execution across **8 runtimes** (Claude Code, Hermes, Gemini CLI, LangGraph, DeepAgents, CrewAI, AutoGen, OpenClaw)
 - Agent Card registration
-- awareness-backed memory integration
+- awareness-backed memory integration; **Memory v2 backed by pgvector** for semantic recall
 - plugin-mounted shared rules/skills
 - hot-reloadable local skills
 - coordinator-only delegation path
@@ -221,6 +225,21 @@ The result is not just “an agent that learns.” It is **an organization that
 - runtime tiers
 - direct workspace inspection through terminal and files
 
+### SaaS (via [`molecule-controlplane`](https://github.com/Molecule-AI/molecule-controlplane))
+
+- multi-tenant on AWS EC2 + Neon (per-tenant Postgres branch) + Cloudflare Tunnels (per-tenant, no public ports)
+- WorkOS AuthKit + Stripe Checkout + Customer Portal
+- AWS KMS envelope encryption (DB / Redis connection strings); AWS Secrets Manager for tenant bootstrap
+- `tenant_resources` audit table + 30-min boot-event-aware reconciler — every CF / AWS lifecycle event recorded, claim vs live state diffed
+
+### Bring your own Claude Code session (via [`molecule-mcp-claude-channel`](https://github.com/Molecule-AI/molecule-mcp-claude-channel))
+
+- Claude Code plugin that bridges Molecule A2A traffic into a local Claude Code session via MCP
+- subscribe to one or more workspaces; peer messages surface as conversation turns; replies route back through Molecule's A2A
+- no tunnel, no public endpoint — the plugin self-registers each watched workspace as `delivery_mode=poll` and long-polls `/activity?since_id=…`
+- multi-tenant friendly: one plugin install can watch workspaces across multiple Molecule tenants (`MOLECULE_PLATFORM_URLS` per-workspace)
+- install via the standard marketplace flow: `/plugin marketplace add Molecule-AI/molecule-mcp-claude-channel` → `/plugin install molecule-channel@molecule-mcp-claude-channel`
+
 ## Built For Teams That Need More Than A Demo
 
 Molecule AI is especially strong when you need to run:
@@ -233,24 +252,30 @@ Molecule AI is especially strong when you need to run:
 ## Architecture
 
 ```text
-Canvas (Next.js :3000)  <--HTTP / WS-->  Platform (Go :8080)  <---> Postgres + Redis
-         |                                          |
-         |                                          +--> Docker provisioner / bundles / templates / secrets
+Canvas (Next.js 15, warm-paper :3000)  <--HTTP / WS-->  Platform (Go 1.25 :8080)  <---> Postgres + Redis
+         |                                                           |
+         |                                                           +--> Provisioner: Docker (local) / EC2 + SSM (prod)
+         |                                                           +--> bundles · templates · secrets · KMS
          |
-         +-------------------- shows --------------------> workspaces, teams, tasks, traces, events
+         +------------------------- shows ------------------------> workspaces, teams, tasks, traces, events
 
-Workspace Runtime (Python image with adapters)
-  - LangGraph / DeepAgents / Claude Code / CrewAI / AutoGen / OpenClaw
-  - Agent Card + A2A server
-  - heartbeat + activity + awareness-backed memory
+Workspace Runtime (Python ≥3.11, image with adapters)
+  - 8 adapters: LangGraph / DeepAgents / Claude Code / CrewAI / AutoGen / Hermes / Gemini CLI / OpenClaw
+  - Agent Card + A2A server (typed-SSOT response path, RFC #2967)
+  - heartbeat + activity + awareness-backed memory (Memory v2 — pgvector semantic recall)
   - skills + plugins + hot reload
+
+SaaS Control Plane (molecule-controlplane, private)
+  - per-tenant EC2 + Neon (Postgres branch) + Cloudflare Tunnel
+  - WorkOS · Stripe · KMS · AWS Secrets Manager
+  - tenant_resources audit + 30-min reconciler
 ```
 
 ## Quick Start
 
 ```bash
-git clone https://github.com/Molecule-AI/molecule-monorepo.git
-cd molecule-monorepo
+git clone https://git.moleculesai.app/molecule-ai/molecule-core.git
+cd molecule-core
 
 cp .env.example .env
 # Defaults boot the stack locally out of the box. See .env.example for
@@ -303,7 +328,11 @@ Then open `http://localhost:3000`:
 
 ## Current Scope
 
-The current `main` branch already includes the core platform, canvas, memory model, six production adapters, skill lifecycle, and operational surfaces. Adjacent runtime work such as **NemoClaw** remains branch-level until merged, and this README keeps that distinction explicit on purpose.
+The current `main` branch ships the core platform, Canvas v4 (warm-paper themed), Memory v2 (pgvector semantic recall), the typed-SSOT A2A response path (RFC #2967), **eight production adapters** (Claude Code, Hermes, Gemini CLI, LangGraph, DeepAgents, CrewAI, AutoGen, OpenClaw), skill lifecycle, and operational surfaces.
+
+The companion private repo [`molecule-controlplane`](https://github.com/Molecule-AI/molecule-controlplane) provides the SaaS surface — multi-tenant orchestration on EC2 + Neon + Cloudflare Tunnels, KMS envelope encryption, WorkOS auth, Stripe billing, and a `tenant_resources` audit table with a 30-min reconciler.
+
+Adjacent runtime work such as **NemoClaw** remains branch-level until merged, and this README keeps that distinction explicit on purpose.
 
 ## License
 

README.zh-CN.md

@@ -1,7 +1,7 @@
 <div align="center">
 
 <p>
-  <img src="./docs/assets/branding/molecule-icon.png" alt="Molecule AI 图案 Logo" width="160" />
+  <img src="./docs/assets/branding/molecule-icon.svg" alt="Molecule AI" width="160" />
 </p>
 
 <p>
@@ -38,8 +38,8 @@
   <a href="./docs/agent-runtime/workspace-runtime.md"><strong>Workspace Runtime</strong></a>
 </p>
 
-[![Deploy on Railway](https://railway.app/button.svg)](https://railway.app/new/template?template=https://github.com/Molecule-AI/molecule-core)
-[![Deploy to Render](https://render.com/images/deploy-to-render-button.svg)](https://render.com/deploy?repo=https://github.com/Molecule-AI/molecule-core)
+[![Deploy on Railway](https://railway.app/button.svg)](https://railway.app/new/template?template=https://git.moleculesai.app/molecule-ai/molecule-core)
+[![Deploy to Render](https://render.com/images/deploy-to-render-button.svg)](https://render.com/deploy?repo=https://git.moleculesai.app/molecule-ai/molecule-core)
 
 </div>
 
@@ -52,8 +52,8 @@ Molecule AI 是目前最强的 AI Agent 组织治理方案之一，用来把 age
 它把过去分散在 demo、内部胶水代码和各类 framework 私有工具里的关键能力，收敛成一个产品：
 
 - 一套组织原生 control plane，管理团队、角色、层级和生命周期
-- 一套 runtime abstraction，让 LangGraph、DeepAgents、Claude Code、CrewAI、AutoGen、OpenClaw 并存运行
-- 一套与组织边界对齐的 memory 模型，把 recall、sharing 和 skill evolution 放进同一体系
+- 一套 runtime abstraction，让 **8 个** agent runtime —— LangGraph、DeepAgents、Claude Code、CrewAI、AutoGen、**Hermes**、**Gemini CLI**、OpenClaw —— 共用一套 workspace 契约
+- 一套与组织边界对齐的 memory 模型，把 recall、sharing 和 skill evolution 放进同一体系（Memory v2 由 pgvector 支撑语义召回）
 - 一套面向线上 workspace 的运维面，统一完成观测、暂停、重启、检查和持续改进
 
 今天很多团队能做好 workflow、单 agent、coding agent，或者自定义 multi-agent graph 中的一种。
@@ -74,7 +74,7 @@ Molecule AI 填的就是这个空白。
 
 ### 3. Runtime 选择不再是死路
 
-LangGraph、DeepAgents、Claude Code、CrewAI、AutoGen、OpenClaw 都可以挂到同一个 workspace abstraction 下。团队可以统一治理方式，而不必统一到底层 runtime。
+LangGraph、DeepAgents、Claude Code、CrewAI、AutoGen、Hermes、Gemini CLI、OpenClaw 都可以挂到同一个 workspace abstraction 下。团队可以统一治理方式，而不必统一到底层 runtime。
 
 ### 4. Memory 被当成基础设施来做
 
@@ -116,6 +116,8 @@ Molecule AI 并不是要替代下面这些 framework，而是把它们纳入更
 | **Claude Code** | `main` 已支持 | 真实编码工作流、CLI-native continuity | 安全 workspace 抽象、A2A delegation、组织边界、共享 control plane |
 | **CrewAI** | `main` 已支持 | 角色型 crew 模式清晰 | 持久 workspace 身份、统一策略、共享 Canvas 和 registry |
 | **AutoGen** | `main` 已支持 | assistant/tool orchestration | 统一部署、层级协作、共享运维平面 |
+| **Hermes 4** | `main` 已支持 | 混合推理、原生工具调用、json_schema 输出（NousResearch/hermes-agent） | Option B 上游 hook、A2A 桥接 OpenAI 兼容 API、多 provider 自动派生 |
+| **Gemini CLI** | `main` 已支持 | Google Gemini CLI 持续会话 | workspace 生命周期、A2A、层级感知协作、共享运维平面 |
 | **OpenClaw** | `main` 已支持 | CLI-native runtime，自有 session 模型 | workspace 生命周期、templates、activity logs、拓扑感知协作 |
 | **NemoClaw** | `feat/nemoclaw-t4-docker` 分支 WIP | NVIDIA 方向 runtime 路线 | 计划并入同一抽象层，但当前还不是 `main` 已合并能力 |
 
@@ -181,9 +183,10 @@ Molecule AI 并不是要替代下面这些 framework，而是把它们纳入更
 
 ## `main` 分支已经具备什么
 
-### Canvas
+### Canvas（v4）
 
 - Next.js 15 + React Flow + Zustand
+- **warm-paper 主题系统** —— light / dark / 跟随系统；SSR cookie + nonce'd boot 脚本 + ThemeProvider；终端与代码面板始终保持深色
 - drag-to-nest 团队构建
 - empty state + onboarding wizard
 - template palette
@@ -192,8 +195,9 @@ Molecule AI 并不是要替代下面这些 framework，而是把它们纳入更
 
 ### Platform
 
-- Go/Gin control plane
-- workspace CRUD 和 provisioning
+- Go 1.25 / Gin control plane（80+ HTTP 端点 + Gorilla WebSocket fanout）
+- workspace CRUD 和 provisioning（可插拔 Provisioner —— 本地 Docker、生产 EC2 + SSM）
+- **A2A 响应路径已收敛为类型化的判别联合（RFC #2967）** —— 冻结 dataclass + 全量 parser；100% 单元测试 + 对抗性 fuzz 覆盖
 - registry 与 heartbeat
 - 浏览器安全的 A2A proxy
 - team expansion/collapse
@@ -203,10 +207,10 @@ Molecule AI 并不是要替代下面这些 framework，而是把它们纳入更
 
 ### Runtime
 
-- 统一 `workspace/` 镜像
-- adapter 驱动执行
+- 统一 `workspace/` 镜像；生产环境采用 thin AMI（us-east-2）
+- adapter 驱动执行，覆盖 **8 个 runtime**（Claude Code、Hermes、Gemini CLI、LangGraph、DeepAgents、CrewAI、AutoGen、OpenClaw）
 - Agent Card 注册
-- awareness-backed memory
+- awareness-backed memory；**Memory v2 由 pgvector 支撑**语义召回
 - plugin 挂载共享 rules/skills
 - 本地 skills 热加载
 - coordinator-only delegation 路径
@@ -220,6 +224,21 @@ Molecule AI 并不是要替代下面这些 framework，而是把它们纳入更
 - runtime tiers
 - 终端与文件层面的 workspace 直接排障
 
+### SaaS（由 [`molecule-controlplane`](https://github.com/Molecule-AI/molecule-controlplane) 提供）
+
+- 多租户运行在 AWS EC2 + Neon（每租户一个 Postgres branch）+ Cloudflare Tunnels（每租户一条隧道，对外不开任何端口）
+- WorkOS AuthKit + Stripe Checkout + Customer Portal
+- AWS KMS 信封加密（DB / Redis 连接串）；AWS Secrets Manager 负责租户 bootstrap
+- `tenant_resources` 审计表 + 30 分钟 boot-event-aware reconciler —— 每个 CF / AWS lifecycle 事件都有记录，每 30 分钟比对 claim 与实际状态
+
+### 在 Claude Code 里直接接入（由 [`molecule-mcp-claude-channel`](https://github.com/Molecule-AI/molecule-mcp-claude-channel) 提供）
+
+- 把 Molecule A2A 流量桥接到本地 Claude Code 会话的 MCP 插件
+- 订阅一个或多个 workspace；peer 的消息会以 user-turn 出现，回复会经 Molecule A2A 路由出去
+- 无需公网隧道、无需公开端点 —— 插件启动时自动把每个 watched workspace 注册成 `delivery_mode=poll`，长轮询 `/activity?since_id=…`
+- 多租户友好：单次安装即可同时 watch 跨多个 Molecule 租户的 workspace（`MOLECULE_PLATFORM_URLS` 按 workspace 配置）
+- 通过标准 marketplace 流程安装：`/plugin marketplace add Molecule-AI/molecule-mcp-claude-channel` → `/plugin install molecule-channel@molecule-mcp-claude-channel`
+
 ## 适合什么团队
 
 Molecule AI 特别适合下面这些场景：
@@ -232,23 +251,29 @@ Molecule AI 特别适合下面这些场景：
 ## 架构总览
 
 ```text
-Canvas (Next.js :3000)  <--HTTP / WS-->  Platform (Go :8080)  <---> Postgres + Redis
-         |                                          |
-         |                                          +--> Docker provisioner / bundles / templates / secrets
+Canvas (Next.js 15, warm-paper :3000)  <--HTTP / WS-->  Platform (Go 1.25 :8080)  <---> Postgres + Redis
+         |                                                           |
+         |                                                           +--> Provisioner: Docker (本地) / EC2 + SSM (生产)
+         |                                                           +--> bundles · templates · secrets · KMS
          |
-         +-------------------- 展示 --------------------> workspaces, teams, tasks, traces, events
+         +------------------------- 展示 ------------------------> workspaces, teams, tasks, traces, events
 
-Workspace Runtime (Python image with adapters)
-  - LangGraph / DeepAgents / Claude Code / CrewAI / AutoGen / OpenClaw
-  - Agent Card + A2A server
-  - heartbeat + activity + awareness-backed memory
+Workspace Runtime (Python ≥3.11，含 adapter 集合的镜像)
+  - 8 个 adapter: LangGraph / DeepAgents / Claude Code / CrewAI / AutoGen / Hermes / Gemini CLI / OpenClaw
+  - Agent Card + A2A server（typed-SSOT 响应路径，RFC #2967）
+  - heartbeat + activity + awareness-backed memory（Memory v2 —— pgvector 语义召回）
   - skills + plugins + hot reload
+
+SaaS Control Plane (molecule-controlplane，私有)
+  - 每租户 EC2 + Neon (Postgres branch) + Cloudflare Tunnel
+  - WorkOS · Stripe · KMS · AWS Secrets Manager
+  - tenant_resources 审计 + 30 分钟 reconciler
 ```
 
 ## 快速开始
 
 ```bash
-git clone https://github.com/Molecule-AI/molecule-core.git
+git clone https://git.moleculesai.app/molecule-ai/molecule-core.git
 cd molecule-core
 
 cp .env.example .env
@@ -296,7 +321,11 @@ npm run dev
 
 ## 当前范围说明
 
-当前 `main` 已经包含核心平台、Canvas、memory model、6 个正式 adapter、skill lifecycle 和主要运维面。像 **NemoClaw** 这样的相邻 runtime 路线仍然属于分支级工作，只有合并后才会进入正式支持列表，这里会明确区分。
+当前 `main` 已经包含核心平台、Canvas v4（warm-paper 主题）、Memory v2（pgvector 语义召回）、typed-SSOT A2A 响应路径（RFC #2967）、**8 个正式 adapter**（Claude Code、Hermes、Gemini CLI、LangGraph、DeepAgents、CrewAI、AutoGen、OpenClaw）、skill lifecycle，以及主要运维面。
+
+配套的私有仓库 [`molecule-controlplane`](https://github.com/Molecule-AI/molecule-controlplane) 提供 SaaS 层 —— 多租户编排（EC2 + Neon + Cloudflare Tunnels）、KMS 信封加密、WorkOS 鉴权、Stripe 计费，以及 `tenant_resources` 审计表加 30 分钟 reconciler。
+
+像 **NemoClaw** 这样的相邻 runtime 路线仍然属于分支级工作，只有合并后才会进入正式支持列表，这里会明确区分。
 
 ## License
 

canvas/src/app/layout.tsx

@@ -3,7 +3,6 @@ import { cookies, headers } from "next/headers";
 import "./globals.css";
 import { AuthGate } from "@/components/AuthGate";
 import { CookieConsent } from "@/components/CookieConsent";
-import { PurchaseSuccessModal } from "@/components/PurchaseSuccessModal";
 import { ThemeProvider } from "@/lib/theme-provider";
 import {
   THEME_COOKIE,
@@ -87,12 +86,6 @@ export default async function RootLayout({
               vercel preview URL, apex) pass through unchanged. */}
           <AuthGate>{children}</AuthGate>
           <CookieConsent />
-          {/* Demo Mock #1: post-purchase success toast. Mounted at the
-              layout level so it persists across page state transitions
-              (loading → hydrated → error) without being unmounted and
-              losing its open-state. Reads ?purchase_success=1 from the
-              URL on first paint, then strips the param. */}
-          <PurchaseSuccessModal />
         </ThemeProvider>
       </body>
     </html>

canvas/src/components/PurchaseSuccessModal.tsx

@@ -1,175 +0,0 @@
-"use client";
-
-/**
- * PurchaseSuccessModal — demo-only post-purchase confirmation.
- *
- * Mounted on the canvas root (`app/page.tsx`). On first paint it inspects
- * `?purchase_success=1[&item=<name>]` on the current URL. If present, it
- * renders a centred modal styled after `ConfirmDialog`, schedules a 5s
- * auto-dismiss, and rewrites the URL via `history.replaceState` to drop
- * the params so a refresh after dismiss does NOT re-show the modal.
- *
- * Mock for the funding demo — there is no real billing surface behind
- * this. The marketplace "Purchase" button on the landing page redirects
- * here with the params; this modal is the only thing the user sees of
- * the "transaction".
- *
- * Styling matches the warm-paper @theme tokens (surface-sunken / line /
- * ink / good) so it tracks light + dark without per-mode overrides.
- */
-
-import { useEffect, useRef, useState } from "react";
-import { createPortal } from "react-dom";
-
-const AUTO_DISMISS_MS = 5000;
-
-function readPurchaseParams(): { open: boolean; item: string | null } {
-  if (typeof window === "undefined") return { open: false, item: null };
-  const sp = new URLSearchParams(window.location.search);
-  const flag = sp.get("purchase_success");
-  if (flag !== "1" && flag !== "true") return { open: false, item: null };
-  return { open: true, item: sp.get("item") };
-}
-
-function stripPurchaseParams() {
-  if (typeof window === "undefined") return;
-  const url = new URL(window.location.href);
-  url.searchParams.delete("purchase_success");
-  url.searchParams.delete("item");
-  // replaceState (not pushState) so back-button doesn't return to the
-  // pre-strip URL and re-trigger the modal.
-  window.history.replaceState({}, "", url.toString());
-}
-
-export function PurchaseSuccessModal() {
-  const [open, setOpen] = useState(false);
-  const [item, setItem] = useState<string | null>(null);
-  const [mounted, setMounted] = useState(false);
-  const dialogRef = useRef<HTMLDivElement>(null);
-
-  // Read the URL params once on mount. We don't subscribe to navigation —
-  // this modal is a one-shot for the demo redirect, not a persistent
-  // listener.
-  useEffect(() => {
-    setMounted(true);
-    const { open: shouldOpen, item: itemName } = readPurchaseParams();
-    if (shouldOpen) {
-      setOpen(true);
-      setItem(itemName);
-      // Clean the URL immediately so a refresh after the modal is closed
-      // (or even while it's still open) does NOT re-trigger it.
-      stripPurchaseParams();
-    }
-  }, []);
-
-  // Auto-dismiss timer + Escape handler.
-  useEffect(() => {
-    if (!open) return;
-    const t = window.setTimeout(() => setOpen(false), AUTO_DISMISS_MS);
-    const onKey = (e: KeyboardEvent) => {
-      if (e.key === "Escape") setOpen(false);
-    };
-    window.addEventListener("keydown", onKey);
-    // Focus the close button so keyboard users land on it after redirect.
-    const raf = requestAnimationFrame(() => {
-      dialogRef.current?.querySelector<HTMLButtonElement>("button")?.focus();
-    });
-    return () => {
-      window.clearTimeout(t);
-      window.removeEventListener("keydown", onKey);
-      cancelAnimationFrame(raf);
-    };
-  }, [open]);
-
-  if (!open || !mounted) return null;
-
-  const itemLabel = item ? decodeURIComponent(item) : "Your new agent";
-
-  return createPortal(
-    <div
-      className="fixed inset-0 z-[9999] flex items-center justify-center"
-      data-testid="purchase-success-modal"
-    >
-      {/* Backdrop — click closes, matches ConfirmDialog backdrop. */}
-      <div
-        className="absolute inset-0 bg-black/60 backdrop-blur-sm"
-        onClick={() => setOpen(false)}
-        aria-hidden="true"
-      />
-
-      <div
-        ref={dialogRef}
-        role="dialog"
-        aria-modal="true"
-        aria-labelledby="purchase-success-title"
-        className="relative bg-surface-sunken border border-line rounded-xl shadow-2xl shadow-black/50 max-w-[420px] w-full mx-4 overflow-hidden"
-      >
-        <div className="px-6 pt-6 pb-4">
-          <div className="flex items-start gap-4">
-            {/* Success glyph — uses --color-good so it tracks the theme.
-                Inline SVG over an emoji so it stays readable + on-brand
-                in both light and dark. */}
-            <div
-              className="flex h-10 w-10 flex-shrink-0 items-center justify-center rounded-full"
-              style={{
-                background:
-                  "color-mix(in srgb, var(--color-good) 15%, transparent)",
-                color: "var(--color-good)",
-              }}
-            >
-              <svg
-                width="22"
-                height="22"
-                viewBox="0 0 24 24"
-                fill="none"
-                aria-hidden="true"
-              >
-                <circle
-                  cx="12"
-                  cy="12"
-                  r="10"
-                  stroke="currentColor"
-                  strokeWidth="1.5"
-                />
-                <path
-                  d="M7.5 12.5L10.5 15.5L16.5 9.5"
-                  stroke="currentColor"
-                  strokeWidth="1.8"
-                  strokeLinecap="round"
-                  strokeLinejoin="round"
-                />
-              </svg>
-            </div>
-            <div className="flex-1">
-              <h3
-                id="purchase-success-title"
-                className="text-base font-semibold text-ink"
-              >
-                Purchase successful
-              </h3>
-              <p className="mt-1.5 text-[13px] leading-relaxed text-ink-mid">
-                <span className="font-medium text-ink">{itemLabel}</span> has
-                been added to your workspace. Provisioning starts in the
-                background — you can keep working while it spins up.
-              </p>
-            </div>
-          </div>
-        </div>
-
-        <div className="flex items-center justify-between gap-3 px-6 py-3 border-t border-line bg-surface/50">
-          <span className="font-mono text-[10.5px] uppercase tracking-[0.12em] text-ink-soft">
-            auto-dismiss · {AUTO_DISMISS_MS / 1000}s
-          </span>
-          <button
-            type="button"
-            onClick={() => setOpen(false)}
-            className="px-3.5 py-1.5 text-[13px] rounded-lg bg-accent hover:bg-accent-strong text-white transition-colors focus:outline-none focus-visible:ring-2 focus-visible:ring-offset-2 focus-visible:ring-offset-surface-sunken focus-visible:ring-accent/60"
-          >
-            Close
-          </button>
-        </div>
-      </div>
-    </div>,
-    document.body,
-  );
-}

canvas/src/components/tabs/ChatTab.tsx

@@ -13,7 +13,6 @@ import { AttachmentPreview } from "./chat/AttachmentPreview";
 import { extractFilesFromTask } from "./chat/message-parser";
 import { AgentCommsPanel } from "./chat/AgentCommsPanel";
 import { appendActivityLine } from "./chat/activityLog";
-import { activityRowToMessages, type ActivityRowForHydration } from "./chat/historyHydration";
 import { runtimeDisplayName } from "@/lib/runtime-names";
 import { ConfirmDialog } from "@/components/ConfirmDialog";
 
@@ -50,38 +49,12 @@ interface A2AResponse {
   };
 }
 
-/** Detect activity-log rows that the workspace's own runtime fired
- *  against itself but were misclassified as canvas-source. The proper
- *  fix is the X-Workspace-ID header from `self_source_headers()` in
- *  workspace/platform_auth.py, which makes the platform record
- *  source_id = workspace_id. But three failure modes still leak a
- *  self-message into "My Chat":
- *
- *    1. Historical rows already in the DB with source_id=NULL.
- *    2. Workspace containers running pre-fix heartbeat.py / main.py
- *       (the fix only takes effect after an image rebuild + redeploy).
- *    3. Future internal triggers added without the helper.
- *
- *  This client-side filter recognises the heartbeat trigger by its
- *  exact prefix — the heartbeat assembles
- *
- *    "Delegation results are ready. Review them and take appropriate
- *     action:\n" + summary_lines + report_instruction
- *
- *  in workspace/heartbeat.py. The prefix is template-fixed so a
- *  string match is reliable. If the heartbeat copy ever changes,
- *  update this constant in the same commit.
- *
- *  This is a backstop, not the primary defence — the X-Workspace-ID
- *  header is. Filtering content is fragile to copy edits, so keep
- *  the list narrow. */
-const INTERNAL_SELF_MESSAGE_PREFIXES = [
-  "Delegation results are ready. Review them and take appropriate action",
-];
-
-function isInternalSelfMessage(text: string): boolean {
-  return INTERNAL_SELF_MESSAGE_PREFIXES.some((p) => text.startsWith(p));
-}
+// Internal-self-message filtering moved server-side in RFC #2945
+// PR-C/D — the platform's /chat-history endpoint applies the
+// IsInternalSelfMessage predicate before returning rows, so the
+// client no longer needs the local backstop on the history path.
+// The proper fix is still X-Workspace-ID header (source_id=workspace_id);
+// the platform-side prefix filter handles the residual cases.
 
 // extractReplyText pulls the agent's text reply out of an A2A response.
 // Concatenates ALL text parts (joined with "\n") rather than returning
@@ -134,8 +107,19 @@ const INITIAL_HISTORY_LIMIT = 10;
 const OLDER_HISTORY_BATCH = 20;
 
 /**
- * Load chat history from the activity_logs database via the platform API.
- * Uses source=canvas to only get user-initiated messages (not agent-to-agent).
+ * Load chat history from the platform's typed /chat-history endpoint.
+ *
+ * Server-side rendering of activity_logs rows into ChatMessage shape
+ * lives in workspace-server/internal/messagestore/postgres_store.go
+ * (RFC #2945 PR-C/D). The server already applies the canvas-source
+ * filter, the internal-self-message predicate, the role decision
+ * (status=error vs agent-error prefix → system), and the v0/v1
+ * file-shape extraction. Canvas just renders what it receives.
+ *
+ * Wire shape (mirrors ChatMessage exactly, no per-row mapping needed):
+ *
+ *   GET /workspaces/:id/chat-history?limit=N&before_ts=T
+ *   200 → {"messages": ChatMessage[], "reached_end": boolean}
  *
  * Pagination:
  *  - Pass `limit` to bound the page size (newest-first from server).
@@ -143,10 +127,10 @@ const OLDER_HISTORY_BATCH = 20;
  *    timestamp. Combined with limit, this yields the next-older page
  *    when scrolling backward through history.
  *
- * `reachedEnd` is true when the server returned fewer rows than asked
- * for — caller uses this to disable further older-batch fetches.
- * (Counts row-level returns, not chat-bubble count: each row may
- * produce 1-2 bubbles.)
+ * `reachedEnd` is propagated from the server. The server computes it
+ * by comparing rowCount vs limit so a partial last page is correctly
+ * detected even when the row→bubble fan-out is non-1:1 (each row
+ * produces 1-2 bubbles).
  */
 async function loadMessagesFromDB(
   workspaceId: string,
@@ -154,25 +138,23 @@ async function loadMessagesFromDB(
   beforeTs?: string,
 ): Promise<{ messages: ChatMessage[]; error: string | null; reachedEnd: boolean }> {
   try {
-    const params = new URLSearchParams({
-      type: "a2a_receive",
-      source: "canvas",
-      limit: String(limit),
-    });
+    const params = new URLSearchParams({ limit: String(limit) });
     if (beforeTs) params.set("before_ts", beforeTs);
-    const activities = await api.get<ActivityRowForHydration[]>(
-      `/workspaces/${workspaceId}/activity?${params.toString()}`,
+    const resp = await api.get<{ messages: ChatMessage[]; reached_end: boolean }>(
+      `/workspaces/${workspaceId}/chat-history?${params.toString()}`,
     );
 
-    const messages: ChatMessage[] = [];
-    // Activities are newest-first, reverse for chronological order.
-    // Per-row mapping lives in chat/historyHydration.ts so it can be
-    // unit-tested without spinning up the full ChatTab component
-    // (regression cover for the timestamp-collapse bug).
-    for (const a of [...activities].reverse()) {
-      messages.push(...activityRowToMessages(a, isInternalSelfMessage));
-    }
-    return { messages, error: null, reachedEnd: activities.length < limit };
+    // Server emits oldest-first within the page (RFC #2945 PR-C-2
+    // post-fix: server reverses row-aware before returning so the
+    // wire is display-ready). Canvas appends/prepends without
+    // reordering — this avoids the pair-flip bug a naive flat
+    // reverse causes when each row produces a (user, agent) pair
+    // with the same timestamp.
+    return {
+      messages: resp.messages ?? [],
+      error: null,
+      reachedEnd: resp.reached_end,
+    };
   } catch (err) {
     return {
       messages: [],

canvas/src/components/tabs/ConfigTab.tsx

@@ -21,20 +21,39 @@ interface Props {
 // --- Agent Card Section ---
 
 function AgentCardSection({ workspaceId }: { workspaceId: string }) {
-  const [card, setCard] = useState<Record<string, unknown> | null>(null);
-  const [loading, setLoading] = useState(true);
+  // Initial card value comes from the canvas store — node.data.agentCard
+  // is hydrated by the platform stream when the workspace appears in the
+  // graph, so reading it here avoids a duplicate `GET /workspaces/${id}`
+  // (the parent ConfigTab.loadConfig already fetches workspace metadata,
+  // and refetching here adds a serialised RTT to the panel-open path —
+  // contributed to the ~20s detail-panel load reported in core#11).
+  // Local state still tracks the edited/saved value so the editor flow
+  // is unchanged.
+  const storeCard = useCanvasStore((s) => {
+    // Defensive against test mocks that omit `nodes` (some test files
+    // stub the store with a minimal shape). In production `nodes` is
+    // always an array — empty or not — so the optional chaining only
+    // matters for the test path.
+    const node = s.nodes?.find?.((n) => n.id === workspaceId);
+    return (node?.data.agentCard as
+      | Record<string, unknown>
+      | null
+      | undefined) ?? null;
+  });
+  const [card, setCard] = useState<Record<string, unknown> | null>(storeCard);
   const [editing, setEditing] = useState(false);
   const [draft, setDraft] = useState("");
   const [saving, setSaving] = useState(false);
   const [error, setError] = useState<string | null>(null);
   const [success, setSuccess] = useState(false);
 
+  // If the store updates while this section is mounted (another tab
+  // pushed an update via the platform event stream), reflect that —
+  // unless the user is mid-edit, in which case we don't clobber their
+  // unsaved draft.
   useEffect(() => {
-    api.get<Record<string, unknown>>(`/workspaces/${workspaceId}`)
-      .then((ws) => setCard((ws.agent_card as Record<string, unknown>) || null))
-      .catch(() => {})
-      .finally(() => setLoading(false));
-  }, [workspaceId]);
+    if (!editing) setCard(storeCard);
+  }, [storeCard, editing]);
 
   const handleSave = async () => {
     setError(null);
@@ -53,9 +72,7 @@ function AgentCardSection({ workspaceId }: { workspaceId: string }) {
 
   return (
     <Section title="Agent Card" defaultOpen={false}>
-      {loading ? (
-        <div className="text-[10px] text-ink-soft">Loading...</div>
-      ) : editing ? (
+      {editing ? (
         <div className="space-y-2">
           <textarea
             aria-label="Agent card JSON editor"
@@ -221,47 +238,51 @@ export function ConfigTab({ workspaceId }: Props) {
     setLoading(true);
     setError(null);
 
-    // ALWAYS load workspace metadata first (runtime + model). These are the
-    // source of truth regardless of whether the runtime uses our config.yaml
-    // template. Without this the form falls back to empty/default values on
-    // a hermes workspace (which doesn't use our template), creating the
-    // appearance that the saved runtime is unset — and worse, clicking Save
-    // would silently flip `runtime` from `hermes` back to the dropdown
-    // default `LangGraph`. See GH #1894.
-    let wsMetadataRuntime = "";
-    let wsMetadataModel = "";
-    let wsMetadataTier: number | null = null;
-    try {
-      const ws = await api.get<{ runtime?: string; tier?: number }>(`/workspaces/${workspaceId}`);
-      wsMetadataRuntime = (ws.runtime || "").trim();
-      if (typeof ws.tier === "number") wsMetadataTier = ws.tier;
-    } catch { /* fall back to config.yaml */ }
-    try {
-      const m = await api.get<{ model?: string }>(`/workspaces/${workspaceId}/model`);
-      wsMetadataModel = (m.model || "").trim();
-    } catch { /* non-fatal */ }
+    // Load workspace metadata (runtime + model + provider) in parallel.
+    // These are independent GETs against three workspace-server endpoints
+    // and used to be awaited serially — for SaaS workspaces each call
+    // round-trips through an EIC SSH tunnel, so the previous serial
+    // pattern stacked 3-5s of tunnel-setup latency per call (core#11).
+    // Promise.all overlaps them; the per-call cost stays the same but
+    // wall time drops to max() instead of sum().
+    //
+    // Each leg has its own .catch handler that yields a sentinel value,
+    // matching the previous semantics:
+    //   - /workspaces/${id}: required source-of-truth for runtime+tier;
+    //     fall back to YAML if the GET fails (rare, network-class only).
+    //   - /workspaces/${id}/model: non-fatal; empty model lets the form
+    //     fall through to YAML runtime_config.model.
+    //   - /workspaces/${id}/provider: non-fatal; old workspace-servers
+    //     return 404, in which case provider="" and Save skips the PUT.
+    //
+    // See GH #1894 for the workspace-row-as-source-of-truth rationale
+    // that motivated splitting from a single config.yaml read.
+    const [wsRes, modelRes, providerRes] = await Promise.all([
+      api.get<{ runtime?: string; tier?: number }>(`/workspaces/${workspaceId}`)
+        .catch(() => ({} as { runtime?: string; tier?: number })),
+      api.get<{ model?: string }>(`/workspaces/${workspaceId}/model`)
+        .catch(() => ({} as { model?: string })),
+      api.get<{ provider?: string }>(`/workspaces/${workspaceId}/provider`)
+        .catch(() => null),
+    ]);
+    const wsMetadataRuntime = (wsRes.runtime || "").trim();
+    const wsMetadataModel = (modelRes.model || "").trim();
+    const wsMetadataTier: number | null =
+      typeof wsRes.tier === "number" ? wsRes.tier : null;
+    if (providerRes !== null) {
+      const loadedProvider = (providerRes.provider || "").trim();
+      setProvider(loadedProvider);
+      setOriginalProvider(loadedProvider);
+    } else {
+      setProvider("");
+      setOriginalProvider("");
+    }
     // originalModel is set further down once the YAML has been parsed —
     // we want it to reflect what the form ACTUALLY rendered, which may
     // be the YAML's runtime_config.model fallback when MODEL_PROVIDER
     // is empty. Setting it here from wsMetadataModel alone would be
     // wrong for hermes/pre-#240 workspaces.
 
-    // Load explicit provider override (Option B PR-5). Endpoint returns
-    // {provider: "", source: "default"} when no override is set, so the
-    // empty string is the legitimate "auto-derive" signal — don't treat
-    // it as a load error. Non-fatal: an older workspace-server that
-    // predates PR-2 returns 404 here; the form falls back to "" and
-    // Save just won't PUT the provider field.
-    try {
-      const p = await api.get<{ provider?: string }>(`/workspaces/${workspaceId}/provider`);
-      const loadedProvider = (p.provider || "").trim();
-      setProvider(loadedProvider);
-      setOriginalProvider(loadedProvider);
-    } catch {
-      setProvider("");
-      setOriginalProvider("");
-    }
-
     // Skip the config.yaml fetch entirely for runtimes that manage
     // their own config (external, hermes, etc.) — they don't have a
     // platform-side template, so the GET would 404. The catch block

canvas/src/components/tabs/__tests__/ChatTab.lazyHistory.test.tsx

@@ -1,13 +1,11 @@
 // @vitest-environment jsdom
 //
-// Pins the lazy-loading chat-history pagination added 2026-05-05.
+// Pins the lazy-loading chat-history pagination.
 //
-// Pre-fix: ChatTab fetched the newest 50 messages on every mount and
-// scrolled to bottom, paying full DOM cost up-front even when the user
-// only wanted to read the last few bubbles. Post-fix: initial load is
-// bounded to 10 newest, and an IntersectionObserver on a top sentinel
-// triggers loadOlder() (batch of 20 with `before_ts` cursor) when the
-// user scrolls up.
+// PR-C-2 (RFC #2945): canvas was migrated from /activity?type=a2a_receive
+// to /chat-history. Server now returns typed ChatMessage[] in
+// display-ready oldest-first order. These tests guard the canvas-side
+// pagination invariants against the new endpoint surface.
 //
 // Pinned branches:
 //   1. Initial fetch carries `limit=10` and NO before_ts (newest-first
@@ -20,11 +18,10 @@
 //      asserting the rendered bubble count matches the full page).
 //   4. The retry button after a failed initial load uses the same
 //      INITIAL_HISTORY_LIMIT (10), not the legacy 50.
-//
-// IntersectionObserver / scroll-anchor restoration is exercised by the
-// E2E synth-canary suite — pinning it in jsdom would require mocking
-// the observer and faking layout, which is brittler than trusting a
-// live-DOM canary against the staging tenant.
+//   5. before_ts cursor is the OLDEST timestamp from the current page,
+//      passed verbatim to walk backward.
+//   6. Inflight guard rejects duplicate IO triggers while a loadOlder
+//      fetch is in flight.
 
 import { describe, it, expect, vi, afterEach, beforeEach } from "vitest";
 import { render, screen, cleanup, waitFor, fireEvent } from "@testing-library/react";
@@ -33,24 +30,31 @@ import React from "react";
 afterEach(cleanup);
 
 // Both ChatTab sub-panels (MyChat + AgentComms) mount simultaneously so
-// keyboard tab order and aria-controls land on a real DOM. Both fire
-// /activity GETs on mount: MyChat's hits `type=a2a_receive&source=canvas`,
-// AgentComms's hits a different filter. Route the mock by URL so each
-// gets a sensible default and only MyChat's call is what the assertions
-// scrutinise.
-const myChatActivityCalls: string[] = [];
-let myChatNextResponse: { ok: true; rows: unknown[] } | { ok: false; err: Error } = {
-  ok: true,
-  rows: [],
-};
+// keyboard tab order and aria-controls land on a real DOM. MyChat's
+// loadMessagesFromDB hits /chat-history; AgentComms's polling hits a
+// different URL. Route the mock by URL so each gets a sensible default
+// and only MyChat's calls land in the assertion array.
+const myChatHistoryCalls: string[] = [];
+let myChatNextResponse:
+  | { ok: true; messages: unknown[]; reachedEnd?: boolean }
+  | { ok: false; err: Error } = { ok: true, messages: [] };
+
 const apiGet = vi.fn((path: string): Promise<unknown> => {
-  if (path.includes("type=a2a_receive") && path.includes("source=canvas")) {
-    myChatActivityCalls.push(path);
-    if (myChatNextResponse.ok) return Promise.resolve(myChatNextResponse.rows);
+  if (path.includes("/chat-history")) {
+    myChatHistoryCalls.push(path);
+    if (myChatNextResponse.ok) {
+      const reached_end =
+        myChatNextResponse.reachedEnd !== undefined
+          ? myChatNextResponse.reachedEnd
+          : myChatNextResponse.messages.length < 10;
+      return Promise.resolve({
+        messages: myChatNextResponse.messages,
+        reached_end,
+      });
+    }
     return Promise.reject(myChatNextResponse.err);
   }
-  // AgentComms / heartbeat / anything else — empty array is a safe
-  // default that won't blow up the corresponding component's .then().
+  // AgentComms / heartbeat / anything else — empty array safe default.
   return Promise.resolve([]);
 });
 const apiPost = vi.fn();
@@ -84,8 +88,8 @@ const ioInstances: IOInstance[] = [];
 beforeEach(() => {
   apiGet.mockClear();
   apiPost.mockReset();
-  myChatActivityCalls.length = 0;
-  myChatNextResponse = { ok: true, rows: [] };
+  myChatHistoryCalls.length = 0;
+  myChatNextResponse = { ok: true, messages: [] };
   ioInstances.length = 0;
   class FakeIO {
     private inst: IOInstance;
@@ -101,20 +105,12 @@ beforeEach(() => {
       this.inst.disconnected = true;
     }
   }
-  // Install on every reachable global — different bundlers / module
-  // graphs can resolve `IntersectionObserver` via `window`, `globalThis`,
-  // or the bare global. Without all three, jsdom's own (pre-existing)
-  // stub silently wins and ioInstances stays empty.
   (window as unknown as { IntersectionObserver: unknown }).IntersectionObserver = FakeIO;
   (globalThis as unknown as { IntersectionObserver: unknown }).IntersectionObserver = FakeIO;
-  // jsdom doesn't implement scrollIntoView; ChatTab calls it after every
-  // messages update.
   Element.prototype.scrollIntoView = vi.fn();
 });
 
 function triggerIntersection(instanceIdx = -1) {
-  // -1 → the latest observer (the live one). Tests targeting an old
-  // (disconnected) instance pass a positive index.
   const inst = ioInstances.at(instanceIdx);
   if (!inst) throw new Error(`no IO instance at ${instanceIdx}`);
   inst.callback(
@@ -125,25 +121,30 @@ function triggerIntersection(instanceIdx = -1) {
 
 import { ChatTab } from "../ChatTab";
 
-function makeActivityRow(seq: number): Record<string, unknown> {
-  // Zero-pad seq into the minute slot so "seq=10" doesn't produce
-  // the invalid timestamp "00:010:00Z" (caught by the loadOlder URL
-  // assertion below — first version of the helper used `0${seq}` and
-  // the test failed on `before_ts` having an extra digit).
+// makeMessagePair returns a (user, agent) pair sharing a timestamp,
+// matching the wire shape /chat-history emits per activity_logs row.
+// Server-side reverseRowChunks ensures the wire is oldest-first across
+// rows but [user, agent] within each row.
+function makeMessagePair(seq: number): unknown[] {
+  // Zero-pad seq into the minute slot so seq=10 produces a valid
+  // timestamp (00:10:00Z, not 00:010:00Z).
   const mm = String(seq).padStart(2, "0");
-  return {
-    activity_type: "a2a_receive",
-    status: "ok",
-    created_at: `2026-05-05T00:${mm}:00Z`,
-    request_body: { params: { message: { parts: [{ kind: "text", text: `user msg ${seq}` }] } } },
-    response_body: { result: `agent reply ${seq}` },
-  };
+  const ts = `2026-05-05T00:${mm}:00Z`;
+  return [
+    { id: `u-${seq}`, role: "user", content: `user msg ${seq}`, timestamp: ts },
+    { id: `a-${seq}`, role: "agent", content: `agent reply ${seq}`, timestamp: ts },
+  ];
 }
 
-// Server returns newest-first; the helper builds a server-shape page
-// so the order in the rendered messages array matches production.
-function newestFirstPage(start: number, count: number): unknown[] {
-  return Array.from({ length: count }, (_, i) => makeActivityRow(start + count - 1 - i));
+// pageOldestFirst builds a wire-shape page (oldest-first within page)
+// of `count` row-pairs starting at seq=`start`. Mirrors the server's
+// post-reverseRowChunks emission order.
+function pageOldestFirst(start: number, count: number): unknown[] {
+  const out: unknown[] = [];
+  for (let i = 0; i < count; i++) {
+    out.push(...makeMessagePair(start + i));
+  }
+  return out;
 }
 
 const minimalData = {
@@ -153,28 +154,30 @@ const minimalData = {
 } as unknown as Parameters<typeof ChatTab>[0]["data"];
 
 describe("ChatTab lazy history pagination", () => {
-  it("initial fetch carries limit=10 (not the legacy 50)", async () => {
-    myChatNextResponse = { ok: true, rows: [makeActivityRow(1)] };
+  it("initial fetch carries limit=10 (not the legacy 50) and hits /chat-history", async () => {
+    myChatNextResponse = { ok: true, messages: makeMessagePair(1) };
     render(<ChatTab workspaceId="ws-1" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
-    const url = myChatActivityCalls[0];
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
+    const url = myChatHistoryCalls[0];
+    expect(url).toContain("/chat-history");
     expect(url).toContain("limit=10");
     expect(url).not.toContain("limit=50");
     // before_ts should NOT be set on the initial fetch — that's the
     // newest-first slice the user lands on.
     expect(url).not.toContain("before_ts");
+    // /chat-history filters source-canvas server-side; client should
+    // NOT pass type/source params (they belonged to /activity).
+    expect(url).not.toContain("type=a2a_receive");
+    expect(url).not.toContain("source=canvas");
   });
 
   it("hides the top sentinel when initial fetch returns fewer than the limit", async () => {
     // 3 < 10 → server says "no more older history exists"; sentinel
     // should NOT mount and the "Loading older messages…" line should
-    // never appear (it can't, since the sentinel is what triggers it).
-    myChatNextResponse = {
-      ok: true,
-      rows: [makeActivityRow(1), makeActivityRow(2), makeActivityRow(3)],
-    };
+    // never appear.
+    myChatNextResponse = { ok: true, messages: pageOldestFirst(1, 3) };
     render(<ChatTab workspaceId="ws-2" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
     await waitFor(() => {
       expect(screen.queryByText(/Loading chat history/i)).toBeNull();
     });
@@ -182,15 +185,15 @@ describe("ChatTab lazy history pagination", () => {
   });
 
   it("renders all messages when initial fetch returns exactly the limit", async () => {
-    // 10 == limit → server might have more older rows; sentinel SHOULD
-    // mount so the IO observer can fire loadOlder() on scroll-up. We
-    // verify by checking the rendered bubble count — if hasMore stayed
-    // true the sentinel render path doesn't crash and all 10 rows
-    // produced their pair of bubbles.
-    const fullPage = Array.from({ length: 10 }, (_, i) => makeActivityRow(i + 1));
-    myChatNextResponse = { ok: true, rows: fullPage };
+    // limit=10 row-pairs → 20 ChatMessages. reachedEnd should be FALSE
+    // so the sentinel mounts. Verified by bubble counts.
+    myChatNextResponse = {
+      ok: true,
+      messages: pageOldestFirst(1, 10),
+      reachedEnd: false,
+    };
     render(<ChatTab workspaceId="ws-3" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
     await waitFor(() => {
       expect(screen.queryByText(/Loading chat history/i)).toBeNull();
     });
@@ -202,54 +205,67 @@ describe("ChatTab lazy history pagination", () => {
     myChatNextResponse = { ok: false, err: new Error("network down") };
     render(<ChatTab workspaceId="ws-4" data={minimalData} />);
     const retry = await screen.findByText(/Retry/);
-    myChatNextResponse = { ok: true, rows: [makeActivityRow(1)] };
+    myChatNextResponse = { ok: true, messages: makeMessagePair(1) };
     fireEvent.click(retry);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(2));
-    const retryUrl = myChatActivityCalls[1];
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(2));
+    const retryUrl = myChatHistoryCalls[1];
+    expect(retryUrl).toContain("/chat-history");
     expect(retryUrl).toContain("limit=10");
     expect(retryUrl).not.toContain("limit=50");
   });
 
   it("loadOlder fetches limit=20 with before_ts=oldest.timestamp", async () => {
-    // Initial page = 10 rows in newest-first order (seq 10..1). After
-    // the component reverses to oldest-first for display, messages[0]
-    // is built from seq=1 — the oldest — and its timestamp is what
-    // before_ts should carry.
-    myChatNextResponse = { ok: true, rows: newestFirstPage(1, 10) };
+    // Initial page = 10 row-pairs in oldest-first order (seq 1..10).
+    // The oldest (and so the cursor for loadOlder) is seq=1's
+    // timestamp 2026-05-05T00:01:00Z.
+    myChatNextResponse = {
+      ok: true,
+      messages: pageOldestFirst(1, 10),
+      reachedEnd: false,
+    };
     render(<ChatTab workspaceId="ws-load-older" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
     await waitFor(() => expect(ioInstances.length).toBeGreaterThan(0));
 
-    // Stage the older-batch response, then fire the IO callback.
-    myChatNextResponse = { ok: true, rows: newestFirstPage(0, 1) };
+    // Stage older-batch response, then fire IO callback.
+    myChatNextResponse = {
+      ok: true,
+      messages: pageOldestFirst(0, 1),
+      reachedEnd: true,
+    };
     triggerIntersection();
 
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(2));
-    const olderUrl = myChatActivityCalls[1];
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(2));
+    const olderUrl = myChatHistoryCalls[1];
+    expect(olderUrl).toContain("/chat-history");
     expect(olderUrl).toContain("limit=20");
     expect(olderUrl).toContain("before_ts=");
     expect(decodeURIComponent(olderUrl)).toContain("before_ts=2026-05-05T00:01:00Z");
   });
 
   it("inflight guard rejects a second IO trigger while first loadOlder is in flight", async () => {
-    myChatNextResponse = { ok: true, rows: newestFirstPage(1, 10) };
+    myChatNextResponse = {
+      ok: true,
+      messages: pageOldestFirst(1, 10),
+      reachedEnd: false,
+    };
     render(<ChatTab workspaceId="ws-inflight" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
     await waitFor(() => expect(ioInstances.length).toBeGreaterThan(0));
 
     // Hold the next loadOlder fetch open with a manual deferred so we
     // can fire the second trigger while the first is in-flight.
-    let release!: (rows: unknown[]) => void;
-    const deferred = new Promise<unknown[]>((res) => {
+    let release!: (resp: unknown) => void;
+    const deferred = new Promise<unknown>((res) => {
       release = res;
     });
     apiGet.mockImplementationOnce((path: string): Promise<unknown> => {
-      myChatActivityCalls.push(path);
+      myChatHistoryCalls.push(path);
       return deferred;
     });
 
     triggerIntersection(); // start loadOlder #1
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(2));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(2));
 
     // Second IO trigger lands while #1 is still pending.
     triggerIntersection();
@@ -258,79 +274,62 @@ describe("ChatTab lazy history pagination", () => {
     // Without the inflight guard, each of these would have started a
     // new fetch. With the guard, none of them do — call count stays 2.
     await new Promise((r) => setTimeout(r, 10));
-    expect(myChatActivityCalls.length).toBe(2);
+    expect(myChatHistoryCalls.length).toBe(2);
 
-    // Release the first fetch. Inflight clears in the finally block;
-    // a subsequent IO trigger is permitted again (verified by checking
-    // we can fire a follow-up after release without hanging the test).
-    release([]);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(2));
+    // Release the first fetch with a valid wire response shape.
+    release({ messages: [], reached_end: true });
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(2));
   });
 
   it("empty older response clears the scroll anchor and unmounts the sentinel", async () => {
-    // The bug we're pinning: if loadOlder returns 0 rows, the
-    // scrollAnchorRef must be cleared so the next paint doesn't try to
-    // restore against a no-op prepend (which would fight the natural
-    // bottom-pin for any subsequent live message). hasMore flipping to
-    // false is the same flag-flip path; sentinel disappearing is the
-    // observable proxy.
-    myChatNextResponse = { ok: true, rows: newestFirstPage(1, 10) };
+    myChatNextResponse = {
+      ok: true,
+      messages: pageOldestFirst(1, 10),
+      reachedEnd: false,
+    };
     render(<ChatTab workspaceId="ws-anchor" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
     await waitFor(() => expect(ioInstances.length).toBeGreaterThan(0));
 
-    myChatNextResponse = { ok: true, rows: [] }; // empty → reachedEnd
+    myChatNextResponse = {
+      ok: true,
+      messages: [],
+      reachedEnd: true,
+    };
     triggerIntersection();
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(2));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(2));
 
-    // After reachedEnd the sentinel unmounts (hasMore=false). We can't
-    // peek scrollAnchorRef directly, but we can assert the consequence:
-    // scrollIntoView (the bottom-pin for live appends) is not blocked
-    // by a stale anchor. Trigger a re-render via an unrelated state
-    // change… in practice the safest assertion here is that the
-    // sentinel disappeared (proving the empty response propagated to
-    // hasMore correctly, which is the same flag-flip path as anchor
-    // clearing).
     await waitFor(() => {
       expect(screen.queryByText(/Loading older messages/i)).toBeNull();
     });
   });
 
   it("IntersectionObserver does not churn when older messages prepend", async () => {
-    // Whole-PR perf invariant: prepending older history (the load-bearing
-    // user gesture) must NOT tear down + re-arm the IO observer.
-    // Triggering loadOlder is the cleanest way to drive a messages
-    // mutation from inside the test, since live agent push goes through
-    // a Zustand store that's harder to drive reliably from jsdom.
-    //
-    // Pre-fix, loadOlder depended on `messages`, so every prepend
-    // recreated loadOlder → re-ran the IO effect → new observer. Each
-    // call to triggerIntersection() produced a fresh disconnected
-    // observer + a new live one. Post-fix, the observer survives.
-    myChatNextResponse = { ok: true, rows: newestFirstPage(1, 10) };
+    myChatNextResponse = {
+      ok: true,
+      messages: pageOldestFirst(1, 10),
+      reachedEnd: false,
+    };
     render(<ChatTab workspaceId="ws-stable-io" data={minimalData} />);
-    await waitFor(() => expect(myChatActivityCalls.length).toBe(1));
+    await waitFor(() => expect(myChatHistoryCalls.length).toBe(1));
     await waitFor(() => expect(ioInstances.length).toBeGreaterThan(0));
 
-    // Snapshot the observer instance after first paint stabilises.
     const observerBefore = ioInstances.at(-1);
     expect(observerBefore).toBeDefined();
     expect(observerBefore!.disconnected).toBe(false);
 
     // Trigger three older-batch prepends. Each batch returns the full
-    // OLDER_HISTORY_BATCH (20 rows) so reachedEnd stays false and the
-    // sentinel keeps mounting. Pre-fix, each prepend mutated `messages`
-    // → recreated loadOlder → re-ran the IO effect → new observer.
+    // OLDER_HISTORY_BATCH (20 row-pairs = 40 messages) so reachedEnd
+    // stays false and the sentinel keeps mounting.
     for (let batch = 0; batch < 3; batch++) {
       myChatNextResponse = {
         ok: true,
-        rows: newestFirstPage(-(batch + 1) * 20, 20),
+        messages: pageOldestFirst(-(batch + 1) * 20, 20),
+        reachedEnd: false,
       };
-      const callsBefore = myChatActivityCalls.length;
+      const callsBefore = myChatHistoryCalls.length;
       triggerIntersection();
-      await waitFor(() =>
-        expect(myChatActivityCalls.length).toBe(callsBefore + 1),
-      );
+      await waitFor(() => expect(myChatHistoryCalls.length).toBe(callsBefore + 1));
     }
 
     // The original observer is still the live one — no churn.

docker-compose.yml

@@ -212,8 +212,8 @@ services:
     #   docker compose pull canvas && docker compose up -d canvas
     # First-time local setup or testing unreleased changes — build from source:
     #   docker compose build canvas && docker compose up -d canvas
-    # Note: GHCR images are private — `docker login ghcr.io` required before pull.
-    image: ghcr.io/molecule-ai/canvas:latest
+    # Note: ECR images require AWS auth — `aws ecr get-login-password --region us-east-2 | docker login --username AWS --password-stdin 153263036946.dkr.ecr.us-east-2.amazonaws.com` before pull.
+    image: 153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/canvas:latest
     build:
       context: ./canvas
       dockerfile: Dockerfile

docs/architecture/canary-release.md

@@ -4,7 +4,7 @@ How a workspace-server code change reaches the prod tenant fleet — and how to
 
 > **⚠️ 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.
 >
-> Current merges gate on manual `promote-latest.yml` dispatches, not canary. See [molecule-controlplane/docs/canary-tenants.md](https://github.com/Molecule-AI/molecule-controlplane/blob/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.
+> 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.
 >
 > **Account-specific identifiers (AWS account ID, IAM role name) referenced below in the original design have been redacted from this public doc.** The actual values — if they exist — are in `Molecule-AI/internal/runbooks/canary-fleet.md`. If you're implementing Phase 2, start there.
 >

docs/architecture/molecule-technical-doc.md

@@ -1,7 +1,7 @@
 # Molecule AI — Comprehensive Technical Documentation
 
 > Definitive technical reference for the Molecule AI Agent Team platform.
-> Based on a full non-invasive scan of the [molecule-monorepo](https://github.com/Molecule-AI/molecule-monorepo) repository.
+> Based on a full non-invasive scan of the [molecule-monorepo](https://git.moleculesai.app/molecule-ai/molecule-monorepo) repository.
 
 ---
 
@@ -1149,11 +1149,11 @@ Molecule AI's workspace abstraction is **runtime-agnostic by design**. A workspa
 
 ## Links
 
-- **GitHub**: https://github.com/Molecule-AI/molecule-monorepo
-- **Architecture Docs**: https://github.com/Molecule-AI/molecule-monorepo/tree/main/docs/architecture
-- **API Protocol**: https://github.com/Molecule-AI/molecule-monorepo/tree/main/docs/api-protocol
-- **Agent Runtime**: https://github.com/Molecule-AI/molecule-monorepo/tree/main/docs/agent-runtime
-- **Product Docs**: https://github.com/Molecule-AI/molecule-monorepo/tree/main/docs/product
+- **GitHub**: https://git.moleculesai.app/molecule-ai/molecule-monorepo
+- **Architecture Docs**: https://git.moleculesai.app/molecule-ai/molecule-monorepo/src/branch/main/docs/architecture
+- **API Protocol**: https://git.moleculesai.app/molecule-ai/molecule-monorepo/src/branch/main/docs/api-protocol
+- **Agent Runtime**: https://git.moleculesai.app/molecule-ai/molecule-monorepo/src/branch/main/docs/agent-runtime
+- **Product Docs**: https://git.moleculesai.app/molecule-ai/molecule-monorepo/src/branch/main/docs/product
 
 ---
 

docs/architecture/secrets-key-custody.md

@@ -79,7 +79,7 @@ For SOC2 / ISO 27001 / customer security questionnaires:
 
 ## Pointers
 
-- KMS envelope code: [`molecule-controlplane/internal/crypto/kms.go`](https://github.com/Molecule-AI/molecule-controlplane/blob/main/internal/crypto/kms.go)
-- Static-key fallback: [`molecule-controlplane/internal/crypto/aes.go`](https://github.com/Molecule-AI/molecule-controlplane/blob/main/internal/crypto/aes.go)
+- KMS envelope code: [`molecule-controlplane/internal/crypto/kms.go`](https://git.moleculesai.app/molecule-ai/molecule-controlplane/src/branch/main/internal/crypto/kms.go)
+- Static-key fallback: [`molecule-controlplane/internal/crypto/aes.go`](https://git.moleculesai.app/molecule-ai/molecule-controlplane/src/branch/main/internal/crypto/aes.go)
 - Tenant secrets handler: [`workspace-server/internal/crypto/aes.go`](../../workspace-server/internal/crypto/aes.go)
 - Tenant secrets schema: [database-schema.md](./database-schema.md#workspace_secrets)

docs/assets/branding/molecule-icon.svg

@@ -0,0 +1,28 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64">
+  <style>
+    .bg { fill: #0a1120; }
+    .accent { fill: #7fe8d6; }
+    .accent-stroke { stroke: #7fe8d6; }
+    @media (prefers-color-scheme: light) {
+      .bg { fill: #f5f7fa; }
+      .accent { fill: #1a8a72; }
+      .accent-stroke { stroke: #1a8a72; }
+    }
+  </style>
+  <rect class="bg" width="64" height="64" rx="14"/>
+  <g class="accent-stroke" stroke-width="2.4" stroke-linecap="round" fill="none">
+    <line x1="32" y1="32" x2="12" y2="14"/>
+    <line x1="32" y1="32" x2="52" y2="18"/>
+    <line x1="32" y1="32" x2="10" y2="40"/>
+    <line x1="32" y1="32" x2="54" y2="44"/>
+    <line x1="32" y1="32" x2="32" y2="56"/>
+  </g>
+  <g class="accent">
+    <circle cx="32" cy="32" r="6.5"/>
+    <circle cx="12" cy="14" r="3.5"/>
+    <circle cx="52" cy="18" r="3.5"/>
+    <circle cx="10" cy="40" r="3.5"/>
+    <circle cx="54" cy="44" r="3.5"/>
+    <circle cx="32" cy="56" r="3.5"/>
+  </g>
+</svg>

docs/assets/branding/molecule-logo.svg

@@ -0,0 +1,17 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64" role="img" aria-label="Molecule AI">
+  <g stroke="#7fe8d6" stroke-width="2.6" stroke-linecap="round" fill="none">
+    <line x1="32" y1="32" x2="12" y2="14"/>
+    <line x1="32" y1="32" x2="52" y2="18"/>
+    <line x1="32" y1="32" x2="10" y2="40"/>
+    <line x1="32" y1="32" x2="54" y2="44"/>
+    <line x1="32" y1="32" x2="32" y2="56"/>
+  </g>
+  <g fill="#7fe8d6">
+    <circle cx="32" cy="32" r="7"/>
+    <circle cx="12" cy="14" r="3.6"/>
+    <circle cx="52" cy="18" r="3.6"/>
+    <circle cx="10" cy="40" r="3.6"/>
+    <circle cx="54" cy="44" r="3.6"/>
+    <circle cx="32" cy="56" r="3.6"/>
+  </g>
+</svg>

docs/blog/2026-04-20-chrome-devtools-mcp-seo/index.md

@@ -299,8 +299,8 @@ Or use the Canvas UI: Workspace → Config → MCP Servers → Add browser MCP s
 
 **Try it free** — Molecule AI is open source and self-hostable. Get a workspace running in under 5 minutes.
 
-→ [Get started on GitHub →](https://github.com/Molecule-AI/molecule-core)
+→ [Get started on GitHub →](https://git.moleculesai.app/molecule-ai/molecule-core)
 
 ---
 
-*Have a browser automation use case you want to see covered? Open a discussion on [GitHub Discussions](https://github.com/Molecule-AI/molecule-core/discussions) — or file an issue with the `enhancement` label.*
+*Have a browser automation use case you want to see covered? File an issue with the `enhancement` label on the [molecule-core issue tracker](https://git.moleculesai.app/molecule-ai/molecule-core/issues).*

docs/blog/2026-04-20-remote-workspaces/index.md

@@ -148,7 +148,7 @@ Then follow the [quick-start guide](/docs/guides/remote-workspaces.md).
 Or run the annotated example directly:
 
 ```bash
-git clone https://github.com/Molecule-AI/molecule-sdk-python
+git clone https://git.moleculesai.app/molecule-ai/molecule-sdk-python
 cd molecule-sdk-python/examples/remote-agent
 # Create workspace with runtime:external, grab the ID, then:
 WORKSPACE_ID=<your-id> PLATFORM_URL=https://acme.moleculesai.app python3 run.py
@@ -160,6 +160,6 @@ The agent appears on the canvas within seconds.
 
 → [Remote Workspaces Guide →](/docs/guides/remote-workspaces.md)
 → [External Agent Registration Reference →](/docs/guides/external-agent-registration.md)
-→ [molecule-sdk-python →](https://github.com/Molecule-AI/molecule-sdk-python)
+→ [molecule-sdk-python →](https://git.moleculesai.app/molecule-ai/molecule-sdk-python)
 
 *Phase 30 shipped in PRs #1075–#1083 and #1085–#1100 on `molecule-core`.*

docs/blog/2026-04-22-a2a-v1-agent-platform/index.md

@@ -133,4 +133,4 @@ With protocol-native A2A, you get:
 
 Molecule AI's external agent registration is production-ready. Documentation is live at [External Agent Registration Guide](https://docs.molecule.ai/docs/guides/external-agent-registration). The npm package for the MCP server is available at [`@molecule-ai/mcp-server`](https://www.npmjs.com/package/@molecule-ai/mcp-server).
 
-Read the full [A2A v1.0 protocol spec](https://github.com/Molecule-AI/molecule-core/blob/main/docs/api-protocol/a2a-protocol.md) on GitHub.
+Read the full [A2A v1.0 protocol spec](https://git.moleculesai.app/molecule-ai/molecule-core/src/branch/main/docs/api-protocol/a2a-protocol.md) on GitHub.

docs/blog/2026-04-22-remote-workspaces/index.md

@@ -45,7 +45,7 @@ canonicalUrl: "https://docs.molecule.ai/blog/remote-workspaces"
   " proficiencyLevel": "Expert",
   "genre": ["technical documentation", "product announcement"],
   "sameAs": [
-    "https://github.com/Molecule-AI/molecule-core",
+    "https://git.moleculesai.app/molecule-ai/molecule-core",
     "https://molecule.ai"
   ]
 }
@@ -270,7 +270,7 @@ Configure it in your project's `.mcp.json` and any AI agent (Claude Code, Cursor
 
 → [External Agent Registration Guide](/docs/guides/external-agent-registration) — full step-by-step with Python and Node.js reference implementations
 
-→ [GitHub: molecule-core](https://github.com/Molecule-AI/molecule-core) — source and issues
+→ [GitHub: molecule-core](https://git.moleculesai.app/molecule-ai/molecule-core) — source and issues
 
 → [Phase 30 Launch Thread on X](https://x.com) — follow for updates
 

docs/blog/a2a-v1-production-reference-2026-04-24.md

@@ -170,4 +170,4 @@ The `staging` branch is now on `a2a-sdk` 1.0.0. The `main` branch still carries
 
 If you're running `a2a-sdk` 0.3.x and planning the 1.0.0 migration, this post is the reference. The four breaking changes are well-contained, the migration is a single PR, and the eight smoke scenarios above will tell you whether the upgrade is clean before you merge.
 
-Questions? The [A2A protocol spec](https://github.com/google-a2a/a2a-specification) is the authoritative source. For Molecule AI's production A2A implementation, see [External Agent Registration](https://docs.molecule.ai/docs/guides/external-agent-registration) or open an issue in the [molecule-core](https://github.com/Molecule-AI/molecule-core) repo.
+Questions? The [A2A protocol spec](https://github.com/google-a2a/a2a-specification) is the authoritative source. For Molecule AI's production A2A implementation, see [External Agent Registration](https://docs.molecule.ai/docs/guides/external-agent-registration) or open an issue in the [molecule-core](https://git.moleculesai.app/molecule-ai/molecule-core) repo.

docs/guides/external-workspace-quickstart.md

@@ -215,7 +215,7 @@ Push mode (this guide) works today but requires an inbound-reachable URL — whi
 
 Your agent makes only outbound HTTPS calls to the platform, pulling messages from an inbox queue and posting replies back. Works behind any NAT/firewall, tolerates offline laptops, no tunnel needed.
 
-See the [design doc](https://github.com/Molecule-AI/internal/blob/main/product/external-workspaces-polling.md) (internal) and [implementation tracking issue](https://github.com/Molecule-AI/molecule-core/issues?q=polling+mode) once opened.
+See the [design doc](https://git.moleculesai.app/molecule-ai/internal/src/branch/main/product/external-workspaces-polling.md) (internal) and the implementation tracking issue (search `polling+mode` on the [molecule-core issue tracker](https://git.moleculesai.app/molecule-ai/molecule-core/issues)).
 
 ---
 

docs/guides/remote-workspaces.md

@@ -143,5 +143,5 @@ The agent appears on the canvas with a **purple REMOTE badge** within seconds. F
 ## Next Steps
 
 - **[External Agent Registration Guide →](/docs/guides/external-agent-registration)** — full endpoint reference, Python + Node.js examples, troubleshooting
-- **[molecule-sdk-python →](https://github.com/Molecule-AI/molecule-sdk-python)** — SDK source, `RemoteAgentClient` API docs
-- **[SDK Examples →](https://github.com/Molecule-AI/molecule-sdk-python/tree/main/examples/remote-agent)** — `run.py` demo script, annotated walkthrough
+- **[molecule-sdk-python →](https://git.moleculesai.app/molecule-ai/molecule-sdk-python)** — SDK source, `RemoteAgentClient` API docs
+- **[SDK Examples →](https://git.moleculesai.app/molecule-ai/molecule-sdk-python/src/branch/main/examples/remote-agent)** — `run.py` demo script, annotated walkthrough

docs/guides/skill-catalog.md

@@ -61,7 +61,7 @@ molecule skills install arxiv-research --from community
 
 Community skills are reviewed by the Molecule AI team before being
 listed. Submit a skill for review by opening a PR against
-[`molecule-ai/skills`](https://github.com/Molecule-AI/skills).
+[`molecule-ai/skills`](https://git.moleculesai.app/molecule-ai/skills).
 
 ## Installing via config.yaml
 
@@ -151,7 +151,7 @@ molecule skills bundle my-custom-skill --output ./org-templates/my-role/
 ```
 
 **Publishing to the community:** Open a PR against
-[`molecule-ai/skills`](https://github.com/Molecule-AI/skills) with a
+[`molecule-ai/skills`](https://git.moleculesai.app/molecule-ai/skills) with a
 complete skill package. Community skills are reviewed for security and
 correctness before listing.
 

docs/integrations/runtime-native-mcp-status.md

@@ -96,7 +96,7 @@ fork needed in production.
   `resolve_platform_id` for plugin-platform-safe deserialization, and
   `self.adapters[adapter.platform]` keying fix (caught by real-subprocess
   test before merge — see below).
-- **Plugin package**: [Molecule-AI/hermes-platform-molecule-a2a](https://github.com/Molecule-AI/hermes-platform-molecule-a2a)
+- **Plugin package**: [Molecule-AI/hermes-platform-molecule-a2a](https://git.moleculesai.app/molecule-ai/hermes-platform-molecule-a2a)
   v0.1.0 — public, MIT-licensed. 11 unit tests + 8 in-process E2E
   + 4 real-subprocess E2E checkpoints all green.
 - **Workspace template patch**: [Molecule-AI/molecule-ai-workspace-template-hermes#32](https://github.com/Molecule-AI/molecule-ai-workspace-template-hermes/pull/32)
@@ -154,7 +154,7 @@ intermediate shim earns its complexity.
 ## Codex (OpenAI Codex CLI)
 
 **Status:** Template SHIPPED. Repo live at
-[`Molecule-AI/molecule-ai-workspace-template-codex`](https://github.com/Molecule-AI/molecule-ai-workspace-template-codex)
+[`Molecule-AI/molecule-ai-workspace-template-codex`](https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-codex)
 (14 files, 1411 LOC, 12/12 tests). molecule-core registration in
 [PR #2512](https://github.com/Molecule-AI/molecule-core/pull/2512).
 E2E with real A2A traffic remains.

docs/quickstart.md

@@ -17,7 +17,7 @@ This path is aligned to the current repository and current UI. It gets you from
 ## The one-command path
 
 ```bash
-git clone https://github.com/Molecule-AI/molecule-monorepo.git
+git clone https://git.moleculesai.app/molecule-ai/molecule-monorepo.git
 cd molecule-monorepo
 ./scripts/dev-start.sh
 ```
@@ -42,7 +42,7 @@ If you'd rather run each component yourself — useful when you're iterating on
 ### Step 1: Clone the repository
 
 ```bash
-git clone https://github.com/Molecule-AI/molecule-monorepo.git
+git clone https://git.moleculesai.app/molecule-ai/molecule-monorepo.git
 cd molecule-monorepo
 ```
 

docs/workspace-runtime-package.md

@@ -98,14 +98,14 @@ Each of the 8 adapter template repos contains:
 
 | Adapter | Repo |
 |---------|------|
-| claude-code | https://github.com/Molecule-AI/molecule-ai-workspace-template-claude-code |
-| langgraph | https://github.com/Molecule-AI/molecule-ai-workspace-template-langgraph |
-| crewai | https://github.com/Molecule-AI/molecule-ai-workspace-template-crewai |
-| autogen | https://github.com/Molecule-AI/molecule-ai-workspace-template-autogen |
-| deepagents | https://github.com/Molecule-AI/molecule-ai-workspace-template-deepagents |
-| hermes | https://github.com/Molecule-AI/molecule-ai-workspace-template-hermes |
-| gemini-cli | https://github.com/Molecule-AI/molecule-ai-workspace-template-gemini-cli |
-| openclaw | https://github.com/Molecule-AI/molecule-ai-workspace-template-openclaw |
+| claude-code | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-claude-code |
+| langgraph | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-langgraph |
+| crewai | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-crewai |
+| autogen | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-autogen |
+| deepagents | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-deepagents |
+| hermes | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-hermes |
+| gemini-cli | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-gemini-cli |
+| openclaw | https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-openclaw |
 
 ## Adapter discovery (ADAPTER_MODULE)
 
@@ -244,7 +244,7 @@ correctness before pushing a `runtime-v*` tag.
 ## Writing a new adapter
 
 Use the GitHub template repo
-[`Molecule-AI/molecule-ai-workspace-template-starter`](https://github.com/Molecule-AI/molecule-ai-workspace-template-starter)
+[`molecule-ai/molecule-ai-workspace-template-starter`](https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-starter) (note: the starter repo did not survive the 2026-05-06 GitHub-org-suspension migration; recreation tracked at internal#41)
 — it ships with the canonical Dockerfile + adapter.py skeleton + config.yaml
 schema + the `repository_dispatch: [runtime-published]` cascade receiver
 already wired up. No follow-up setup PR required.
@@ -256,7 +256,7 @@ gh repo create Molecule-AI/molecule-ai-workspace-template-<runtime> \
   --public \
   --description "Molecule AI workspace template: <runtime>"
 
-git clone https://github.com/Molecule-AI/molecule-ai-workspace-template-<runtime>
+git clone https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-<runtime>.git
 cd molecule-ai-workspace-template-<runtime>
 ```
 
@@ -286,7 +286,7 @@ After `git push`:
 If the canonical shape changes (e.g. `config.yaml` schema gets a new field,
 the `BaseAdapter` interface adds a method, the reusable CI workflow
 signature changes), update the
-[starter](https://github.com/Molecule-AI/molecule-ai-workspace-template-starter)
+[starter](https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-starter) (recreation pending — see note above)
 **first**. Existing templates can either migrate at their own pace or be
 touched in a coordinated cleanup PR. Either way, future templates pick up
 the new shape from day one.

manifest.json

@@ -41,7 +41,6 @@
     {"name": "medo-smoke", "repo": "Molecule-AI/molecule-ai-org-template-medo-smoke", "ref": "main"},
     {"name": "molecule-worker-gemini", "repo": "Molecule-AI/molecule-ai-org-template-molecule-worker-gemini", "ref": "main"},
     {"name": "reno-stars", "repo": "Molecule-AI/molecule-ai-org-template-reno-stars", "ref": "main"},
-    {"name": "ux-ab-lab", "repo": "Molecule-AI/molecule-ai-org-template-ux-ab-lab", "ref": "main"},
-    {"name": "mock-bigorg", "repo": "Molecule-AI/molecule-ai-org-template-mock-bigorg", "ref": "main"}
+    {"name": "ux-ab-lab", "repo": "Molecule-AI/molecule-ai-org-template-ux-ab-lab", "ref": "main"}
   ]
 }

scripts/README.md

@@ -11,7 +11,7 @@ There are three related scripts; pick the right one:
 |---|---|---|
 | `measure-coordinator-task-bounds.sh` | **Canonical** v1 harness for the RFC #2251 / Issue 4 reproduction. Provisions a PM coordinator + Researcher child via `claude-code-default` + `langgraph` templates, sends a synthesis-heavy A2A kickoff, observes elapsed time + activity trace. | OSS-shape platform — localhost or any `/workspaces`-shaped endpoint. Has tenant/admin-token guards for non-localhost runs. |
 | `measure-coordinator-task-bounds-runner.sh` | Generalised runner for the same measurement contract but with **arbitrary template + secret + model combinations** (Hermes/MiniMax, etc.). Useful for cross-runtime variants without modifying the canonical harness. | Same as above (local or SaaS via `MODE=saas`). |
-| `measure-coordinator-task-bounds.sh` (in [molecule-controlplane](https://github.com/Molecule-AI/molecule-controlplane)) | **Production-shape** variant that bootstraps a real staging tenant via `POST /cp/admin/orgs`, then runs the same measurement against `<slug>.staging.moleculesai.app`. | Staging controlplane only — refuses to run against production. |
+| `measure-coordinator-task-bounds.sh` (in [molecule-controlplane](https://git.moleculesai.app/molecule-ai/molecule-controlplane)) | **Production-shape** variant that bootstraps a real staging tenant via `POST /cp/admin/orgs`, then runs the same measurement against `<slug>.staging.moleculesai.app`. | Staging controlplane only — refuses to run against production. |
 
 See `reference_harness_pair_pattern` (auto-memory) for when to use which
 and the cross-repo design rationale.

scripts/build_runtime_package.py

@@ -278,7 +278,7 @@ include = ["molecule_runtime*"]
 README_TEMPLATE = """\
 # molecule-ai-workspace-runtime
 
-Shared workspace runtime for [Molecule AI](https://github.com/Molecule-AI/molecule-core)
+Shared workspace runtime for [Molecule AI](https://git.moleculesai.app/molecule-ai/molecule-core)
 agent adapters. Installed by every workspace template image
 (`workspace-template-claude-code`, `-langgraph`, `-hermes`, etc.) to provide
 A2A delegation, heartbeat, memory, plugin loading, and skill management.
@@ -396,7 +396,7 @@ If you don't need real-time push, the default poll path works
 universally with no extra setup; both modes converge on the same
 `inbox_pop` ack so messages never duplicate.
 
-See [`docs/workspace-runtime-package.md`](https://github.com/Molecule-AI/molecule-core/blob/main/docs/workspace-runtime-package.md)
+See [`docs/workspace-runtime-package.md`](https://git.moleculesai.app/molecule-ai/molecule-core/src/branch/main/docs/workspace-runtime-package.md)
 for the publish flow and architecture.
 """
 

scripts/clone-manifest.sh

@@ -45,18 +45,11 @@ clone_category() {
             continue
         fi
 
-        # Post-2026-05-06 GitHub-org-suspension: clone from Gitea instead.
-        # manifest.json paths still read "Molecule-AI/..." (the historic
-        # github.com slug); Gitea lowercases the org part to "molecule-ai/".
-        # Lowercase the org segment on the fly so we don't need to rewrite
-        # every manifest entry.
-        repo_gitea="$(echo "$repo" | awk -F/ '{ printf "%s", tolower($1); for (i=2; i<=NF; i++) printf "/%s", $i; print "" }')"
-
-        echo "  cloning $repo_gitea -> $target_dir/$name (ref=$ref)"
+        echo "  cloning $repo -> $target_dir/$name (ref=$ref)"
         if [ "$ref" = "main" ]; then
-            git clone --depth=1 -q "https://git.moleculesai.app/${repo_gitea}.git" "$target_dir/$name"
+            git clone --depth=1 -q "https://github.com/${repo}.git" "$target_dir/$name"
         else
-            git clone --depth=1 -q --branch "$ref" "https://git.moleculesai.app/${repo_gitea}.git" "$target_dir/$name"
+            git clone --depth=1 -q --branch "$ref" "https://github.com/${repo}.git" "$target_dir/$name"
         fi
         CLONED=$((CLONED + 1))
         i=$((i + 1))

scripts/demo-freeze.sh

@@ -10,11 +10,11 @@
 #           → PyPI auto-bumps molecule-ai-workspace-runtime patch version
 #           → repository_dispatch fans out to 8 workspace-template-* repos
 #           → each template repo rebuilds and re-tags
-#             ghcr.io/molecule-ai/workspace-template-<runtime>:latest
+#             153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/workspace-template-<runtime>:latest
 #
 #   PATH 2: any merge to a workspace-template-* repo's main branch
 #           → that repo's publish-image.yml fires
-#           → ghcr.io/molecule-ai/workspace-template-<runtime>:latest
+#           → 153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/workspace-template-<runtime>:latest
 #             gets re-tagged
 #
 #   provisioner.go:296 RuntimeImages[runtime] reads `:latest` at every

scripts/refresh-workspace-images.sh

@@ -51,7 +51,7 @@ log "pulling latest images for: ${RUNTIMES[*]}"
 PULLED=()
 FAILED=()
 for rt in "${RUNTIMES[@]}"; do
-  IMG="ghcr.io/molecule-ai/workspace-template-$rt:latest"
+  IMG="153263036946.dkr.ecr.us-east-2.amazonaws.com/molecule-ai/workspace-template-$rt:latest"
   if docker pull "$IMG" >/dev/null 2>&1; then
     log "  ✓ $rt"
     PULLED+=("$rt")

scripts/rollback-latest.sh

@@ -1,9 +1,10 @@
 #!/bin/bash
-# rollback-latest.sh — moves the :latest tag on ghcr.io/molecule-ai/platform
-# (and the matching tenant image) back to a prior :staging-<sha> digest
-# without rebuilding anything. Prod tenants auto-pull :latest every 5
-# min, so this is the fast path when a canary-verified image turns out
-# to have a runtime regression that canary didn't catch.
+# rollback-latest.sh — moves the :latest tag on the platform image
+# (and the matching tenant image) on AWS ECR back to a prior
+# :staging-<sha> digest without rebuilding anything. Prod tenants
+# auto-pull :latest every 5 min, so this is the fast path when a
+# canary-verified image turns out to have a runtime regression that
+# canary didn't catch.
 #
 # Usage:
 #   scripts/rollback-latest.sh <sha>
@@ -12,12 +13,14 @@
 # Prereqs:
 #   - crane on $PATH (brew install crane OR download from
 #     https://github.com/google/go-containerregistry/releases)
-#   - GHCR token exported as GITHUB_TOKEN with write:packages scope
+#   - aws CLI authenticated for region us-east-2 with ECR pull/push
+#     access to the molecule-ai/platform + platform-tenant repositories.
+#     `aws sts get-caller-identity` should succeed.
 #
 # What it does (per image — platform + tenant):
-#   crane digest ghcr.io/…:<sha>         # verify the target sha exists
-#   crane tag    ghcr.io/…:<sha> latest  # retag remotely, single API call
-#   crane digest ghcr.io/…:latest        # confirm the move
+#   crane digest <ecr>:<sha>         # verify the target sha exists
+#   crane tag    <ecr>:<sha> latest  # retag remotely, single API call
+#   crane digest <ecr>:latest        # confirm the move
 #
 # Exit codes: 0 = both retagged, 1 = tag missing / crane error, 2 = bad args.
 
@@ -30,21 +33,23 @@ if [ "${1:-}" = "" ]; then
 fi
 
 TARGET_SHA="$1"
-PLATFORM=ghcr.io/molecule-ai/platform
-TENANT=ghcr.io/molecule-ai/platform-tenant
+ECR_HOST=153263036946.dkr.ecr.us-east-2.amazonaws.com
+PLATFORM=$ECR_HOST/molecule-ai/platform
+TENANT=$ECR_HOST/molecule-ai/platform-tenant
 
 if ! command -v crane >/dev/null; then
   echo "ERROR: crane not installed. brew install crane" >&2
   exit 1
 fi
-if [ -z "${GITHUB_TOKEN:-}" ]; then
-  echo "ERROR: GITHUB_TOKEN unset. export it with write:packages scope." >&2
+if ! command -v aws >/dev/null; then
+  echo "ERROR: aws CLI not installed. brew install awscli" >&2
   exit 1
 fi
 
-# Log in once. crane stores creds in a config file keyed by registry;
-# re-running is cheap.
-printf '%s\n' "$GITHUB_TOKEN" | crane auth login ghcr.io -u "${GITHUB_ACTOR:-$(whoami)}" --password-stdin >/dev/null
+# Log in once. ECR auth is via short-lived password from `aws ecr
+# get-login-password`. crane stores creds in a config file keyed by
+# registry; re-running is cheap.
+aws ecr get-login-password --region us-east-2 | crane auth login "$ECR_HOST" -u AWS --password-stdin >/dev/null
 
 roll() {
   local image="$1"

tools/check-template-parity.sh

@@ -18,7 +18,7 @@
 #
 # Or inline via curl:
 #
-#     bash <(curl -fsSL https://raw.githubusercontent.com/Molecule-AI/molecule-core/main/tools/check-template-parity.sh) \
+#     bash <(curl -fsSL https://git.moleculesai.app/molecule-ai/molecule-core/raw/branch/main/tools/check-template-parity.sh) \
 #          install.sh start.sh
 #
 # Exit codes:

workspace-server/cmd/server/bind_test.go

@@ -1,89 +0,0 @@
-package main
-
-import "testing"
-
-// TestResolveBindHost pins the precedence: BIND_ADDR explicit > dev-mode
-// fail-open default of 127.0.0.1 > production-shape empty (all interfaces).
-//
-// Mutation-test invariant: removing the IsDevModeFailOpen() branch makes
-// "no_bindaddr_devmode_unset_admin" fail (returns "" instead of "127.0.0.1").
-// Removing the BIND_ADDR branch makes "explicit_bindaddr_*" cases fail.
-func TestResolveBindHost(t *testing.T) {
-	cases := []struct {
-		name       string
-		bindAddr   string
-		adminToken string
-		molEnv     string
-		want       string
-	}{
-		{
-			name:       "no_bindaddr_devmode_unset_admin",
-			bindAddr:   "",
-			adminToken: "",
-			molEnv:     "dev",
-			want:       "127.0.0.1",
-		},
-		{
-			name:       "no_bindaddr_devmode_unset_admin_full_word",
-			bindAddr:   "",
-			adminToken: "",
-			molEnv:     "development",
-			want:       "127.0.0.1",
-		},
-		{
-			name:       "no_bindaddr_admin_set_in_dev_env",
-			bindAddr:   "",
-			adminToken: "secret",
-			molEnv:     "dev",
-			want:       "", // ADMIN_TOKEN flips IsDevModeFailOpen to false → all interfaces
-		},
-		{
-			name:       "no_bindaddr_production_env",
-			bindAddr:   "",
-			adminToken: "",
-			molEnv:     "production",
-			want:       "", // production is not a dev value → all interfaces
-		},
-		{
-			name:       "no_bindaddr_unset_env",
-			bindAddr:   "",
-			adminToken: "",
-			molEnv:     "",
-			want:       "", // unset MOLECULE_ENV → not dev → all interfaces
-		},
-		{
-			name:       "explicit_bindaddr_loopback_overrides_devmode",
-			bindAddr:   "127.0.0.1",
-			adminToken: "",
-			molEnv:     "dev",
-			want:       "127.0.0.1",
-		},
-		{
-			name:       "explicit_bindaddr_wildcard_overrides_devmode_default",
-			bindAddr:   "0.0.0.0",
-			adminToken: "",
-			molEnv:     "dev",
-			want:       "0.0.0.0",
-		},
-		{
-			name:       "explicit_bindaddr_in_production",
-			bindAddr:   "10.0.5.7",
-			adminToken: "secret",
-			molEnv:     "production",
-			want:       "10.0.5.7",
-		},
-	}
-
-	for _, tc := range cases {
-		t.Run(tc.name, func(t *testing.T) {
-			t.Setenv("BIND_ADDR", tc.bindAddr)
-			t.Setenv("ADMIN_TOKEN", tc.adminToken)
-			t.Setenv("MOLECULE_ENV", tc.molEnv)
-			got := resolveBindHost()
-			if got != tc.want {
-				t.Errorf("resolveBindHost() = %q, want %q (BIND_ADDR=%q ADMIN_TOKEN=%q MOLECULE_ENV=%q)",
-					got, tc.want, tc.bindAddr, tc.adminToken, tc.molEnv)
-			}
-		})
-	}
-}

workspace-server/cmd/server/main.go

@@ -19,7 +19,6 @@ import (
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/handlers"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/imagewatch"
 	memwiring "github.com/Molecule-AI/molecule-monorepo/platform/internal/memory/wiring"
-	"github.com/Molecule-AI/molecule-monorepo/platform/internal/middleware"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/pendinguploads"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/provisioner"
 	"github.com/Molecule-AI/molecule-monorepo/platform/internal/registry"
@@ -249,6 +248,19 @@ func main() {
 		})
 	}
 
+	// CP-mode orphan sweeper — SaaS counterpart to the Docker sweeper
+	// above. Re-issues cpProv.Stop for any workspace at status='removed'
+	// with a non-NULL instance_id, healing the deprovision split-write
+	// race documented in #2989: tenant marks status='removed' BEFORE
+	// calling CP DELETE, so a transient CP failure leaves the EC2
+	// running with no retry path. cpProv.Stop is idempotent against
+	// already-terminated instances; on success we clear instance_id.
+	if cpProv != nil {
+		go supervised.RunWithRecover(ctx, "cp-orphan-sweeper", func(c context.Context) {
+			registry.StartCPOrphanSweeper(c, cpProv)
+		})
+	}
+
 	// Pending-uploads GC sweep — deletes acked rows past their retention
 	// window plus unacked rows past expires_at. Without this the
 	// pending_uploads table grows unbounded; even with the 24h hard TTL,
@@ -320,23 +332,15 @@ func main() {
 	// Router
 	r := router.Setup(hub, broadcaster, prov, platformURL, configsDir, wh, channelMgr, memBundle)
 
-	// HTTP server with graceful shutdown.
-	//
-	// Bind host: in dev-mode (no ADMIN_TOKEN, MOLECULE_ENV=dev|development)
-	// the AdminAuth chain fails open by design; pairing that with a wildcard
-	// bind would expose unauth /workspaces to any same-LAN peer. Default to
-	// loopback when fail-open is active. Operators who need LAN exposure set
-	// BIND_ADDR=0.0.0.0 explicitly. Production (ADMIN_TOKEN set) is unchanged.
-	// See molecule-core#7.
-	bindHost := resolveBindHost()
+	// HTTP server with graceful shutdown
 	srv := &http.Server{
-		Addr:    fmt.Sprintf("%s:%s", bindHost, port),
+		Addr:    fmt.Sprintf(":%s", port),
 		Handler: r,
 	}
 
 	// Start server in goroutine
 	go func() {
-		log.Printf("Platform starting on %s:%s (dev-mode-fail-open=%v)", bindHost, port, middleware.IsDevModeFailOpen())
+		log.Printf("Platform starting on :%s", port)
 		if err := srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
 			log.Fatalf("Server failed: %v", err)
 		}
@@ -371,29 +375,6 @@ func envOr(key, fallback string) string {
 	return fallback
 }
 
-// resolveBindHost picks the listener interface for the HTTP server.
-//
-// Precedence:
-//  1. BIND_ADDR — explicit operator override (any value, including "0.0.0.0").
-//  2. dev-mode fail-open active → "127.0.0.1" (loopback only).
-//  3. otherwise → "" (Go binds every interface; existing prod/self-host shape).
-//
-// Coupling the loopback default to middleware.IsDevModeFailOpen() means the
-// two safety levers — bind narrowness and auth strength — move together. A
-// production deploy (ADMIN_TOKEN set) keeps binding to all interfaces because
-// the auth chain is doing its job; a dev Mac (no ADMIN_TOKEN, MOLECULE_ENV=dev)
-// is reachable only via loopback because the auth chain is fail-open. See
-// molecule-core#7 for the original LAN exposure finding.
-func resolveBindHost() string {
-	if v := os.Getenv("BIND_ADDR"); v != "" {
-		return v
-	}
-	if middleware.IsDevModeFailOpen() {
-		return "127.0.0.1"
-	}
-	return ""
-}
-
 func findConfigsDir() string {
 	candidates := []string{
 		"workspace-configs-templates",

workspace-server/internal/handlers/a2a_proxy.go

@@ -413,23 +413,6 @@ func (h *WorkspaceHandler) proxyA2ARequest(ctx context.Context, workspaceID stri
 		return http.StatusOK, respBody, nil
 	}
 
-	// Mock-runtime short-circuit. Workspaces with runtime='mock' have
-	// no container, no EC2, no URL — every reply is synthesised here
-	// from a small canned-variant pool. Built for the "200-workspace
-	// mock org" demo: a CEO/VPs/Managers/ICs hierarchy that renders
-	// at scale on the canvas without burning real LLM credits or
-	// provisioning 200 EC2 instances. See mock_runtime.go for the
-	// full rationale + reply shape contract.
-	//
-	// Position: AFTER poll-mode (mock isn't a delivery mode, it's a
-	// runtime; treating poll-set-on-mock as poll matches operator
-	// intent if anyone ever does that), BEFORE resolveAgentURL (mock
-	// has no URL — going through resolveAgentURL would 404 on the
-	// SELECT url since the row is provisioned as NULL).
-	if status, respBody, handled := h.handleMockA2A(ctx, workspaceID, callerID, body, a2aMethod, logActivity); handled {
-		return status, respBody, nil
-	}
-
 	agentURL, proxyErr := h.resolveAgentURL(ctx, workspaceID)
 	if proxyErr != nil {
 		return 0, nil, proxyErr

workspace-server/internal/handlers/eic_tunnel_pool.go

@@ -0,0 +1,437 @@
+package handlers
+
+// eic_tunnel_pool.go — refcounted pool for EIC SSH tunnels keyed on
+// instanceID. Reuses one tunnel across N file ops, amortising the
+// ssh-keygen + SendSSHPublicKey + open-tunnel + waitForPort cost
+// (~3-5s) over multiple cats/finds (~50-200ms each).
+//
+// Origin: core#11 — canvas detail-panel config + filesystem load
+// took ~20s. ConfigTab fans out 4 GETs serially; the slowest is
+// /files/config.yaml which dispatches to readFileViaEIC. Without a
+// pool, every readFileViaEIC + listFilesViaEIC + writeFileViaEIC +
+// deleteFileViaEIC pays the full setup cost even when fired
+// back-to-back on the same workspace EC2.
+//
+// The pool keeps one eicSSHSession alive per instanceID for up to
+// poolTTL. SendSSHPublicKey grants a 60s key validity, so poolTTL
+// must stay strictly below that to avoid serving requests on a
+// just-expired key. We default to 50s with a 10s safety margin.
+//
+// Concurrency model:
+//
+//   - Single mutex guards the entries map.
+//   - Slow path (tunnel setup) runs OUTSIDE the lock, gated by an
+//     "intent" placeholder so concurrent acquires for the same
+//     instanceID don't both build a tunnel — the loser drops its
+//     setup and uses the winner's.
+//   - Refcount on each entry; eviction blocked while refcount > 0.
+//   - Janitor goroutine sweeps every poolJanitorInterval, drops
+//     entries where refcount == 0 && expiresAt < now.
+//
+// Test injection:
+//
+//   - poolSetupTunnel is a package-level var so tests can swap the
+//     slow path for a counting stub. Production wires it to
+//     realWithEICTunnel-style setup.
+//   - withEICTunnel (the public, single-shot API) is also a var
+//     (already, see template_files_eic.go). It's rebound here to
+//     pooledWithEICTunnel which routes through globalEICTunnelPool.
+//   - Tests that need single-shot behaviour can set poolTTL = 0,
+//     which makes pooledWithEICTunnel fall through to the underlying
+//     setup directly (no pool entry kept).
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+)
+
+// poolTTL is the maximum age of a pooled tunnel. Must be strictly
+// less than the SendSSHPublicKey grant window (60s) so we never
+// serve a request through a key that's about to expire mid-op.
+//
+// Configurable via init-time wiring (see initEICTunnelPool); not a
+// const so tests can pin TTL=0 (disable pooling) or TTL=50ms (drive
+// eviction tests).
+var poolTTL = 50 * time.Second
+
+// poolJanitorInterval is how often the janitor goroutine sweeps for
+// expired idle entries. Tighter than poolTTL so eviction is timely;
+// loose enough that the goroutine doesn't burn CPU.
+var poolJanitorInterval = 10 * time.Second
+
+// poolMaxEntries caps simultaneous instanceIDs the pool tracks.
+// Beyond this, new acquires evict the LRU entry. Defends against a
+// pathological caller (e.g. a sweep over hundreds of workspace
+// EC2s) from leaking unbounded tunnel processes. 32 is a generous
+// ceiling for the canvas use case (one human navigates ≤ ~5
+// workspaces at a time).
+var poolMaxEntries = 32
+
+// poolSetupTunnel is the slow-path tunnel constructor. Wrapped in a
+// var so tests can inject a counter stub. Returns a session and a
+// cleanup function (closes the open-tunnel subprocess + scrubs the
+// ephemeral keydir). nil session + non-nil err means setup failed
+// and there is nothing to clean up.
+//
+// Production wiring lives in eic_tunnel_pool_setup.go (a thin shim
+// over the existing realWithEICTunnel logic).
+var poolSetupTunnel = func(ctx context.Context, instanceID string) (
+	sess eicSSHSession, cleanup func(), err error) {
+	return setupRealEICTunnel(ctx, instanceID)
+}
+
+// pooledTunnel is one entry in the pool. session is shared by N
+// concurrent fn calls; cleanup runs once when refcount returns to
+// zero AND the entry is past expiresAt or evicted.
+//
+// lastUsed tracks the most recent acquire time for LRU bookkeeping
+// (overflow eviction). expiresAt is set at construction and not
+// extended on use — a tunnel cannot live past poolTTL even if it's
+// hot, because the underlying SendSSHPublicKey grant expires.
+type pooledTunnel struct {
+	session   eicSSHSession
+	cleanup   func()
+	expiresAt time.Time
+	lastUsed  time.Time
+	refcount  int
+	poisoned  bool // true if a fn returned a tunnel-fatal error; do not reuse
+}
+
+// eicTunnelPool is the package-level pool. Single instance lives
+// in globalEICTunnelPool; constructor runs lazily on first acquire.
+type eicTunnelPool struct {
+	mu      sync.Mutex
+	entries map[string]*pooledTunnel
+	// pendingSetups guards concurrent setup for the same instanceID.
+	// First acquirer takes the slot; later ones wait on the channel.
+	pendingSetups map[string]chan struct{}
+	stopJanitor   chan struct{}
+}
+
+var (
+	globalEICTunnelPool     *eicTunnelPool
+	globalEICTunnelPoolOnce sync.Once
+)
+
+// getEICTunnelPool returns the singleton pool, lazy-initialising on
+// first call. Idempotent.
+func getEICTunnelPool() *eicTunnelPool {
+	globalEICTunnelPoolOnce.Do(func() {
+		globalEICTunnelPool = newEICTunnelPool()
+		go globalEICTunnelPool.janitor()
+	})
+	return globalEICTunnelPool
+}
+
+// newEICTunnelPool constructs an empty pool. Exported so tests can
+// build isolated pools without sharing the singleton.
+func newEICTunnelPool() *eicTunnelPool {
+	return &eicTunnelPool{
+		entries:       map[string]*pooledTunnel{},
+		pendingSetups: map[string]chan struct{}{},
+		stopJanitor:   make(chan struct{}),
+	}
+}
+
+// acquire returns a usable session for instanceID. If a healthy entry
+// exists, refcount++ and return it. If a setup is in flight for the
+// same instanceID, wait for it. Otherwise build one (slow path).
+//
+// done() must be called by the caller when the op finishes. It
+// decrements refcount and triggers cleanup if the entry is past
+// TTL or poisoned and refcount==0.
+//
+// Errors from the slow path propagate; pool state is not modified
+// for failed setups (no poisoned entry created — that's only for
+// fn-returned errors on a previously-good session).
+func (p *eicTunnelPool) acquire(ctx context.Context, instanceID string) (
+	sess eicSSHSession, done func(poisoned bool), err error) {
+
+	if poolTTL <= 0 {
+		// Pool disabled (TTL=0 mode for tests / opt-out). Fall
+		// through to a direct setup with caller-driven cleanup.
+		s, cleanup, err := poolSetupTunnel(ctx, instanceID)
+		if err != nil {
+			return eicSSHSession{}, nil, err
+		}
+		return s, func(_ bool) { cleanup() }, nil
+	}
+
+	for {
+		p.mu.Lock()
+		if pt, ok := p.entries[instanceID]; ok && !pt.poisoned && pt.expiresAt.After(time.Now()) {
+			pt.refcount++
+			pt.lastUsed = time.Now()
+			p.mu.Unlock()
+			return pt.session, p.releaser(instanceID, pt), nil
+		}
+		// Either no entry, expired entry, or poisoned entry. If a
+		// setup is already in flight, wait and retry.
+		if pending, ok := p.pendingSetups[instanceID]; ok {
+			p.mu.Unlock()
+			select {
+			case <-pending:
+				continue // re-check the entries map
+			case <-ctx.Done():
+				return eicSSHSession{}, nil, ctx.Err()
+			}
+		}
+		// Drop expired/poisoned entry now (we'll cleanup outside
+		// the lock — the entry is unreferenced or we'd not be here).
+		var oldCleanup func()
+		if pt, ok := p.entries[instanceID]; ok {
+			if pt.refcount == 0 {
+				oldCleanup = pt.cleanup
+				delete(p.entries, instanceID)
+			}
+		}
+		// Reserve the setup slot.
+		signal := make(chan struct{})
+		p.pendingSetups[instanceID] = signal
+		p.mu.Unlock()
+
+		if oldCleanup != nil {
+			go oldCleanup()
+		}
+
+		// Slow path: build a new tunnel. Anything that goes wrong
+		// here cleans up the pendingSetups slot and propagates to
+		// the caller without leaving the pool in a state where the
+		// next acquire blocks waiting on a signal that never fires.
+		newSess, cleanup, setupErr := poolSetupTunnel(ctx, instanceID)
+
+		p.mu.Lock()
+		delete(p.pendingSetups, instanceID)
+		close(signal)
+
+		if setupErr != nil {
+			p.mu.Unlock()
+			return eicSSHSession{}, nil, fmt.Errorf("eic tunnel setup: %w", setupErr)
+		}
+
+		// Enforce LRU bound BEFORE inserting so we don't briefly
+		// exceed the cap even by one entry.
+		p.evictLRUIfFullLocked(instanceID)
+
+		pt := &pooledTunnel{
+			session:   newSess,
+			cleanup:   cleanup,
+			expiresAt: time.Now().Add(poolTTL),
+			lastUsed:  time.Now(),
+			refcount:  1,
+		}
+		p.entries[instanceID] = pt
+		p.mu.Unlock()
+		return pt.session, p.releaser(instanceID, pt), nil
+	}
+}
+
+// releaser returns a closure that decrements refcount and triggers
+// cleanup if (a) the entry is past TTL or (b) the caller signalled
+// poison. Idempotent against double-release (decrements once via the
+// captured pt; pool entry may have been replaced by then).
+func (p *eicTunnelPool) releaser(instanceID string, pt *pooledTunnel) func(poisoned bool) {
+	released := false
+	return func(poisoned bool) {
+		p.mu.Lock()
+		defer p.mu.Unlock()
+		if released {
+			return
+		}
+		released = true
+		pt.refcount--
+		if poisoned {
+			pt.poisoned = true
+		}
+		// Evict immediately if poisoned-and-idle OR expired-and-idle.
+		// Hot entries (refcount > 0) defer eviction to the last release.
+		if pt.refcount == 0 && (pt.poisoned || pt.expiresAt.Before(time.Now())) {
+			// If the entry in the map is still us, remove it.
+			if cur, ok := p.entries[instanceID]; ok && cur == pt {
+				delete(p.entries, instanceID)
+			}
+			go pt.cleanup()
+		}
+	}
+}
+
+// evictLRUIfFullLocked drops the least-recently-used IDLE entry
+// when the pool is at capacity. Caller must hold p.mu. The new
+// instanceID about to be inserted is excluded so we don't evict
+// ourselves. If no idle entries exist, no eviction happens — the
+// new entry will push us above the soft cap until something releases.
+func (p *eicTunnelPool) evictLRUIfFullLocked(skipInstance string) {
+	if len(p.entries) < poolMaxEntries {
+		return
+	}
+	var oldestKey string
+	var oldest *pooledTunnel
+	for k, pt := range p.entries {
+		if k == skipInstance {
+			continue
+		}
+		if pt.refcount > 0 {
+			continue
+		}
+		if oldest == nil || pt.lastUsed.Before(oldest.lastUsed) {
+			oldestKey = k
+			oldest = pt
+		}
+	}
+	if oldest == nil {
+		return // every entry is in use; no eviction possible
+	}
+	delete(p.entries, oldestKey)
+	go oldest.cleanup()
+}
+
+// janitor periodically scans for entries that are idle AND expired,
+// closing their tunnels. Runs forever (per pool lifetime); cancelled
+// by close(p.stopJanitor) for tests that build short-lived pools.
+func (p *eicTunnelPool) janitor() {
+	t := time.NewTicker(poolJanitorInterval)
+	defer t.Stop()
+	for {
+		select {
+		case <-t.C:
+			p.sweep()
+		case <-p.stopJanitor:
+			return
+		}
+	}
+}
+
+// sweep is one janitor pass. Drops idle expired entries.
+func (p *eicTunnelPool) sweep() {
+	p.mu.Lock()
+	now := time.Now()
+	var toClose []func()
+	for k, pt := range p.entries {
+		if pt.refcount == 0 && pt.expiresAt.Before(now) {
+			toClose = append(toClose, pt.cleanup)
+			delete(p.entries, k)
+		}
+	}
+	p.mu.Unlock()
+	for _, c := range toClose {
+		go c()
+	}
+}
+
+// stop terminates the janitor and closes all idle entries. Hot
+// (refcount > 0) entries are NOT force-closed — callers running
+// against them would see a use-after-free. In practice stop is only
+// called by tests that have already drained their callers.
+func (p *eicTunnelPool) stop() {
+	close(p.stopJanitor)
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	for k, pt := range p.entries {
+		if pt.refcount == 0 {
+			go pt.cleanup()
+			delete(p.entries, k)
+		}
+	}
+}
+
+// pooledWithEICTunnel is the pool-backed replacement for
+// realWithEICTunnel. The signature matches `var withEICTunnel`
+// exactly so the rebind (in initEICTunnelPool) is a drop-in.
+//
+// Errors from `fn` itself are forwarded to the caller AND mark the
+// pool entry as poisoned, so the next acquire builds a fresh
+// tunnel. This catches the case where the workspace EC2 was
+// restarted out-of-band (tunnel still appears alive locally but
+// every cat/find errors out).
+func pooledWithEICTunnel(ctx context.Context, instanceID string,
+	fn func(s eicSSHSession) error) error {
+	pool := getEICTunnelPool()
+	sess, done, err := pool.acquire(ctx, instanceID)
+	if err != nil {
+		return err
+	}
+	// poisoned defaults to true so a panic from fn poisons the
+	// entry on the way through the deferred release. Without the
+	// defer, a panicking fn would leak refcount=1 forever and
+	// permanently block eviction of this entry. The fn-error path
+	// resets poisoned to its real classification before return.
+	poisoned := true
+	defer func() { done(poisoned) }()
+	fnErr := fn(sess)
+	poisoned = fnErrIndicatesTunnelFault(fnErr)
+	return fnErr
+}
+
+// fnErrIndicatesTunnelFault returns true for fn errors whose nature
+// suggests the underlying tunnel is no longer reusable (auth gone,
+// network gone, ssh process dead). Returning true poisons the pool
+// entry so the next acquire builds fresh.
+//
+// Conservative: only marks tunnel-faulty for clearly tunnel-level
+// failures (connection refused, broken pipe, ssh exit-status from
+// fatal-channel signals). A `cat` returning os.ErrNotExist on a
+// missing file is NOT a tunnel fault — that's the file path being
+// wrong, the tunnel is fine.
+func fnErrIndicatesTunnelFault(err error) bool {
+	if err == nil {
+		return false
+	}
+	msg := err.Error()
+	// stderr substrings produced by ssh when the tunnel is broken.
+	for _, marker := range []string{
+		"connection refused",
+		"connection closed",
+		"broken pipe",
+		"Connection reset by peer",
+		"kex_exchange_identification",
+		"port forwarding failed",
+		"Permission denied",
+		"Authentication failed",
+	} {
+		if containsCaseInsensitive(msg, marker) {
+			return true
+		}
+	}
+	return false
+}
+
+// containsCaseInsensitive avoids importing strings just for this
+// (the file already needs ssh stderr matching elsewhere — this
+// keeps the helper local to avoid a cross-file dependency).
+func containsCaseInsensitive(s, substr string) bool {
+	if len(substr) > len(s) {
+		return false
+	}
+	// Manual lowercase compare loop; ssh error markers are ASCII so
+	// no need for unicode-aware folding.
+	low := func(b byte) byte {
+		if b >= 'A' && b <= 'Z' {
+			return b + 32
+		}
+		return b
+	}
+	for i := 0; i+len(substr) <= len(s); i++ {
+		match := true
+		for j := 0; j < len(substr); j++ {
+			if low(s[i+j]) != low(substr[j]) {
+				match = false
+				break
+			}
+		}
+		if match {
+			return true
+		}
+	}
+	return false
+}
+
+// initEICTunnelPool rebinds the package-level withEICTunnel var to
+// the pooled implementation. Called once at package init via the
+// init() in eic_tunnel_pool_setup.go (split file so the rebind
+// itself is testable without dragging in the production setup
+// shim's exec/aws dependencies).
+func initEICTunnelPool() {
+	withEICTunnel = pooledWithEICTunnel
+}

workspace-server/internal/handlers/eic_tunnel_pool_setup.go

@@ -0,0 +1,136 @@
+package handlers
+
+// eic_tunnel_pool_setup.go — production setup shim.
+//
+// setupRealEICTunnel decomposes the existing realWithEICTunnel into
+// its slow half (build the tunnel) and its caller half (run fn). The
+// pool calls the slow half once and shares the resulting session
+// across N callers, holding cleanup until the last release.
+//
+// Why decompose instead of refactoring realWithEICTunnel: the
+// existing function and its test stub-vars (withEICTunnel,
+// sendSSHPublicKey, openTunnelCmd) are load-bearing for the
+// dispatch tests. Extracting a sibling setup function preserves the
+// existing single-shot path verbatim — the pool wraps it by calling
+// realWithEICTunnel through a thin adapter, leaving the tested
+// surface unchanged.
+//
+// The pool's acquire() invokes poolSetupTunnel, which is a `var`
+// pointing to setupRealEICTunnel for production and a counting stub
+// for tests.
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+	"time"
+)
+
+// setupRealEICTunnel is the slow path that the pool consumes when
+// no warm entry exists. Mirrors realWithEICTunnel's setup half but
+// returns the session + cleanup instead of running fn inline.
+//
+// The cleanup func owns the tunnel subprocess, ephemeral key dir,
+// and a one-time wait. Idempotent — calling it twice is safe; the
+// pool guarantees one call per session, but defence-in-depth helps
+// when tests run pools in parallel and racy sweeps re-trigger.
+func setupRealEICTunnel(ctx context.Context, instanceID string) (
+	eicSSHSession, func(), error) {
+
+	if instanceID == "" {
+		return eicSSHSession{}, nil,
+			fmt.Errorf("workspace has no instance_id — not a SaaS EC2 workspace")
+	}
+	osUser := os.Getenv("WORKSPACE_EC2_OS_USER")
+	if osUser == "" {
+		osUser = "ubuntu"
+	}
+	region := os.Getenv("AWS_REGION")
+	if region == "" {
+		region = "us-east-2"
+	}
+
+	keyDir, err := os.MkdirTemp("", "molecule-eic-pool-*")
+	if err != nil {
+		return eicSSHSession{}, nil, fmt.Errorf("keydir mkdir: %w", err)
+	}
+	keyPath := keyDir + "/id"
+	if out, kerr := exec.CommandContext(ctx, "ssh-keygen",
+		"-t", "ed25519", "-f", keyPath, "-N", "", "-q",
+		"-C", "molecule-eic-pool",
+	).CombinedOutput(); kerr != nil {
+		_ = os.RemoveAll(keyDir)
+		return eicSSHSession{}, nil,
+			fmt.Errorf("ssh-keygen: %w (%s)", kerr, strings.TrimSpace(string(out)))
+	}
+	pubKey, err := os.ReadFile(keyPath + ".pub")
+	if err != nil {
+		_ = os.RemoveAll(keyDir)
+		return eicSSHSession{}, nil, fmt.Errorf("read pubkey: %w", err)
+	}
+
+	if err := sendSSHPublicKey(ctx, region, instanceID, osUser,
+		strings.TrimSpace(string(pubKey))); err != nil {
+		_ = os.RemoveAll(keyDir)
+		return eicSSHSession{}, nil, fmt.Errorf("send-ssh-public-key: %w", err)
+	}
+
+	localPort, err := pickFreePort()
+	if err != nil {
+		_ = os.RemoveAll(keyDir)
+		return eicSSHSession{}, nil, fmt.Errorf("pick free port: %w", err)
+	}
+
+	tunnel := openTunnelCmd(eicSSHOptions{
+		InstanceID:     instanceID,
+		OSUser:         osUser,
+		Region:         region,
+		LocalPort:      localPort,
+		PrivateKeyPath: keyPath,
+	})
+	tunnel.Env = os.Environ()
+	if err := tunnel.Start(); err != nil {
+		_ = os.RemoveAll(keyDir)
+		return eicSSHSession{}, nil, fmt.Errorf("open-tunnel start: %w", err)
+	}
+
+	if err := waitForPort(ctx, "127.0.0.1", localPort, 10*time.Second); err != nil {
+		if tunnel.Process != nil {
+			_ = tunnel.Process.Kill()
+		}
+		_ = tunnel.Wait()
+		_ = os.RemoveAll(keyDir)
+		return eicSSHSession{}, nil, fmt.Errorf("tunnel never listened: %w", err)
+	}
+
+	cleanedUp := false
+	cleanup := func() {
+		if cleanedUp {
+			return
+		}
+		cleanedUp = true
+		if tunnel.Process != nil {
+			_ = tunnel.Process.Kill()
+		}
+		_ = tunnel.Wait()
+		_ = os.RemoveAll(keyDir)
+	}
+
+	return eicSSHSession{
+		keyPath:    keyPath,
+		localPort:  localPort,
+		osUser:     osUser,
+		instanceID: instanceID,
+	}, cleanup, nil
+}
+
+// init wires the pool into the package-level withEICTunnel var so
+// every read/write/list/delete EIC op uses pooled tunnels by default.
+// Test files that need single-shot behaviour can swap withEICTunnel
+// back via the existing stubWithEICTunnel pattern, OR set poolTTL=0
+// to disable pooling without rebinding the var.
+func init() {
+	initEICTunnelPool()
+}

workspace-server/internal/handlers/eic_tunnel_pool_test.go

@@ -0,0 +1,467 @@
+package handlers
+
+// eic_tunnel_pool_test.go — tests for the refcounted EIC tunnel pool
+// added in core#11. Stubs poolSetupTunnel with a counter so the
+// tests don't fork ssh-keygen / aws subprocesses.
+//
+// Per memory feedback_assert_exact_not_substring: each test pins
+// exact expected counts (not "at least N") so a regression that
+// silently double-sets-up surfaces here.
+
+import (
+	"context"
+	"errors"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+// withPoolSetupStub swaps poolSetupTunnel for a counting fake that
+// returns a sentinel session and a cleanup func that records its
+// invocation. Restores on test cleanup.
+//
+// setupSignal blocks each setup until released — for concurrent-
+// acquire tests where we want to gate setup completion.
+func withPoolSetupStub(t *testing.T) (
+	setupCount *int64, cleanupCount *int64, restore func(), unblock func()) {
+	t.Helper()
+	prev := poolSetupTunnel
+	prevTTL := poolTTL
+	prevJanitor := poolJanitorInterval
+
+	var sc, cc int64
+	setupCount, cleanupCount = &sc, &cc
+
+	gate := make(chan struct{}, 1)
+	gate <- struct{}{} // allow the first setup through immediately
+	unblock = func() { gate <- struct{}{} }
+
+	poolSetupTunnel = func(ctx context.Context, instanceID string) (
+		eicSSHSession, func(), error) {
+		select {
+		case <-gate:
+		case <-ctx.Done():
+			return eicSSHSession{}, nil, ctx.Err()
+		}
+		atomic.AddInt64(&sc, 1)
+		sess := eicSSHSession{
+			instanceID: instanceID,
+			osUser:     "ubuntu",
+			localPort:  10000 + int(atomic.LoadInt64(&sc)),
+			keyPath:    "/tmp/molecule-eic-test-" + instanceID,
+		}
+		cleanup := func() { atomic.AddInt64(&cc, 1) }
+		return sess, cleanup, nil
+	}
+
+	restore = func() {
+		poolSetupTunnel = prev
+		poolTTL = prevTTL
+		poolJanitorInterval = prevJanitor
+	}
+	t.Cleanup(restore)
+	return
+}
+
+// freshPool returns an isolated pool (NOT the global) so tests run
+// independently. Stops the janitor on cleanup.
+func freshPool(t *testing.T) *eicTunnelPool {
+	t.Helper()
+	p := newEICTunnelPool()
+	t.Cleanup(p.stop)
+	return p
+}
+
+// TestEICTunnelPool_FourOpsAmortise pins the core invariant: four
+// sequential acquire/release cycles on the same instanceID share
+// ONE underlying tunnel setup. Mutation: delete the cache hit branch
+// in acquire() → setupCount goes 1 → 4 → test fails.
+func TestEICTunnelPool_FourOpsAmortise(t *testing.T) {
+	setupCount, cleanupCount, _, _ := withPoolSetupStub(t)
+	// Refill gate after each setup so concurrent stubs aren't blocked
+	// (we want every test to be able to set up if it needs to).
+	t.Cleanup(func() { /* no-op; defer is enough */ })
+	poolTTL = 50 * time.Second
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	for i := 0; i < 4; i++ {
+		sess, done, err := pool.acquire(ctx, "i-test-1")
+		if err != nil {
+			t.Fatalf("op %d: acquire: %v", i, err)
+		}
+		if sess.instanceID != "i-test-1" {
+			t.Fatalf("op %d: session has wrong instanceID: %q", i, sess.instanceID)
+		}
+		done(false)
+	}
+
+	if got := atomic.LoadInt64(setupCount); got != 1 {
+		t.Errorf("expected exactly 1 tunnel setup across 4 ops, got %d", got)
+	}
+	if got := atomic.LoadInt64(cleanupCount); got != 0 {
+		t.Errorf("expected 0 cleanups while entry is hot (TTL=50s), got %d", got)
+	}
+}
+
+// TestEICTunnelPool_DifferentInstancesDoNotShare pins that two
+// different instanceIDs each get their own tunnel — the pool is
+// keyed on instanceID, not a single global slot.
+func TestEICTunnelPool_DifferentInstancesDoNotShare(t *testing.T) {
+	setupCount, _, _, unblock := withPoolSetupStub(t)
+	poolTTL = 50 * time.Second
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	// First instance setup uses the initial gate slot.
+	_, doneA, err := pool.acquire(ctx, "i-a")
+	if err != nil {
+		t.Fatalf("acquire A: %v", err)
+	}
+	doneA(false)
+
+	// Second instance needs a new slot through the gate.
+	unblock()
+	_, doneB, err := pool.acquire(ctx, "i-b")
+	if err != nil {
+		t.Fatalf("acquire B: %v", err)
+	}
+	doneB(false)
+
+	if got := atomic.LoadInt64(setupCount); got != 2 {
+		t.Errorf("expected 2 setups (one per instance), got %d", got)
+	}
+}
+
+// TestEICTunnelPool_TTLEviction: a short TTL forces the second op
+// to build a fresh tunnel after the first expires.
+func TestEICTunnelPool_TTLEviction(t *testing.T) {
+	setupCount, cleanupCount, _, unblock := withPoolSetupStub(t)
+	poolTTL = 50 * time.Millisecond
+	poolJanitorInterval = 1 * time.Second // keep janitor away
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	_, done, err := pool.acquire(ctx, "i-ttl")
+	if err != nil {
+		t.Fatalf("acquire 1: %v", err)
+	}
+	done(false)
+
+	time.Sleep(80 * time.Millisecond) // past TTL
+
+	unblock() // allow next setup
+	_, done, err = pool.acquire(ctx, "i-ttl")
+	if err != nil {
+		t.Fatalf("acquire 2: %v", err)
+	}
+	done(false)
+
+	if got := atomic.LoadInt64(setupCount); got != 2 {
+		t.Errorf("expected 2 setups (TTL eviction between), got %d", got)
+	}
+	// First entry should have been cleaned up when the second
+	// acquire evicted it on the slow path. Cleanup runs in a
+	// goroutine; poll briefly for it to land.
+	deadline := time.Now().Add(500 * time.Millisecond)
+	for atomic.LoadInt64(cleanupCount) < 1 && time.Now().Before(deadline) {
+		time.Sleep(5 * time.Millisecond)
+	}
+	if got := atomic.LoadInt64(cleanupCount); got < 1 {
+		t.Errorf("expected ≥1 cleanup (first entry evicted), got %d", got)
+	}
+}
+
+// TestEICTunnelPool_FailureInvalidates pins the poison-on-fault
+// behavior — fn returning a tunnel-fatal error marks the entry
+// unusable so the next acquire builds fresh.
+func TestEICTunnelPool_FailureInvalidates(t *testing.T) {
+	setupCount, _, _, unblock := withPoolSetupStub(t)
+	poolTTL = 50 * time.Second
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	_, done, err := pool.acquire(ctx, "i-fault")
+	if err != nil {
+		t.Fatalf("acquire 1: %v", err)
+	}
+	done(true) // signal poison
+
+	unblock() // let the next setup through
+	_, done, err = pool.acquire(ctx, "i-fault")
+	if err != nil {
+		t.Fatalf("acquire 2: %v", err)
+	}
+	done(false)
+
+	if got := atomic.LoadInt64(setupCount); got != 2 {
+		t.Errorf("expected 2 setups (poison forced rebuild), got %d", got)
+	}
+}
+
+// TestEICTunnelPool_ConcurrentAcquireSingleSetup pins that N
+// concurrent acquires for the same instanceID before any release
+// only trigger ONE tunnel setup — the rest wait via pendingSetups.
+//
+// Without this guard each concurrent acquire would spawn its own
+// tunnel and the loser-cleanup would still leak refcount. Mutation:
+// delete the pendingSetups gate → setupCount goes 1 → N → fails.
+func TestEICTunnelPool_ConcurrentAcquireSingleSetup(t *testing.T) {
+	setupCount, _, _, _ := withPoolSetupStub(t)
+	// Pause setup so all goroutines pile into the pending slot.
+	prev := poolSetupTunnel
+	gate := make(chan struct{})
+	poolSetupTunnel = func(ctx context.Context, instanceID string) (
+		eicSSHSession, func(), error) {
+		<-gate
+		atomic.AddInt64(setupCount, 1)
+		return eicSSHSession{instanceID: instanceID}, func() {}, nil
+	}
+	t.Cleanup(func() { poolSetupTunnel = prev })
+
+	poolTTL = 50 * time.Second
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	const N = 8
+	type result struct {
+		done func(bool)
+		err  error
+	}
+	results := make(chan result, N)
+	var startWg sync.WaitGroup
+	startWg.Add(N)
+	for i := 0; i < N; i++ {
+		go func() {
+			startWg.Done()
+			_, done, err := pool.acquire(ctx, "i-concurrent")
+			results <- result{done, err}
+		}()
+	}
+	startWg.Wait()
+	// give all N goroutines time to enter pool.acquire
+	time.Sleep(20 * time.Millisecond)
+	close(gate)
+
+	for i := 0; i < N; i++ {
+		r := <-results
+		if r.err != nil {
+			t.Fatalf("acquire %d: %v", i, r.err)
+		}
+		r.done(false)
+	}
+
+	if got := atomic.LoadInt64(setupCount); got != 1 {
+		t.Errorf("expected 1 setup across %d concurrent acquires, got %d", N, got)
+	}
+}
+
+// TestEICTunnelPool_TTLZeroDisablesPooling pins the escape hatch:
+// poolTTL=0 means every acquire goes straight through to setup +
+// cleanup, no entry kept. Useful for tests / opt-out.
+func TestEICTunnelPool_TTLZeroDisablesPooling(t *testing.T) {
+	setupCount, cleanupCount, _, unblock := withPoolSetupStub(t)
+	poolTTL = 0
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	_, done, err := pool.acquire(ctx, "i-ttlzero")
+	if err != nil {
+		t.Fatalf("acquire 1: %v", err)
+	}
+	done(false)
+
+	unblock()
+	_, done, err = pool.acquire(ctx, "i-ttlzero")
+	if err != nil {
+		t.Fatalf("acquire 2: %v", err)
+	}
+	done(false)
+
+	if got := atomic.LoadInt64(setupCount); got != 2 {
+		t.Errorf("expected 2 setups with TTL=0 (pool disabled), got %d", got)
+	}
+	if got := atomic.LoadInt64(cleanupCount); got != 2 {
+		t.Errorf("expected 2 cleanups with TTL=0 (each release closes), got %d", got)
+	}
+}
+
+// TestEICTunnelPool_LRUEvictionAtCap pins the LRU defence: when the
+// pool reaches poolMaxEntries, a new acquire for an unseen
+// instanceID evicts the LRU idle entry instead of growing unbounded.
+func TestEICTunnelPool_LRUEvictionAtCap(t *testing.T) {
+	setupCount, cleanupCount, _, _ := withPoolSetupStub(t)
+	prev := poolMaxEntries
+	poolMaxEntries = 2
+	t.Cleanup(func() { poolMaxEntries = prev })
+	poolTTL = 50 * time.Second
+
+	// Replace stub with one that doesn't gate so we can fill quickly.
+	poolSetupTunnel = func(ctx context.Context, instanceID string) (
+		eicSSHSession, func(), error) {
+		atomic.AddInt64(setupCount, 1)
+		return eicSSHSession{instanceID: instanceID}, func() {
+			atomic.AddInt64(cleanupCount, 1)
+		}, nil
+	}
+
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	for _, id := range []string{"i-1", "i-2"} {
+		_, done, err := pool.acquire(ctx, id)
+		if err != nil {
+			t.Fatalf("acquire %s: %v", id, err)
+		}
+		done(false)
+	}
+	// Both entries idle, pool at cap.
+	_, done, err := pool.acquire(ctx, "i-3")
+	if err != nil {
+		t.Fatalf("acquire i-3: %v", err)
+	}
+	done(false)
+
+	// Wait for the goroutine'd cleanup of the evicted entry.
+	deadline := time.Now().Add(500 * time.Millisecond)
+	for atomic.LoadInt64(cleanupCount) < 1 && time.Now().Before(deadline) {
+		time.Sleep(10 * time.Millisecond)
+	}
+
+	if got := atomic.LoadInt64(setupCount); got != 3 {
+		t.Errorf("expected 3 setups (one per unique instance), got %d", got)
+	}
+	if got := atomic.LoadInt64(cleanupCount); got < 1 {
+		t.Errorf("expected ≥1 cleanup (LRU eviction), got %d", got)
+	}
+}
+
+// TestEICTunnelPool_PoisonedClassification pins the heuristic that
+// distinguishes tunnel-fatal errors (poison the entry) from
+// app-level errors (file not found, validation) that should NOT
+// invalidate the tunnel.
+func TestEICTunnelPool_PoisonedClassification(t *testing.T) {
+	cases := []struct {
+		name string
+		err  error
+		want bool
+	}{
+		{"nil", nil, false},
+		{"file not found", errors.New("os: file does not exist"), false},
+		{"validation", errors.New("invalid path: must be relative"), false},
+		{"connection refused", errors.New("ssh: connect to host: connection refused"), true},
+		{"connection refused upper", errors.New("Connection Refused"), true},
+		{"broken pipe", errors.New("write tunnel: broken pipe"), true},
+		{"permission denied", errors.New("Permission denied (publickey)"), true},
+		{"auth failed", errors.New("Authentication failed"), true},
+		{"connection reset", errors.New("Connection reset by peer"), true},
+		{"port forward", errors.New("port forwarding failed"), true},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			got := fnErrIndicatesTunnelFault(tc.err)
+			if got != tc.want {
+				t.Errorf("fnErrIndicatesTunnelFault(%v) = %v, want %v",
+					tc.err, got, tc.want)
+			}
+		})
+	}
+}
+
+// TestEICTunnelPool_RefcountBlocksEviction pins that an entry past
+// TTL is NOT evicted while a caller still holds it — preventing
+// use-after-free in the holder.
+func TestEICTunnelPool_RefcountBlocksEviction(t *testing.T) {
+	setupCount, cleanupCount, _, _ := withPoolSetupStub(t)
+	poolTTL = 30 * time.Millisecond
+	poolJanitorInterval = 5 * time.Millisecond
+	pool := freshPool(t)
+	ctx := context.Background()
+
+	_, done, err := pool.acquire(ctx, "i-hold")
+	if err != nil {
+		t.Fatalf("acquire: %v", err)
+	}
+
+	// Sleep past TTL while holding the session. Janitor sweeps
+	// every 5ms but must skip our entry because refcount=1.
+	time.Sleep(80 * time.Millisecond)
+
+	if got := atomic.LoadInt64(cleanupCount); got != 0 {
+		t.Errorf("expected 0 cleanups while holder is active, got %d", got)
+	}
+
+	done(false)
+	// Now refcount=0 and entry is past TTL; releaser triggers cleanup.
+	deadline := time.Now().Add(200 * time.Millisecond)
+	for atomic.LoadInt64(cleanupCount) < 1 && time.Now().Before(deadline) {
+		time.Sleep(5 * time.Millisecond)
+	}
+	if got := atomic.LoadInt64(cleanupCount); got != 1 {
+		t.Errorf("expected 1 cleanup after release of expired entry, got %d", got)
+	}
+	if got := atomic.LoadInt64(setupCount); got != 1 {
+		t.Errorf("setupCount tracking: got %d, want 1", got)
+	}
+}
+
+// TestPooledWithEICTunnel_PanicPoisonsEntry pins that a panic
+// from fn poisons the pool entry on the way out — refcount goes
+// back to zero (no leak) and the entry is marked unusable so the
+// next acquire builds fresh. Without the defer-release pattern, a
+// panic would leave refcount=1 forever and the entry would never
+// evict.
+func TestPooledWithEICTunnel_PanicPoisonsEntry(t *testing.T) {
+	setupCount, _, _, _ := withPoolSetupStub(t)
+	poolTTL = 50 * time.Second
+	globalEICTunnelPool = newEICTunnelPool()
+	t.Cleanup(globalEICTunnelPool.stop)
+
+	func() {
+		defer func() {
+			if r := recover(); r == nil {
+				t.Errorf("expected panic to bubble up, got nil")
+			}
+		}()
+		_ = pooledWithEICTunnel(context.Background(), "i-panic",
+			func(s eicSSHSession) error { panic("boom") })
+	}()
+
+	// Replenish the gate so the next setup can run.
+	prev := poolSetupTunnel
+	poolSetupTunnel = func(ctx context.Context, instanceID string) (
+		eicSSHSession, func(), error) {
+		atomic.AddInt64(setupCount, 1)
+		return eicSSHSession{instanceID: instanceID}, func() {}, nil
+	}
+	t.Cleanup(func() { poolSetupTunnel = prev })
+
+	// Next acquire must build fresh — entry was poisoned by panic.
+	if err := pooledWithEICTunnel(context.Background(), "i-panic",
+		func(s eicSSHSession) error { return nil }); err != nil {
+		t.Fatalf("post-panic acquire: %v", err)
+	}
+	if got := atomic.LoadInt64(setupCount); got != 2 {
+		t.Errorf("expected 2 setups (panic poisoned, rebuild), got %d", got)
+	}
+}
+
+// TestPooledWithEICTunnel_PreservesFnErr pins that errors from the
+// inner fn pass through to the caller verbatim — pool wrapping
+// should not swallow or transform error semantics for app code.
+func TestPooledWithEICTunnel_PreservesFnErr(t *testing.T) {
+	withPoolSetupStub(t)
+	poolTTL = 50 * time.Second
+
+	// Reset the global pool so this test is isolated from any prior
+	// test that may have populated it.
+	globalEICTunnelPool = newEICTunnelPool()
+
+	want := errors.New("file does not exist")
+	got := pooledWithEICTunnel(context.Background(), "i-fn-err",
+		func(s eicSSHSession) error { return want })
+	if !errors.Is(got, want) {
+		t.Errorf("pooledWithEICTunnel returned %v, want %v", got, want)
+	}
+}

workspace-server/internal/handlers/mock_runtime.go

@@ -1,223 +0,0 @@
-package handlers
-
-// mock_runtime.go — "mock" runtime: a virtual workspace that has no
-// container, no EC2, no LLM, just hardcoded canned A2A replies. Built
-// for the funding-demo "200-workspace mock org" so hongming can show
-// investors a CEO/VPs/Managers/ICs hierarchy at scale without burning
-// 200 EC2 instances or 200 Anthropic keys.
-//
-// Wire model:
-//   - org template declares `runtime: mock` on every workspace
-//   - createWorkspaceTree skips provisioning, sets status='online'
-//     directly (mirrors the `external` short-circuit, minus the URL +
-//     awaiting_agent dance)
-//   - proxyA2ARequest short-circuits on a mock-runtime target and
-//     returns a canned JSON-RPC reply; never calls resolveAgentURL,
-//     never opens an HTTP connection, never touches Docker/EC2
-//
-// The reply is JSON-RPC 2.0 + a2a-sdk v0.3 shape so the canvas's
-// extractAgentText / extractTextsFromParts read it without any
-// special-casing. We rotate over a small variant pool so a screen
-// full of replies doesn't all read identical — gives the demo a bit
-// of life without pretending to be a real agent.
-
-import (
-	"context"
-	"crypto/sha1"
-	"database/sql"
-	"encoding/binary"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"log"
-	"net/http"
-	"strings"
-	"time"
-
-	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
-	"github.com/gin-gonic/gin"
-	"github.com/google/uuid"
-)
-
-// MockRuntimeName is the canonical runtime string a workspace row
-// carries to opt into the canned-reply short-circuit. Kept as a const
-// so the proxy's runtime-check + the org-import skip-block reference
-// the same literal.
-const MockRuntimeName = "mock"
-
-// mockReplyVariants is the pool of canned strings the mock runtime
-// rotates through. Picked to read like a busy-but-short reply from a
-// real human in a hierarchy — a CEO would NOT respond with "On it!",
-// but for the demo every node is shown to be reachable, so we lean
-// into the variety. Variant selection is deterministic per
-// (workspaceID, request-id) pair so a screen recording replays the
-// same reply for the same input.
-var mockReplyVariants = []string{
-	"On it!",
-	"Got it, on it now.",
-	"On it, boss.",
-	"Working on it.",
-	"Acknowledged — on it.",
-	"On it, will report back.",
-	"Roger that, on it.",
-	"Copy that. On it.",
-	"On it — ETA shortly.",
-	"On it. Standby for update.",
-}
-
-// pickMockReply returns a canned reply for the given workspaceID +
-// requestID. Deterministic so the same (workspace, message-id) pair
-// always picks the same variant — useful for screen recordings and
-// flake-free e2e snapshots. Falls back to variant[0] if the inputs
-// are empty.
-func pickMockReply(workspaceID, requestID string) string {
-	if len(mockReplyVariants) == 0 {
-		return "On it!"
-	}
-	if workspaceID == "" && requestID == "" {
-		return mockReplyVariants[0]
-	}
-	h := sha1.Sum([]byte(workspaceID + ":" + requestID))
-	idx := int(binary.BigEndian.Uint32(h[0:4]) % uint32(len(mockReplyVariants)))
-	return mockReplyVariants[idx]
-}
-
-// lookupRuntime returns the workspace's runtime string. Empty when the
-// row is missing / DB hiccup so callers fall through to the existing
-// dispatch path (which will then 404 / 502 normally). Fail-open here
-// because a transient DB error must not silently flip a real workspace
-// into mock-mode and start handing out canned replies in place of
-// genuine agent traffic.
-func lookupRuntime(ctx context.Context, workspaceID string) string {
-	var runtime sql.NullString
-	err := db.DB.QueryRowContext(ctx,
-		`SELECT runtime FROM workspaces WHERE id = $1`, workspaceID,
-	).Scan(&runtime)
-	if err != nil {
-		if !errors.Is(err, sql.ErrNoRows) {
-			log.Printf("ProxyA2A: lookupRuntime(%s) failed (%v) — falling through to dispatch path", workspaceID, err)
-		}
-		return ""
-	}
-	if !runtime.Valid {
-		return ""
-	}
-	return runtime.String
-}
-
-// buildMockA2AResponse synthesises a JSON-RPC 2.0 success envelope that
-// matches the a2a-sdk v0.3 reply shape the canvas's extractAgentText
-// already understands: `{result: {parts: [{kind: "text", text: ...}]}}`.
-// `requestID` is the JSON-RPC `id` of the inbound request — A2A
-// implementations echo it on the reply so callers can correlate. We
-// extract it from the normalized payload in the caller and pass it in
-// here so this function stays JSON-only (no payload parsing).
-//
-// Returns marshalled bytes ready to write straight to the HTTP body.
-// Marshal failure is logged + a tiny fallback envelope returned, since
-// failing the whole request because of a JSON encoding hiccup on a
-// constant-shaped payload would defeat the "mock always works" guarantee.
-func buildMockA2AResponse(workspaceID, requestID, replyText string) []byte {
-	if requestID == "" {
-		requestID = uuid.New().String()
-	}
-	envelope := map[string]any{
-		"jsonrpc": "2.0",
-		"id":      requestID,
-		"result": map[string]any{
-			"parts": []map[string]any{
-				{"kind": "text", "text": replyText},
-			},
-		},
-	}
-	out, err := json.Marshal(envelope)
-	if err != nil {
-		log.Printf("ProxyA2A: mock-runtime response marshal failed for %s: %v — emitting fallback", workspaceID, err)
-		// Hand-rolled minimal envelope. Safe because every value is a
-		// hardcoded constant string with no characters that need
-		// escaping in a JSON string literal.
-		fallback := fmt.Sprintf(
-			`{"jsonrpc":"2.0","id":%q,"result":{"parts":[{"kind":"text","text":%q}]}}`,
-			requestID, replyText,
-		)
-		return []byte(fallback)
-	}
-	return out
-}
-
-// extractRequestID pulls the JSON-RPC `id` out of an already-normalized
-// A2A payload. Returns "" when the field is absent or not a string —
-// caller substitutes a fresh UUID. Tolerant of every shape
-// normalizeA2APayload could produce.
-func extractRequestID(body []byte) string {
-	var top map[string]json.RawMessage
-	if err := json.Unmarshal(body, &top); err != nil {
-		return ""
-	}
-	raw, ok := top["id"]
-	if !ok {
-		return ""
-	}
-	var s string
-	if json.Unmarshal(raw, &s) == nil {
-		return s
-	}
-	// JSON-RPC permits numeric IDs too; canvas issues UUIDs but be
-	// defensive against alternative SDKs.
-	var n json.Number
-	if json.Unmarshal(raw, &n) == nil {
-		return n.String()
-	}
-	return ""
-}
-
-// handleMockA2A is the proxy short-circuit for mock-runtime workspaces.
-// Returns (status, body, true) when the target is mock — caller writes
-// the response and returns. Returns (_, _, false) when the target is
-// not mock — caller continues to the real dispatch path.
-//
-// Side-effects: writes a synthetic activity_logs row via logA2ASuccess
-// when logActivity is true so the canvas's "Agent Comms" tab shows the
-// mock reply in the trace alongside real-agent traffic. Without this
-// the demo would render messages on the canvas chat panel but a peer
-// node clicking through to its activity tab would see an empty list.
-func (h *WorkspaceHandler) handleMockA2A(ctx context.Context, workspaceID, callerID string, body []byte, a2aMethod string, logActivity bool) (int, []byte, bool) {
-	if lookupRuntime(ctx, workspaceID) != MockRuntimeName {
-		return 0, nil, false
-	}
-	requestID := extractRequestID(body)
-	replyText := pickMockReply(workspaceID, requestID)
-	respBody := buildMockA2AResponse(workspaceID, requestID, replyText)
-
-	// Tiny artificial delay so the canvas chat UI has time to render
-	// the user's outgoing bubble before the agent reply appears.
-	// Without it the reply lands the same animation frame and feels
-	// robotic. 80ms is too fast to look "real" but masks the React
-	// double-render race that drops the user bubble entirely on slow
-	// machines (observed locally on M1 Air, 2026-05-07). Below 200ms
-	// keeps a 200-node demo snappy when investors fan out 30 messages
-	// at once.
-	time.Sleep(80 * time.Millisecond)
-
-	if logActivity {
-		// Reuse the existing success-logger so the activity feed shape
-		// is identical to a real agent reply. Status 200 + duration 0
-		// is the "synthesised reply" marker; activity_logs.duration_ms
-		// being 0 is harmless (real fast paths can hit 0 too).
-		h.logA2ASuccess(ctx, workspaceID, callerID, body, respBody, a2aMethod, http.StatusOK, 0)
-	}
-	return http.StatusOK, respBody, true
-}
-
-// IsMockRuntime is a small public helper for callers outside this
-// package (tests, the org importer) that need to ask the question
-// without depending on the unexported constant. Trims + lower-cases
-// so a typoed YAML cell like "  Mock " still resolves correctly.
-func IsMockRuntime(runtime string) bool {
-	return strings.EqualFold(strings.TrimSpace(runtime), MockRuntimeName)
-}
-
-// gin import is unused at file scope but kept as a tag so a future
-// addition of a thin HTTP handler (e.g. POST /workspaces/:id/mock/replies
-// for an admin-set custom reply pool) doesn't need an import re-order.
-var _ = gin.H{}

workspace-server/internal/handlers/mock_runtime_test.go

@@ -1,266 +0,0 @@
-package handlers
-
-// mock_runtime_test.go — locks the contract for the mock-runtime
-// short-circuit added for the funding-demo "200-workspace mock org"
-// template. Three invariants:
-//
-//   1. ProxyA2A on a workspace with runtime='mock' must return 200
-//      with a JSON-RPC reply containing one text part. NO HTTP
-//      dispatch, NO resolveAgentURL DB read (mock workspaces have
-//      no URL — that read would 404 and break the demo).
-//
-//   2. The reply text must be one of the canned variants and must be
-//      deterministic for a given (workspace_id, request_id) pair so
-//      screen recordings replay identically.
-//
-//   3. Workspaces with runtime != 'mock' must NOT be affected — the
-//      mock check fails fast and falls through to the existing
-//      dispatch path. Same kind of regression guard the poll-mode
-//      tests carry.
-
-import (
-	"bytes"
-	"encoding/json"
-	"net/http"
-	"net/http/httptest"
-	"testing"
-	"time"
-
-	"github.com/DATA-DOG/go-sqlmock"
-	"github.com/gin-gonic/gin"
-)
-
-// TestProxyA2A_MockRuntime_ReturnsCannedReply is the happy-path
-// contract. A workspace flagged runtime='mock' must:
-//   - return 200 with JSON-RPC envelope {result:{parts:[{kind:text,text:...}]}}
-//   - not dispatch HTTP (no SELECT url SQL expected)
-//   - reply text is one of mockReplyVariants
-func TestProxyA2A_MockRuntime_ReturnsCannedReply(t *testing.T) {
-	mock := setupTestDB(t)
-	setupTestRedis(t)
-	broadcaster := newTestBroadcaster()
-	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
-
-	const wsID = "ws-mock-canned"
-
-	// Budget check fires before runtime lookup (same as the poll-mode
-	// short-circuit) — keeps mock workspaces honest if a tenant ever
-	// sets a budget on one. Unlikely on a demo, but the guard stays
-	// uniform so future "monthly_spend on mock = 0" assertions don't
-	// drift.
-	expectBudgetCheck(mock, wsID)
-
-	// lookupDeliveryMode runs first — return push so the poll
-	// short-circuit doesn't fire and we hit the mock check.
-	mock.ExpectQuery("SELECT delivery_mode FROM workspaces WHERE id").
-		WithArgs(wsID).
-		WillReturnRows(sqlmock.NewRows([]string{"delivery_mode"}).AddRow("push"))
-
-	// lookupRuntime SELECT — returns 'mock', triggering the canned-reply
-	// short-circuit. CRITICAL: NO ExpectQuery for `SELECT url, status
-	// FROM workspaces` (resolveAgentURL's query). If the short-circuit
-	// fails to fire, sqlmock will surface "unexpected query" on the URL
-	// SELECT and the test fails loudly — that's the dispatch-leak detector.
-	mock.ExpectQuery("SELECT runtime FROM workspaces WHERE id").
-		WithArgs(wsID).
-		WillReturnRows(sqlmock.NewRows([]string{"runtime"}).AddRow("mock"))
-
-	// Activity log: logA2ASuccess writes the synthetic reply to
-	// activity_logs so the canvas's Agent Comms tab shows it alongside
-	// real-agent traffic.
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WillReturnResult(sqlmock.NewResult(0, 1))
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: wsID}}
-
-	body := `{"jsonrpc":"2.0","id":"req-mock-1","method":"message/send","params":{"message":{"role":"user","parts":[{"kind":"text","text":"hello mock"}]}}}`
-	c.Request = httptest.NewRequest("POST", "/workspaces/"+wsID+"/a2a", bytes.NewBufferString(body))
-	c.Request.Header.Set("Content-Type", "application/json")
-
-	handler.ProxyA2A(c)
-
-	// logA2ASuccess fires async — give it a moment to settle so
-	// ExpectationsWereMet doesn't flake.
-	time.Sleep(200 * time.Millisecond)
-
-	if w.Code != http.StatusOK {
-		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
-	}
-	var resp map[string]interface{}
-	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
-		t.Fatalf("response is not valid JSON: %v", err)
-	}
-	if resp["jsonrpc"] != "2.0" {
-		t.Errorf("response.jsonrpc = %v, want 2.0", resp["jsonrpc"])
-	}
-	if resp["id"] != "req-mock-1" {
-		t.Errorf("response.id = %v, want %q (echoed from request)", resp["id"], "req-mock-1")
-	}
-	result, _ := resp["result"].(map[string]interface{})
-	if result == nil {
-		t.Fatalf("response.result missing or wrong type: %v", resp["result"])
-	}
-	parts, _ := result["parts"].([]interface{})
-	if len(parts) != 1 {
-		t.Fatalf("expected exactly one part, got %d: %v", len(parts), parts)
-	}
-	part, _ := parts[0].(map[string]interface{})
-	if part["kind"] != "text" {
-		t.Errorf("part.kind = %v, want text", part["kind"])
-	}
-	text, _ := part["text"].(string)
-	if text == "" {
-		t.Error("part.text is empty — canned reply not populated")
-	}
-	// Reply must be one of the variants.
-	matched := false
-	for _, v := range mockReplyVariants {
-		if v == text {
-			matched = true
-			break
-		}
-	}
-	if !matched {
-		t.Errorf("reply text %q is not in mockReplyVariants", text)
-	}
-
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet sqlmock expectations: %v", err)
-	}
-}
-
-// TestProxyA2A_NonMockRuntime_NoShortCircuit verifies the symmetric
-// contract: a workspace with a real runtime (claude-code, hermes, etc.)
-// must NOT be affected by the mock check — it falls through to the
-// real dispatch path. Without this guard, a regression in
-// lookupRuntime could silently flip every workspace into mock-mode
-// and start handing out canned replies in place of real-agent traffic.
-func TestProxyA2A_NonMockRuntime_NoShortCircuit(t *testing.T) {
-	mock := setupTestDB(t)
-	mr := setupTestRedis(t)
-	allowLoopbackForTest(t)
-	broadcaster := newTestBroadcaster()
-	handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", t.TempDir())
-
-	const wsID = "ws-real-runtime"
-
-	dispatched := false
-	agentServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		dispatched = true
-		w.Header().Set("Content-Type", "application/json")
-		w.Write([]byte(`{"jsonrpc":"2.0","id":"1","result":{"status":"ok"}}`))
-	}))
-	defer agentServer.Close()
-	mr.Set("ws:"+wsID+":url", agentServer.URL)
-
-	expectBudgetCheck(mock, wsID)
-
-	// poll-mode SELECT — return push so we proceed past the poll
-	// short-circuit.
-	mock.ExpectQuery("SELECT delivery_mode FROM workspaces WHERE id").
-		WithArgs(wsID).
-		WillReturnRows(sqlmock.NewRows([]string{"delivery_mode"}).AddRow("push"))
-
-	// runtime SELECT — return claude-code so the mock check falls
-	// through.
-	mock.ExpectQuery("SELECT runtime FROM workspaces WHERE id").
-		WithArgs(wsID).
-		WillReturnRows(sqlmock.NewRows([]string{"runtime"}).AddRow("claude-code"))
-
-	mock.ExpectExec("INSERT INTO activity_logs").
-		WillReturnResult(sqlmock.NewResult(0, 1))
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: wsID}}
-	body := `{"jsonrpc":"2.0","id":"real-1","method":"message/send","params":{"message":{"role":"user","parts":[{"kind":"text","text":"hi"}]}}}`
-	c.Request = httptest.NewRequest("POST", "/workspaces/"+wsID+"/a2a", bytes.NewBufferString(body))
-	c.Request.Header.Set("Content-Type", "application/json")
-
-	handler.ProxyA2A(c)
-
-	time.Sleep(50 * time.Millisecond)
-
-	if w.Code != http.StatusOK {
-		t.Fatalf("expected 200, got %d: %s", w.Code, w.Body.String())
-	}
-	if !dispatched {
-		t.Error("non-mock runtime: expected the agent server to receive the request, but it did not — mock short-circuit may be over-firing")
-	}
-	if err := mock.ExpectationsWereMet(); err != nil {
-		t.Errorf("unmet sqlmock expectations: %v", err)
-	}
-}
-
-// TestPickMockReply_Deterministic locks the determinism contract:
-// the same (workspaceID, requestID) input must yield the same variant
-// every call. Required for screen recordings + flake-free e2e
-// snapshots.
-func TestPickMockReply_Deterministic(t *testing.T) {
-	cases := []struct {
-		ws, req string
-	}{
-		{"ws-1", "req-A"},
-		{"ws-1", "req-B"},
-		{"ws-2", "req-A"},
-		{"", ""},
-	}
-	for _, tc := range cases {
-		first := pickMockReply(tc.ws, tc.req)
-		for i := 0; i < 10; i++ {
-			next := pickMockReply(tc.ws, tc.req)
-			if next != first {
-				t.Errorf("pickMockReply(%q,%q) is not deterministic: got %q then %q",
-					tc.ws, tc.req, first, next)
-			}
-		}
-	}
-}
-
-// TestIsMockRuntime_TrimsAndCaseInsensitive — typos and stray
-// whitespace in YAML must still resolve to mock so a single
-// runtime: " Mock " entry doesn't silently get dispatched.
-func TestIsMockRuntime_TrimsAndCaseInsensitive(t *testing.T) {
-	cases := map[string]bool{
-		"mock":      true,
-		"MOCK":      true,
-		"  Mock  ":  true,
-		"mocky":     false,
-		"":          false,
-		"external":  false,
-		"claude-code": false,
-	}
-	for in, want := range cases {
-		if got := IsMockRuntime(in); got != want {
-			t.Errorf("IsMockRuntime(%q) = %v, want %v", in, got, want)
-		}
-	}
-}
-
-// TestBuildMockA2AResponse_EchoesRequestID — JSON-RPC requires the
-// reply id to match the request id so callers can correlate. Mock
-// must hold this contract or canvas's correlation logic breaks.
-func TestBuildMockA2AResponse_EchoesRequestID(t *testing.T) {
-	out := buildMockA2AResponse("ws-x", "req-echo-7", "On it!")
-	var resp map[string]interface{}
-	if err := json.Unmarshal(out, &resp); err != nil {
-		t.Fatalf("response is not valid JSON: %v", err)
-	}
-	if resp["id"] != "req-echo-7" {
-		t.Errorf("id = %v, want req-echo-7", resp["id"])
-	}
-	if resp["jsonrpc"] != "2.0" {
-		t.Errorf("jsonrpc = %v, want 2.0", resp["jsonrpc"])
-	}
-	result, _ := resp["result"].(map[string]interface{})
-	parts, _ := result["parts"].([]interface{})
-	if len(parts) != 1 {
-		t.Fatalf("expected 1 part, got %d", len(parts))
-	}
-	p, _ := parts[0].(map[string]interface{})
-	if p["text"] != "On it!" {
-		t.Errorf("part.text = %v, want On it!", p["text"])
-	}
-}

workspace-server/internal/handlers/org_import.go

@@ -250,21 +250,6 @@ func (h *OrgHandler) createWorkspaceTree(ws OrgWorkspace, parentID *string, absX
 		h.broadcaster.RecordAndBroadcast(ctx, string(events.EventWorkspaceOnline), id, map[string]interface{}{
 			"name": ws.Name, "external": true,
 		})
-	} else if IsMockRuntime(runtime) {
-		// Mock-runtime workspaces have no container, no EC2, no URL —
-		// the proxyA2ARequest short-circuit synthesises every reply
-		// from a canned variant pool (see mock_runtime.go). Status
-		// goes straight to 'online' so the canvas renders the node
-		// as reachable + the chat tab's send button is enabled. No
-		// URL is set; the proxy never tries to resolve one for mock
-		// runtimes. Built for the funding-demo "200-workspace mock
-		// org" template — visual scale without real backend cost.
-		if _, err := db.DB.ExecContext(ctx, `UPDATE workspaces SET status = $1 WHERE id = $2`, models.StatusOnline, id); err != nil {
-			log.Printf("Org import: mock workspace status update failed for %s: %v", ws.Name, err)
-		}
-		h.broadcaster.RecordAndBroadcast(ctx, string(events.EventWorkspaceOnline), id, map[string]interface{}{
-			"name": ws.Name, "mock": true, "runtime": runtime,
-		})
 	} else if h.workspace.HasProvisioner() {
 		// Provision container — either backend (CP for SaaS, local Docker
 		// for self-hosted) is fine. Pre-2026-05-05 this gate was
@@ -690,23 +675,7 @@ func (h *OrgHandler) recurseChildrenForImport(ws OrgWorkspace, parentID string,
 		if err := h.createWorkspaceTree(child, &parentID, childAbsX, childAbsY, slotX, slotY, defaults, orgBaseDir, results, provisionSem); err != nil {
 			return err
 		}
-		// Pacing exists to throttle Docker container-spawn thundering
-		// during a self-hosted import. Mock-runtime children spawn no
-		// container — no Docker pressure, no LLM bursts, just DB
-		// inserts + a broadcast. Skipping the 2s sleep collapses a
-		// 200-workspace mock-org import from ~7min → ~5s, which is
-		// the difference between a snappy demo and a "did it freeze?"
-		// staring contest. Real (containerful) runtimes still pace.
-		// Inheritance: if the child itself doesn't declare a runtime,
-		// fall back to defaults.runtime — the org template sets
-		// runtime: mock once at the org level, not on every IC node.
-		childRuntime := child.Runtime
-		if childRuntime == "" {
-			childRuntime = defaults.Runtime
-		}
-		if !IsMockRuntime(childRuntime) {
-			time.Sleep(workspaceCreatePacingMs * time.Millisecond)
-		}
+		time.Sleep(workspaceCreatePacingMs * time.Millisecond)
 	}
 	return nil
 }

workspace-server/internal/handlers/plugins.go

@@ -4,7 +4,6 @@ import (
 	"bytes"
 	"context"
 	"io"
-	"log"
 	"os"
 	"path/filepath"
 	"strings"
@@ -178,42 +177,16 @@ func strDefault(m map[string]interface{}, key, fallback string) string {
 	return fallback
 }
 
-// findRunningContainer returns the live container name for workspaceID, or ""
-// when the container is genuinely not running OR the daemon errored
-// transiently. Routed through provisioner.RunningContainerName as the SSOT
-// (molecule-core#10) so this handler agrees with healthsweep on the same
-// inputs. Transient daemon errors are logged distinctly so triage doesn't
-// confuse a flaky daemon with a stopped container.
 func (h *PluginsHandler) findRunningContainer(ctx context.Context, workspaceID string) string {
-	name, err := provisioner.RunningContainerName(ctx, h.docker, workspaceID)
-	if err != nil {
-		log.Printf("plugins: docker inspect transient error for %s: %v (treating as not-running for this request)", workspaceID, err)
+	if h.docker == nil {
 		return ""
 	}
-	return name
-}
-
-// isExternalRuntime reports whether the workspace's runtime is the
-// `external` (remote-pull) shape introduced in Phase 30. External
-// workspaces have no local container — `POST /plugins` (push-install via
-// docker exec) doesn't apply to them; they pull via the download endpoint
-// instead. Returns false (allow-install) if the lookup is unwired or
-// errors — failing open here is safe because the downstream
-// findRunningContainer step still gates on a real container being there.
-//
-// Background — molecule-core#10: without this check, external workspaces
-// fall through to findRunningContainer's NotFound path and return a
-// misleading 503 "container not running" instead of a clear "use the
-// pull endpoint" message.
-func (h *PluginsHandler) isExternalRuntime(workspaceID string) bool {
-	if h.runtimeLookup == nil {
-		return false
+	name := provisioner.ContainerName(workspaceID)
+	info, err := h.docker.ContainerInspect(ctx, name)
+	if err == nil && info.State.Running {
+		return name
 	}
-	runtime, err := h.runtimeLookup(workspaceID)
-	if err != nil {
-		return false
-	}
-	return runtime == "external"
+	return ""
 }
 
 func (h *PluginsHandler) execAsRoot(ctx context.Context, containerName string, cmd []string) (string, error) {

workspace-server/internal/handlers/plugins_findrunning_ssot_test.go

@@ -1,176 +0,0 @@
-package handlers
-
-import (
-	"go/ast"
-	"go/parser"
-	"go/token"
-	"strings"
-	"testing"
-)
-
-// TestFindRunningContainer_RoutesThroughProvisionerSSOT is a behavior-based
-// AST gate: it pins the invariant that PluginsHandler.findRunningContainer
-// MUST go through provisioner.RunningContainerName for its is-running check,
-// instead of carrying its own copy of cli.ContainerInspect logic.
-//
-// Background — molecule-core#10: a parallel impl of "is the workspace's
-// container running" used to live in plugins.go. It drifted from the
-// canonical impl in healthsweep (which goes through Provisioner.IsRunning
-// → RunningContainerName) on edge cases like "transient daemon error" —
-// the duplicate would 503 with a misleading message while healthsweep
-// correctly stayed defensive. Consolidating onto RunningContainerName as
-// the SSOT prevents any future copy from re-introducing that drift.
-//
-// Mutation invariant: if a future PR replaces the provisioner call with
-// `h.docker.ContainerInspect(...)` directly, this test fails. That's the
-// signal to either (a) extend RunningContainerName's contract OR (b)
-// document why this call site needs to differ. Either way: the drift
-// gets a reviewer's attention instead of shipping silently.
-func TestFindRunningContainer_RoutesThroughProvisionerSSOT(t *testing.T) {
-	fset := token.NewFileSet()
-	file, err := parser.ParseFile(fset, "plugins.go", nil, parser.ParseComments)
-	if err != nil {
-		t.Fatalf("parse plugins.go: %v", err)
-	}
-
-	var fn *ast.FuncDecl
-	ast.Inspect(file, func(n ast.Node) bool {
-		f, ok := n.(*ast.FuncDecl)
-		if !ok || f.Name.Name != "findRunningContainer" {
-			return true
-		}
-		// Confirm receiver is *PluginsHandler so we don't pick up an unrelated
-		// helper of the same name. ast.Recv is a FieldList — receivers carry
-		// at most one field.
-		if f.Recv == nil || len(f.Recv.List) == 0 {
-			return true
-		}
-		fn = f
-		return false
-	})
-
-	if fn == nil {
-		t.Fatal("findRunningContainer not found in plugins.go — was it renamed? update this test or the SSOT routing assumption")
-	}
-
-	var (
-		callsRunningContainerName bool
-		callsContainerInspectRaw  bool
-	)
-	ast.Inspect(fn.Body, func(n ast.Node) bool {
-		call, ok := n.(*ast.CallExpr)
-		if !ok {
-			return true
-		}
-		sel, ok := call.Fun.(*ast.SelectorExpr)
-		if !ok {
-			return true
-		}
-		// Pkg.Func form: provisioner.RunningContainerName(...)
-		if pkgIdent, ok := sel.X.(*ast.Ident); ok {
-			if pkgIdent.Name == "provisioner" && sel.Sel.Name == "RunningContainerName" {
-				callsRunningContainerName = true
-			}
-		}
-		// Receiver-then-method form: h.docker.ContainerInspect(...) /
-		// p.cli.ContainerInspect(...) — anything ending in
-		// .ContainerInspect that's NOT routed through provisioner.
-		if sel.Sel.Name == "ContainerInspect" {
-			callsContainerInspectRaw = true
-		}
-		return true
-	})
-
-	if !callsRunningContainerName {
-		t.Errorf(
-			"findRunningContainer must call provisioner.RunningContainerName for the SSOT inspect — see molecule-core#10. Found no such call.",
-		)
-	}
-	if callsContainerInspectRaw {
-		t.Errorf(
-			"findRunningContainer carries a direct ContainerInspect call. This is the parallel-impl drift molecule-core#10 fixed. " +
-				"Either route through provisioner.RunningContainerName OR — if a new use case truly needs a different inspect — extend RunningContainerName's contract first and update this gate to allow the specific delta.",
-		)
-	}
-}
-
-// TestProvisionerIsRunning_RoutesThroughRunningContainerName mirrors the
-// gate above but for the OTHER consumer of the SSOT — Provisioner.IsRunning
-// (called by healthsweep). If a future refactor makes IsRunning carry its
-// own ContainerInspect again, the two consumers' edge-case behaviors will
-// silently drift. Keep them yoked.
-func TestProvisionerIsRunning_RoutesThroughRunningContainerName(t *testing.T) {
-	fset := token.NewFileSet()
-	file, err := parser.ParseFile(fset, "../provisioner/provisioner.go", nil, parser.ParseComments)
-	if err != nil {
-		t.Fatalf("parse provisioner.go: %v", err)
-	}
-
-	var fn *ast.FuncDecl
-	ast.Inspect(file, func(n ast.Node) bool {
-		f, ok := n.(*ast.FuncDecl)
-		if !ok || f.Name.Name != "IsRunning" || f.Recv == nil {
-			return true
-		}
-		// The receiver type must be *Provisioner specifically. CPProvisioner
-		// has its own IsRunning that talks HTTP to the controlplane and is
-		// out of scope for this gate.
-		if !receiverIs(f, "Provisioner") {
-			return true
-		}
-		fn = f
-		return false
-	})
-	if fn == nil {
-		t.Fatal("Provisioner.IsRunning not found — was it renamed? update this test")
-	}
-
-	var (
-		callsRunningContainerName bool
-		callsContainerInspectRaw  bool
-	)
-	ast.Inspect(fn.Body, func(n ast.Node) bool {
-		call, ok := n.(*ast.CallExpr)
-		if !ok {
-			return true
-		}
-		// Same-package call: bare identifier (e.g. RunningContainerName(...)).
-		if id, ok := call.Fun.(*ast.Ident); ok && id.Name == "RunningContainerName" {
-			callsRunningContainerName = true
-			return true
-		}
-		// Selector call: pkg.Func (e.g. provisioner.RunningContainerName)
-		// OR recv.Method (e.g. p.cli.ContainerInspect).
-		sel, ok := call.Fun.(*ast.SelectorExpr)
-		if !ok {
-			return true
-		}
-		switch sel.Sel.Name {
-		case "RunningContainerName":
-			callsRunningContainerName = true
-		case "ContainerInspect":
-			callsContainerInspectRaw = true
-		}
-		return true
-	})
-
-	if !callsRunningContainerName {
-		t.Errorf("Provisioner.IsRunning must call RunningContainerName for the SSOT inspect — see molecule-core#10")
-	}
-	if callsContainerInspectRaw {
-		t.Errorf("Provisioner.IsRunning carries a direct ContainerInspect call; route through RunningContainerName instead")
-	}
-}
-
-// receiverIs reports whether fn's receiver is `*<typeName>` or `<typeName>`.
-func receiverIs(fn *ast.FuncDecl, typeName string) bool {
-	if fn.Recv == nil || len(fn.Recv.List) == 0 {
-		return false
-	}
-	expr := fn.Recv.List[0].Type
-	if star, ok := expr.(*ast.StarExpr); ok {
-		expr = star.X
-	}
-	id, ok := expr.(*ast.Ident)
-	return ok && strings.EqualFold(id.Name, typeName)
-}

workspace-server/internal/handlers/plugins_install.go

@@ -32,18 +32,6 @@ import (
 // inside the workspace at startup.
 func (h *PluginsHandler) Install(c *gin.Context) {
 	workspaceID := c.Param("id")
-	// External-runtime guard (molecule-core#10): push-install via docker
-	// exec is meaningless for `runtime='external'` workspaces — they have
-	// no local container. Reject early with a hint pointing at the
-	// pull-mode endpoint, instead of falling through to a misleading
-	// "container not running" 503 from findRunningContainer.
-	if h.isExternalRuntime(workspaceID) {
-		c.JSON(http.StatusUnprocessableEntity, gin.H{
-			"error": "plugin install via push is not supported for external runtimes",
-			"hint":  "external workspaces pull plugins via GET /workspaces/:id/plugins/:name/download",
-		})
-		return
-	}
 	// Cap the JSON body so a pathological POST can't exhaust parser memory.
 	bodyMax := envx.Int64("PLUGIN_INSTALL_BODY_MAX_BYTES", defaultInstallBodyMaxBytes)
 	c.Request.Body = http.MaxBytesReader(c.Writer, c.Request.Body, bodyMax)
@@ -105,16 +93,6 @@ func (h *PluginsHandler) Uninstall(c *gin.Context) {
 	pluginName := c.Param("name")
 	ctx := c.Request.Context()
 
-	// Mirror Install's external-runtime guard (molecule-core#10) so the
-	// two endpoints reject the same shape with the same message.
-	if h.isExternalRuntime(workspaceID) {
-		c.JSON(http.StatusUnprocessableEntity, gin.H{
-			"error": "plugin uninstall via docker exec is not supported for external runtimes",
-			"hint":  "external workspaces manage their own plugin directory; remove it locally",
-		})
-		return
-	}
-
 	if err := validatePluginName(pluginName); err != nil {
 		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid plugin name"})
 		return

workspace-server/internal/handlers/plugins_install_external_test.go

@@ -1,176 +0,0 @@
-package handlers
-
-import (
-	"bytes"
-	"net/http"
-	"net/http/httptest"
-	"strings"
-	"testing"
-
-	"github.com/gin-gonic/gin"
-)
-
-// TestPluginInstall_ExternalRuntime_Returns422 — molecule-core#10.
-// Install on a `runtime='external'` workspace must NOT fall through to
-// findRunningContainer (which would 503 with a misleading "container not
-// running"). It must return 422 with a hint pointing at the pull-mode
-// download endpoint.
-func TestPluginInstall_ExternalRuntime_Returns422(t *testing.T) {
-	h := NewPluginsHandler(t.TempDir(), nil, nil).
-		WithRuntimeLookup(func(workspaceID string) (string, error) {
-			return "external", nil
-		})
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "ba1789b0-4d21-4f4f-a878-fa226bf77cf5"}}
-	c.Request = httptest.NewRequest(
-		"POST",
-		"/workspaces/ba1789b0-4d21-4f4f-a878-fa226bf77cf5/plugins",
-		bytes.NewBufferString(`{"source":"local://my-plugin"}`),
-	)
-	c.Request.Header.Set("Content-Type", "application/json")
-
-	h.Install(c)
-
-	if w.Code != http.StatusUnprocessableEntity {
-		t.Errorf("expected 422 (Unprocessable Entity) for runtime='external', got %d: %s", w.Code, w.Body.String())
-	}
-	if !strings.Contains(w.Body.String(), "external runtimes") {
-		t.Errorf("expected error body to mention 'external runtimes', got: %s", w.Body.String())
-	}
-	if !strings.Contains(w.Body.String(), "download") {
-		t.Errorf("expected error body to point at the download endpoint, got: %s", w.Body.String())
-	}
-}
-
-// TestPluginUninstall_ExternalRuntime_Returns422 — symmetric guard on the
-// uninstall path (DELETE /workspaces/:id/plugins/:name). External
-// workspaces manage their own plugin directory locally; the platform
-// can't docker-exec into them.
-func TestPluginUninstall_ExternalRuntime_Returns422(t *testing.T) {
-	h := NewPluginsHandler(t.TempDir(), nil, nil).
-		WithRuntimeLookup(func(workspaceID string) (string, error) {
-			return "external", nil
-		})
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{
-		{Key: "id", Value: "ba1789b0-4d21-4f4f-a878-fa226bf77cf5"},
-		{Key: "name", Value: "my-plugin"},
-	}
-	c.Request = httptest.NewRequest(
-		"DELETE",
-		"/workspaces/ba1789b0-4d21-4f4f-a878-fa226bf77cf5/plugins/my-plugin",
-		nil,
-	)
-
-	h.Uninstall(c)
-
-	if w.Code != http.StatusUnprocessableEntity {
-		t.Errorf("expected 422 for runtime='external', got %d: %s", w.Code, w.Body.String())
-	}
-	if !strings.Contains(w.Body.String(), "external runtimes") {
-		t.Errorf("expected error body to mention 'external runtimes', got: %s", w.Body.String())
-	}
-}
-
-// TestPluginInstall_ContainerBackedRuntime_FallsThroughGuard — the runtime
-// guard MUST NOT short-circuit container-backed runtimes. With
-// `runtime='claude-code'` the install proceeds past the guard; without a
-// real plugin source it'll fail downstream (here: 404 from local resolver
-// because no plugin staged), which is the correct error to surface.
-//
-// This is the mutation-test partner: deleting the `runtime == "external"`
-// check would still pass TestPluginInstall_ExternalRuntime (because Install
-// would 404 instead of 422 — but the test asserts 422), and would still
-// pass this test (because both pre-fix and post-fix produce 404 here).
-// What this case pins is "non-external still falls through," catching
-// any over-eager guard that rejects all runtimes.
-func TestPluginInstall_ContainerBackedRuntime_FallsThroughGuard(t *testing.T) {
-	h := NewPluginsHandler(t.TempDir(), nil, nil).
-		WithRuntimeLookup(func(workspaceID string) (string, error) {
-			return "claude-code", nil
-		})
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "c7c28c0b-4ea5-4e75-9728-3ba860081708"}}
-	c.Request = httptest.NewRequest(
-		"POST",
-		"/workspaces/c7c28c0b-4ea5-4e75-9728-3ba860081708/plugins",
-		bytes.NewBufferString(`{"source":"local://nonexistent-plugin"}`),
-	)
-	c.Request.Header.Set("Content-Type", "application/json")
-
-	h.Install(c)
-
-	if w.Code == http.StatusUnprocessableEntity {
-		t.Errorf("runtime='claude-code' must fall through the external guard; got 422: %s", w.Body.String())
-	}
-	// The local resolver will fail to find the plugin → 404. Anything
-	// other than 422 (which would mean we mis-classified) is fine.
-	if w.Code != http.StatusNotFound {
-		t.Errorf("expected 404 (plugin not found in registry), got %d: %s", w.Code, w.Body.String())
-	}
-}
-
-// TestPluginInstall_NoRuntimeLookup_FailsOpen — when the runtime lookup
-// is unwired (test fixtures, niche deploy shapes) the guard MUST default
-// to allowing the install attempt. The downstream findRunningContainer
-// step still gates on a real container, so failing open here doesn't
-// expose a bypass — it just preserves backwards-compat with deployments
-// that haven't wired the lookup.
-func TestPluginInstall_NoRuntimeLookup_FailsOpen(t *testing.T) {
-	h := NewPluginsHandler(t.TempDir(), nil, nil) // NO WithRuntimeLookup
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "ws-no-lookup"}}
-	c.Request = httptest.NewRequest(
-		"POST",
-		"/workspaces/ws-no-lookup/plugins",
-		bytes.NewBufferString(`{"source":"local://nonexistent"}`),
-	)
-	c.Request.Header.Set("Content-Type", "application/json")
-
-	h.Install(c)
-
-	if w.Code == http.StatusUnprocessableEntity {
-		t.Errorf("nil runtimeLookup must fall through (fail-open); got 422: %s", w.Body.String())
-	}
-}
-
-// TestPluginInstall_RuntimeLookupErrors_FailsOpen — same fail-open story
-// for transient DB errors in the lookup. We don't want a momentary
-// Postgres hiccup to flip every plugin install into a 422.
-func TestPluginInstall_RuntimeLookupErrors_FailsOpen(t *testing.T) {
-	h := NewPluginsHandler(t.TempDir(), nil, nil).
-		WithRuntimeLookup(func(workspaceID string) (string, error) {
-			return "", errFakeDB
-		})
-
-	w := httptest.NewRecorder()
-	c, _ := gin.CreateTestContext(w)
-	c.Params = gin.Params{{Key: "id", Value: "ws-db-flake"}}
-	c.Request = httptest.NewRequest(
-		"POST",
-		"/workspaces/ws-db-flake/plugins",
-		bytes.NewBufferString(`{"source":"local://nonexistent"}`),
-	)
-	c.Request.Header.Set("Content-Type", "application/json")
-
-	h.Install(c)
-
-	if w.Code == http.StatusUnprocessableEntity {
-		t.Errorf("runtimeLookup error must fall through (fail-open); got 422: %s", w.Body.String())
-	}
-}
-
-// errFakeDB is a sentinel for the fail-open lookup-error case.
-var errFakeDB = &fakeError{msg: "synthetic db error"}
-
-type fakeError struct{ msg string }
-
-func (e *fakeError) Error() string { return e.msg }

workspace-server/internal/handlers/runtime_registry.go

@@ -78,10 +78,6 @@ var fallbackRuntimes = map[string]struct{}{
 	"openclaw":    {},
 	"codex":       {},
 	"external":    {},
-	// mock — virtual workspace with hardcoded canned A2A replies.
-	// No container, no EC2, no template repo. See mock_runtime.go
-	// for the full rationale (200-workspace funding-demo org).
-	"mock": {},
 }
 
 // loadRuntimesFromManifest builds the runtime allowlist from
@@ -108,10 +104,6 @@ func loadRuntimesFromManifest(path string) (map[string]struct{}, error) {
 		// the manifest doesn't know about it. Injected here so we
 		// don't need a special-case in every caller.
 		"external": {},
-		// mock is ALWAYS available for the same reason as external:
-		// virtual workspace, no template repo, never spawns a
-		// container. See mock_runtime.go.
-		"mock": {},
 	}
 	for _, e := range m.WorkspaceTemplates {
 		name := strings.TrimSpace(e.Name)

workspace-server/internal/handlers/workspace_restart.go

@@ -112,19 +112,6 @@ func (h *WorkspaceHandler) Restart(c *gin.Context) {
 		return
 	}
 
-	// runtime=mock: virtual workspace with canned A2A replies. No
-	// container, no EC2, no provisioning state to recycle. Mirror
-	// the external no-op so the canvas's Restart button doesn't
-	// silently fail or leak through to the (template-less) provisioner.
-	if dbRuntime == "mock" {
-		c.JSON(http.StatusOK, gin.H{
-			"status":  "noop",
-			"runtime": "mock",
-			"message": "mock workspaces have no container — restart is a no-op",
-		})
-		return
-	}
-
 	// SaaS mode: cpProv handles workspace EC2 lifecycle. Self-hosted mode:
 	// provisioner handles local Docker containers. At least one must be
 	// available — previously only `provisioner` was checked, which broke
@@ -545,9 +532,7 @@ func (h *WorkspaceHandler) runRestartCycle(workspaceID string) {
 	}
 
 	// Don't auto-restart external workspaces (no Docker container)
-	// or mock workspaces (no container, every reply is canned —
-	// see workspace-server/internal/handlers/mock_runtime.go).
-	if dbRuntime == "external" || dbRuntime == "mock" {
+	if dbRuntime == "external" {
 		return
 	}
 

workspace-server/internal/messagestore/postgres_store.go

@@ -110,10 +110,55 @@ func (s *PostgresMessageStore) List(ctx context.Context, workspaceID string, opt
 		return nil, false, err
 	}
 
+	// Wire order: oldest-first within the page so canvas (and any
+	// future client) can render chronologically without per-pair
+	// reordering. The SQL is `ORDER BY created_at DESC LIMIT N` for
+	// pagination correctness, and activityRowToChatMessages emits
+	// [user, agent] within a row — so a naive client-side flat-reverse
+	// would swap the pair (agent before user at the same timestamp).
+	// Reversing ROW-AWARE here keeps the wire shape display-ready.
+	//
+	// Algorithm: group consecutive same-timestamp messages into row
+	// chunks (1-2 messages each), reverse the chunk order, flatten.
+	// Within-row [user, agent] order is preserved. Single-message
+	// rows (no agent reply yet, or attachments-only) collapse to
+	// 1-element chunks and still reverse correctly.
+	messages = reverseRowChunks(messages)
+
 	reachedEnd := rowCount < opts.Limit
 	return messages, reachedEnd, nil
 }
 
+// reverseRowChunks groups msgs by adjacent same-Timestamp runs and
+// reverses the run order, preserving within-run order. Pairs of
+// (user, agent) emitted by activityRowToChatMessages share a
+// timestamp, so this keeps each pair internally ordered while
+// reversing the row sequence.
+func reverseRowChunks(msgs []ChatMessage) []ChatMessage {
+	if len(msgs) == 0 {
+		return msgs
+	}
+	var chunks [][]ChatMessage
+	cur := []ChatMessage{msgs[0]}
+	for i := 1; i < len(msgs); i++ {
+		if msgs[i].Timestamp == cur[len(cur)-1].Timestamp {
+			cur = append(cur, msgs[i])
+		} else {
+			chunks = append(chunks, cur)
+			cur = []ChatMessage{msgs[i]}
+		}
+	}
+	chunks = append(chunks, cur)
+	for i, j := 0, len(chunks)-1; i < j; i, j = i+1, j-1 {
+		chunks[i], chunks[j] = chunks[j], chunks[i]
+	}
+	out := make([]ChatMessage, 0, len(msgs))
+	for _, chunk := range chunks {
+		out = append(out, chunk...)
+	}
+	return out
+}
+
 // queryActivityRows is split from List so unit tests can exercise the
 // parser without spinning a real DB. Internal — alternative impls
 // shouldn't depend on the SQL shape.

workspace-server/internal/messagestore/postgres_store_test.go

@@ -14,10 +14,13 @@ package messagestore
 // legacy source the server replaces; divergence == regression.
 
 import (
+	"context"
 	"encoding/json"
 	"strings"
 	"testing"
 	"time"
+
+	"github.com/DATA-DOG/go-sqlmock"
 )
 
 const fixedTimestamp = "2026-04-25T18:00:00Z"
@@ -282,6 +285,145 @@ func TestChatHistory_NoAgentMessageWhenResponseHasNoTextNoFiles(t *testing.T) {
 	}
 }
 
+// =====================================================================
+// List() integration — sqlmock-backed end-to-end via the real handler
+// =====================================================================
+
+// TestList_WireOrderIsOldestFirstAcrossPagedRows pins the integration
+// invariant: List() returns wire-display-ready messages even though
+// the underlying SQL is `ORDER BY created_at DESC`. This is the
+// load-bearing test for PR-C-2 — without the row-aware reversal,
+// canvas would render every paired bubble in the wrong order on every
+// chat reload (agent before user within each timestamp).
+//
+// Mutation-test cover: removing the `messages = reverseRowChunks(...)`
+// call in List() must turn this test red. (The lower-level
+// TestReverseRowChunks_PreservesPairOrderAcrossRows pins the helper
+// itself; this test pins that List ACTUALLY CALLS the helper.)
+func TestList_WireOrderIsOldestFirstAcrossPagedRows(t *testing.T) {
+	db, mock, err := sqlmock.New()
+	if err != nil {
+		t.Fatalf("sqlmock.New: %v", err)
+	}
+	defer db.Close()
+
+	// Server's SQL is ORDER BY created_at DESC. Build mock rows in
+	// THAT order so the row-aware reversal has work to do.
+	rows := sqlmock.NewRows([]string{"created_at", "status", "request_body", "response_body"}).
+		AddRow(mustParseTime(t, "2026-05-05T00:03:00Z"), "ok",
+			`{"params":{"message":{"parts":[{"kind":"text","text":"u3"}]}}}`,
+			`{"result":"a3"}`).
+		AddRow(mustParseTime(t, "2026-05-05T00:02:00Z"), "ok",
+			`{"params":{"message":{"parts":[{"kind":"text","text":"u2"}]}}}`,
+			`{"result":"a2"}`).
+		AddRow(mustParseTime(t, "2026-05-05T00:01:00Z"), "ok",
+			`{"params":{"message":{"parts":[{"kind":"text","text":"u1"}]}}}`,
+			`{"result":"a1"}`)
+
+	mock.ExpectQuery(`SELECT created_at, status, request_body::text, response_body::text`).
+		WillReturnRows(rows)
+
+	store := NewPostgresMessageStore(db)
+	msgs, reachedEnd, err := store.List(context.Background(), "ws-1", ListOptions{Limit: 10})
+	if err != nil {
+		t.Fatalf("List: %v", err)
+	}
+
+	wantContents := []string{"u1", "a1", "u2", "a2", "u3", "a3"}
+	if len(msgs) != len(wantContents) {
+		t.Fatalf("len(msgs)=%d want %d; got=%v", len(msgs), len(wantContents), msgs)
+	}
+	for i, w := range wantContents {
+		if msgs[i].Content != w {
+			t.Errorf("idx %d: got %q want %q (full slice ordering broken; reverseRowChunks regressed?)", i, msgs[i].Content, w)
+		}
+	}
+	if !reachedEnd {
+		t.Errorf("3 rows < limit 10 should reach end, got reachedEnd=false")
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Errorf("sqlmock expectations: %v", err)
+	}
+}
+
+// =====================================================================
+// reverseRowChunks — wire-order helper added in PR-C-2
+// =====================================================================
+
+// TestReverseRowChunks_PreservesPairOrderAcrossRows pins the
+// row-aware reversal that List() applies before returning. Server's
+// SQL is `ORDER BY created_at DESC`, so messages come out
+// newest-row-first; activityRowToChatMessages emits [user, agent]
+// per row with same timestamp. A naive flat reversal of the messages
+// slice would flip each pair (agent before user). reverseRowChunks
+// reverses ROWS, preserving pair-internal order. Without this, canvas
+// would render every paired bubble in the wrong order on every chat
+// reload — the canvas-side reverse used to do the right thing because
+// it reversed ROWS BEFORE flattening, but PR-C/D moved the flattening
+// into the server, so the row-awareness has to live there too.
+func TestReverseRowChunks_PreservesPairOrderAcrossRows(t *testing.T) {
+	// Build messages newest-row-first as List() collects them. Each
+	// row is a pair sharing a timestamp, with [user, agent] order.
+	in := []ChatMessage{
+		{Role: "user", Content: "user_3", Timestamp: "2026-05-05T00:03:00Z"},
+		{Role: "agent", Content: "agent_3", Timestamp: "2026-05-05T00:03:00Z"},
+		{Role: "user", Content: "user_2", Timestamp: "2026-05-05T00:02:00Z"},
+		{Role: "agent", Content: "agent_2", Timestamp: "2026-05-05T00:02:00Z"},
+		{Role: "user", Content: "user_1", Timestamp: "2026-05-05T00:01:00Z"},
+		{Role: "agent", Content: "agent_1", Timestamp: "2026-05-05T00:01:00Z"},
+	}
+	got := reverseRowChunks(in)
+
+	want := []struct {
+		role, content string
+	}{
+		{"user", "user_1"}, {"agent", "agent_1"},
+		{"user", "user_2"}, {"agent", "agent_2"},
+		{"user", "user_3"}, {"agent", "agent_3"},
+	}
+	if len(got) != len(want) {
+		t.Fatalf("len(got)=%d len(want)=%d", len(got), len(want))
+	}
+	for i, w := range want {
+		if got[i].Role != w.role || got[i].Content != w.content {
+			t.Errorf("idx %d: got role=%q content=%q want role=%q content=%q",
+				i, got[i].Role, got[i].Content, w.role, w.content)
+		}
+	}
+}
+
+// TestReverseRowChunks_HandlesSingleMessageRows pins the case where
+// a row has only a user OR only an agent message (e.g., agent reply
+// not yet recorded, attachments-only user upload). Naive reversal
+// still works for single-message chunks; the test guards against a
+// future change that special-cases the 2-message-row path.
+func TestReverseRowChunks_HandlesSingleMessageRows(t *testing.T) {
+	in := []ChatMessage{
+		{Role: "user", Content: "u3", Timestamp: "2026-05-05T00:03:00Z"},
+		{Role: "user", Content: "u2", Timestamp: "2026-05-05T00:02:00Z"}, // single, no agent
+		{Role: "agent", Content: "a2", Timestamp: "2026-05-05T00:02:00Z"},
+		{Role: "user", Content: "u1", Timestamp: "2026-05-05T00:01:00Z"},
+	}
+	got := reverseRowChunks(in)
+	wantContents := []string{"u1", "u2", "a2", "u3"}
+	if len(got) != len(wantContents) {
+		t.Fatalf("len got=%d want=%d", len(got), len(wantContents))
+	}
+	for i, w := range wantContents {
+		if got[i].Content != w {
+			t.Errorf("idx %d: got %q want %q", i, got[i].Content, w)
+		}
+	}
+}
+
+// TestReverseRowChunks_EmptyInput returns nil/empty without panic.
+func TestReverseRowChunks_EmptyInput(t *testing.T) {
+	got := reverseRowChunks(nil)
+	if len(got) != 0 {
+		t.Errorf("nil input should return empty, got %v", got)
+	}
+}
+
 // =====================================================================
 // end-to-end shape — paired user + agent with same timestamp
 // =====================================================================

workspace-server/internal/provisioner/provisioner.go

@@ -1073,53 +1073,18 @@ func (p *Provisioner) IsRunning(ctx context.Context, workspaceID string) (bool,
 	if p == nil || p.cli == nil {
 		return false, ErrNoBackend
 	}
-	name, err := RunningContainerName(ctx, p.cli, workspaceID)
+	name := ContainerName(workspaceID)
+	info, err := p.cli.ContainerInspect(ctx, name)
 	if err != nil {
+		if isContainerNotFound(err) {
+			return false, nil
+		}
 		// Transient daemon error: caller treats !running as dead + restarts.
 		// Returning true + the underlying error preserves the error for
 		// metrics/logging without triggering the destructive path.
 		return true, err
 	}
-	return name != "", nil
-}
-
-// RunningContainerName returns the container name for workspaceID iff the
-// container exists AND is in the Running state. Single source of truth for
-// "what live container should I exec into for this workspace?" — used by
-// both Provisioner.IsRunning (healthsweep) and the plugins handler.
-//
-// Distinguishes three outcomes so callers can pick their own policy:
-//
-//   - ("ws-<id>", nil): container is running. Caller can exec into it.
-//   - ("",        nil): container does not exist OR exists but is stopped
-//                       (NotFound, Exited, Created, Restarting…). Caller
-//                       should treat as a definitive "not running."
-//   - ("",        err): transient daemon error (timeout, socket EOF, ctx
-//                       cancel). Caller should NOT infer "not running" —
-//                       this could be a flaky daemon under load. Decide
-//                       per-callsite whether to fail soft or hard.
-//
-// Background — molecule-core#10: the plugins handler used to carry its own
-// copy of this inspect logic (`findRunningContainer`) which collapsed
-// transient errors into the same "" return as a genuinely-stopped container.
-// That hid daemon flakes as misleading 503 "container not running" responses
-// AND let the two impls drift on edge-case behavior. This is the SSOT.
-func RunningContainerName(ctx context.Context, cli *client.Client, workspaceID string) (string, error) {
-	if cli == nil {
-		return "", ErrNoBackend
-	}
-	name := ContainerName(workspaceID)
-	info, err := cli.ContainerInspect(ctx, name)
-	if err != nil {
-		if isContainerNotFound(err) {
-			return "", nil
-		}
-		return "", err
-	}
-	if info.State.Running {
-		return name, nil
-	}
-	return "", nil
+	return info.State.Running, nil
 }
 
 // isContainerNotFound returns true when the Docker client indicates the

workspace-server/internal/registry/cp_orphan_sweeper.go

@@ -0,0 +1,149 @@
+package registry
+
+// cp_orphan_sweeper.go — SaaS-mode counterpart to orphan_sweeper.go.
+//
+// The Docker sweeper (StartOrphanSweeper) runs only when prov != nil
+// (single-tenant Docker mode); SaaS tenants run cpProv != nil and prov
+// == nil, so they get no sweep coverage from that path. This file fills
+// the gap for the deprovision split-write race documented in #2989:
+//
+//	1. handlers/workspace_crud.go:365 marks workspaces.status = 'removed'.
+//	2. workspace_crud.go:439 calls StopWorkspaceAuto → cpProv.Stop, which
+//	   issues DELETE /cp/workspaces/:id?instance_id=… to controlplane.
+//	3. If step 2 fails (CP transient 5xx, network blip, AWS hiccup), the
+//	   inline path returns a 500 to the canvas — but the DB row is already
+//	   at status='removed' with instance_id still populated. There's no
+//	   retry, and the EC2 lives forever.
+//
+// This sweeper closes that gap by re-issuing cpProv.Stop on every cycle
+// for any workspace at status='removed' with a non-NULL instance_id.
+// Stop is idempotent: AWS TerminateInstance on an already-terminated
+// instance is a no-op (per AWS docs), and CP's Deprovision handler
+// (controlplane/internal/handlers/workspace_provision.go:289) handles
+// the already-terminated and already-deleted-DNS cases via best-effort
+// guards. On Stop success, the sweeper clears instance_id so the next
+// cycle skips the row.
+//
+// Cadence + safety filters mirror the Docker sweeper:
+//   - 60s tick (OrphanSweepInterval)
+//   - 30s per-cycle deadline (orphanSweepDeadline)
+//   - LIMIT 100 per cycle so a sustained CP outage that backs up many
+//     orphans doesn't blow the request timeout; subsequent cycles drain.
+//
+// SSOT note: Stop's idempotency (no-op on empty instance_id, AWS
+// terminate on already-terminated) is the load-bearing invariant. Any
+// future change that adds non-idempotent side effects to cpProv.Stop
+// must also gate this sweeper, or it will re-execute those side effects
+// every 60s for every cleared-but-not-yet-NULL row.
+
+import (
+	"context"
+	"log"
+	"time"
+
+	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
+)
+
+// CPOrphanReaper is the dependency the SaaS-mode sweeper takes from
+// the CP provisioner. *provisioner.CPProvisioner satisfies this
+// naturally; tests inject fakes.
+type CPOrphanReaper interface {
+	Stop(ctx context.Context, workspaceID string) error
+}
+
+// cpSweepLimit caps the per-cycle row count so a sustained CP outage
+// can't make a single sweep cycle blow orphanSweepDeadline. With a
+// 60s cadence and 100-row limit, drain rate is up to 100 orphans/min,
+// which has never been approached even during the worst leak windows.
+const cpSweepLimit = 100
+
+// StartCPOrphanSweeper runs the SaaS-mode reconcile loop until ctx is
+// cancelled. nil reaper makes the loop a no-op (matches the Docker
+// sweeper's nil-tolerant pattern).
+//
+// Caller is expected to gate on `cpProv != nil` (matching how
+// StartOrphanSweeper is gated on `prov != nil` at the call site in
+// cmd/server/main.go) — passing a nil *CPProvisioner here would also
+// short-circuit but the gate at the wiring site keeps the call shape
+// symmetric across the two sweepers.
+func StartCPOrphanSweeper(ctx context.Context, reaper CPOrphanReaper) {
+	if reaper == nil {
+		log.Println("CP orphan sweeper: reaper is nil — sweeper disabled")
+		return
+	}
+	log.Printf("CP orphan sweeper started — reconciling every %s", OrphanSweepInterval)
+	ticker := time.NewTicker(OrphanSweepInterval)
+	defer ticker.Stop()
+	cpSweepOnce(ctx, reaper)
+	for {
+		select {
+		case <-ctx.Done():
+			log.Println("CP orphan sweeper: shutdown")
+			return
+		case <-ticker.C:
+			cpSweepOnce(ctx, reaper)
+		}
+	}
+}
+
+// cpSweepOnce executes one reconcile pass. Defensive against db.DB
+// being nil so a misconfigured boot doesn't panic.
+func cpSweepOnce(parent context.Context, reaper CPOrphanReaper) {
+	if db.DB == nil {
+		return
+	}
+	ctx, cancel := context.WithTimeout(parent, orphanSweepDeadline)
+	defer cancel()
+
+	rows, err := db.DB.QueryContext(ctx, `
+		SELECT id::text
+		  FROM workspaces
+		 WHERE status = 'removed'
+		   AND instance_id IS NOT NULL
+		   AND instance_id != ''
+		 ORDER BY updated_at DESC
+		 LIMIT $1
+	`, cpSweepLimit)
+	if err != nil {
+		log.Printf("CP orphan sweeper: DB query failed: %v", err)
+		return
+	}
+	defer rows.Close()
+
+	var orphanIDs []string
+	for rows.Next() {
+		var id string
+		if scanErr := rows.Scan(&id); scanErr != nil {
+			log.Printf("CP orphan sweeper: row scan failed: %v", scanErr)
+			continue
+		}
+		orphanIDs = append(orphanIDs, id)
+	}
+	if iterErr := rows.Err(); iterErr != nil {
+		log.Printf("CP orphan sweeper: rows iteration failed: %v", iterErr)
+		return
+	}
+
+	for _, id := range orphanIDs {
+		log.Printf("CP orphan sweeper: terminating leaked EC2 for removed workspace %s", id)
+		if stopErr := reaper.Stop(ctx, id); stopErr != nil {
+			// CP-side error — transient 5xx, network, AWS hiccup. Leave
+			// instance_id populated so the next cycle retries. Loud-fail
+			// only at the log layer; the user-visible 500 was already
+			// returned by the inline path that triggered this orphan.
+			log.Printf("CP orphan sweeper: Stop failed for %s: %v — retry next cycle", id, stopErr)
+			continue
+		}
+		// Stop succeeded — clear instance_id so the next cycle skips this
+		// row. We can't use a tombstone column (no schema change in this
+		// PR); NULL'ing instance_id is the SSOT signal for "no live
+		// EC2 attached." The matching SELECT predicate above stays in
+		// sync with this UPDATE.
+		if _, updErr := db.DB.ExecContext(ctx,
+			`UPDATE workspaces SET instance_id = NULL, updated_at = now() WHERE id = $1`,
+			id,
+		); updErr != nil {
+			log.Printf("CP orphan sweeper: clear instance_id failed for %s: %v — next cycle will re-Stop (idempotent)", id, updErr)
+		}
+	}
+}

workspace-server/internal/registry/cp_orphan_sweeper_test.go

@@ -0,0 +1,266 @@
+package registry
+
+import (
+	"context"
+	"errors"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/DATA-DOG/go-sqlmock"
+
+	"github.com/Molecule-AI/molecule-monorepo/platform/internal/db"
+)
+
+// fakeCPReaper is a hand-rolled CPOrphanReaper for the SaaS-mode
+// sweeper tests. Records every Stop call so tests can assert which
+// workspace IDs were re-issued.
+type fakeCPReaper struct {
+	mu        sync.Mutex
+	stopErr   map[string]error
+	stopCalls []string
+}
+
+func (f *fakeCPReaper) Stop(_ context.Context, wsID string) error {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.stopCalls = append(f.stopCalls, wsID)
+	return f.stopErr[wsID]
+}
+
+// TestCPSweepOnce_StopSucceeds_ClearsInstanceID — happy path. Single
+// removed-row with non-NULL instance_id; Stop succeeds; instance_id
+// gets NULL'd so the next cycle won't re-sweep it.
+func TestCPSweepOnce_StopSucceeds_ClearsInstanceID(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{}
+
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces\s+WHERE status = 'removed'\s+AND instance_id IS NOT NULL\s+AND instance_id != ''\s+ORDER BY updated_at DESC\s+LIMIT \$1`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow("ws-uuid-1"))
+	mock.ExpectExec(`UPDATE workspaces SET instance_id = NULL, updated_at = now\(\) WHERE id = \$1`).
+		WithArgs("ws-uuid-1").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 1 || reaper.stopCalls[0] != "ws-uuid-1" {
+		t.Fatalf("expected Stop(ws-uuid-1), got %v", reaper.stopCalls)
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+// TestCPSweepOnce_StopFails_KeepsInstanceID — CP transient failure.
+// Stop returns an error; instance_id MUST stay populated so the next
+// cycle retries. UPDATE must NOT fire.
+func TestCPSweepOnce_StopFails_KeepsInstanceID(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{
+		stopErr: map[string]error{"ws-uuid-1": errors.New("CP returned 503")},
+	}
+
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).AddRow("ws-uuid-1"))
+	// No ExpectExec for the UPDATE — sqlmock fails the test if the
+	// UPDATE fires.
+
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 1 || reaper.stopCalls[0] != "ws-uuid-1" {
+		t.Fatalf("expected Stop(ws-uuid-1), got %v", reaper.stopCalls)
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations (UPDATE should NOT have fired): %v", err)
+	}
+}
+
+// TestCPSweepOnce_NoOrphans — empty result set is the steady state in
+// healthy operation. No Stop, no UPDATE.
+func TestCPSweepOnce_NoOrphans(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{}
+
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}))
+
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 0 {
+		t.Fatalf("expected zero Stop calls, got %v", reaper.stopCalls)
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+// TestCPSweepOnce_MultipleOrphans — all rows in the batch get Stop'd
+// independently; one failure doesn't block others.
+func TestCPSweepOnce_MultipleOrphans(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{
+		stopErr: map[string]error{"ws-uuid-2": errors.New("CP 503 on ws-uuid-2")},
+	}
+
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).
+			AddRow("ws-uuid-1").
+			AddRow("ws-uuid-2").
+			AddRow("ws-uuid-3"))
+	// ws-uuid-1 succeeds → UPDATE fires.
+	mock.ExpectExec(`UPDATE workspaces SET instance_id = NULL`).
+		WithArgs("ws-uuid-1").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	// ws-uuid-2 fails → no UPDATE.
+	// ws-uuid-3 succeeds → UPDATE fires.
+	mock.ExpectExec(`UPDATE workspaces SET instance_id = NULL`).
+		WithArgs("ws-uuid-3").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 3 {
+		t.Fatalf("expected Stop on all 3 ids, got %v", reaper.stopCalls)
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+// TestCPSweepOnce_QueryError — DB transient failure. Sweep returns
+// without panicking. No Stop calls.
+func TestCPSweepOnce_QueryError(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{}
+
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnError(errors.New("connection refused"))
+
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 0 {
+		t.Fatalf("expected zero Stop calls on query error, got %v", reaper.stopCalls)
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+// TestCPSweepOnce_UpdateError_LogsButContinues — Stop succeeded but
+// the UPDATE to clear instance_id failed. Subsequent rows in the batch
+// must still process; comment in cpSweepOnce promises idempotent re-Stop
+// next cycle.
+func TestCPSweepOnce_UpdateError_LogsButContinues(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{}
+
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}).
+			AddRow("ws-uuid-1").
+			AddRow("ws-uuid-2"))
+	mock.ExpectExec(`UPDATE workspaces SET instance_id = NULL`).
+		WithArgs("ws-uuid-1").
+		WillReturnError(errors.New("UPDATE timeout"))
+	mock.ExpectExec(`UPDATE workspaces SET instance_id = NULL`).
+		WithArgs("ws-uuid-2").
+		WillReturnResult(sqlmock.NewResult(0, 1))
+
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 2 {
+		t.Fatalf("expected Stop on both ids despite UPDATE error on first, got %v", reaper.stopCalls)
+	}
+	if err := mock.ExpectationsWereMet(); err != nil {
+		t.Fatalf("unmet expectations: %v", err)
+	}
+}
+
+// TestCPSweepOnce_NilDB — defensive against db.DB being nil. Must not
+// panic; must not call Stop.
+func TestCPSweepOnce_NilDB(t *testing.T) {
+	saved := db.DB
+	db.DB = nil
+	t.Cleanup(func() { db.DB = saved })
+
+	reaper := &fakeCPReaper{}
+	cpSweepOnce(context.Background(), reaper)
+
+	if len(reaper.stopCalls) != 0 {
+		t.Fatalf("expected zero Stop calls when db.DB is nil, got %v", reaper.stopCalls)
+	}
+}
+
+// TestStartCPOrphanSweeper_NilReaperDisabled — boot-safety: a SaaS CP
+// without cpProv configured must not start the loop (immediate return,
+// no goroutine leak).
+func TestStartCPOrphanSweeper_NilReaperDisabled(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	done := make(chan struct{})
+	go func() {
+		StartCPOrphanSweeper(ctx, nil)
+		close(done)
+	}()
+	select {
+	case <-done:
+		// expected — nil reaper short-circuits.
+	case <-time.After(500 * time.Millisecond):
+		t.Fatal("StartCPOrphanSweeper(nil) did not return immediately")
+	}
+}
+
+// TestStartCPOrphanSweeper_RunsOnceImmediatelyAndOnTick — cadence
+// contract: kick off one sweep at boot (so a platform restart starts
+// healing immediately), then once per OrphanSweepInterval. Verifies
+// the loop terminates on ctx cancel.
+func TestStartCPOrphanSweeper_RunsOnceImmediatelyAndOnTick(t *testing.T) {
+	mock := setupTestDB(t)
+	reaper := &fakeCPReaper{}
+
+	// Two sweeps within the test window: one immediate, one on the
+	// first tick. We can't shrink OrphanSweepInterval (it's a const),
+	// so assert "at least one immediate sweep" and let cancel close
+	// the loop.
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}))
+	// The ticker may or may not fire in the test window depending on
+	// scheduler; tolerate both shapes by registering a second optional
+	// expectation. sqlmock fails on UNREGISTERED queries, so register
+	// one more then accept either 1 or 2 fires.
+	mock.ExpectQuery(`(?s)^\s*SELECT id::text\s+FROM workspaces`).
+		WithArgs(cpSweepLimit).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}))
+
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+	go func() {
+		StartCPOrphanSweeper(ctx, reaper)
+		close(done)
+	}()
+	// 100ms is well past the boot-sweep but well shy of the 60s
+	// interval, so the second query expectation is intentionally
+	// unmet — that's fine, sqlmock distinguishes "expected but not
+	// received" (we don't enforce here) from "unexpected query"
+	// (which would fail).
+	time.Sleep(100 * time.Millisecond)
+	cancel()
+	select {
+	case <-done:
+		// expected
+	case <-time.After(2 * time.Second):
+		t.Fatal("StartCPOrphanSweeper did not exit on ctx cancel")
+	}
+
+	// Boot sweep must have happened — without it, an operator restart
+	// after a CP outage would leave a 60s gap before the first heal.
+	// We don't assert mock.ExpectationsWereMet() here because the
+	// second query is intentionally optional.
+}

workspace-server/internal/registry/healthsweep.go

@@ -71,15 +71,9 @@ func StartHealthSweep(ctx context.Context, checker ContainerChecker, interval ti
 }
 
 func sweepOnlineWorkspaces(ctx context.Context, checker ContainerChecker, onOffline OfflineHandler) {
-	// Skip external + mock workspaces — neither has a Docker container.
-	// external: agent runs on operator's laptop, polled via heartbeat.
-	// mock: virtual workspace, every reply is canned (see
-	// workspace-server/internal/handlers/mock_runtime.go). Both would
-	// false-positive as "container gone" on every sweep tick and
-	// auto-restart would loop forever (provisioner has no template
-	// for either runtime).
+	// Skip external workspaces (runtime='external') — they have no Docker container
 	rows, err := db.DB.QueryContext(ctx,
-		`SELECT id FROM workspaces WHERE status IN ('online', 'degraded') AND COALESCE(runtime, 'langgraph') NOT IN ('external', 'mock')`)
+		`SELECT id FROM workspaces WHERE status IN ('online', 'degraded') AND COALESCE(runtime, 'langgraph') != 'external'`)
 	if err != nil {
 		log.Printf("Health sweep: query error: %v", err)
 		return

workspace-server/internal/registry/orphan_sweeper.go

@@ -413,20 +413,22 @@ func sweepStaleTokensWithoutContainer(ctx context.Context, reaper OrphanReaper)
 	// `"5m0s"` mismatch with Postgres interval grammar; passing seconds
 	// as an int keeps the binding portable.
 	graceSeconds := int(staleTokenGrace.Seconds())
-	// `runtime NOT IN ('external','mock')` is load-bearing: neither
-	// runtime has a local container, so the "no live container"
-	// predicate below would match every row and revoke its token.
-	// external: token is the off-host agent's only credential —
-	// revoking breaks the entire external-runtime feature
-	// (incident 2026-05-03). mock: same shape — no container by
-	// design, see workspace-server/internal/handlers/mock_runtime.go.
+	// `runtime != 'external'` is load-bearing: external workspaces have NO
+	// local container by design (the agent runs off-host), so the
+	// "no live container" predicate below would match every external
+	// workspace and revoke its token. The token is the off-host agent's
+	// only authentication credential — revoking breaks the entire
+	// external-runtime feature. Discovered 2026-05-03 when a fresh
+	// external workspace had its token silently revoked ~6 minutes after
+	// creation by this sweep, killing the operator's MCP heartbeat and
+	// inbox poll with `HTTP 401 — token may be revoked`.
 	rows, qErr := db.DB.QueryContext(ctx, `
 		SELECT DISTINCT t.workspace_id::text
 		  FROM workspace_auth_tokens t
 		  JOIN workspaces w ON w.id = t.workspace_id
 		 WHERE t.revoked_at IS NULL
 		   AND w.status NOT IN ('removed', 'provisioning')
-		   AND w.runtime NOT IN ('external', 'mock')
+		   AND w.runtime != 'external'
 		   AND COALESCE(t.last_used_at, t.created_at) < now() - make_interval(secs => $2)
 		   AND (
 		         cardinality($1::text[]) = 0

workspace-server/internal/registry/orphan_sweeper_test.go

@@ -26,7 +26,7 @@ import (
 // accidentally matching a future query that opens with the same column
 // name OR a regression that drops one of the load-bearing predicates.
 func expectStaleTokenSweepNoOp(mock sqlmock.Sqlmock) {
-	mock.ExpectQuery(`(?s)^\s*SELECT DISTINCT t\.workspace_id::text\s+FROM workspace_auth_tokens.*status NOT IN \('removed', 'provisioning'\).*runtime NOT IN \('external', 'mock'\)`).
+	mock.ExpectQuery(`(?s)^\s*SELECT DISTINCT t\.workspace_id::text\s+FROM workspace_auth_tokens.*status NOT IN \('removed', 'provisioning'\).*runtime != 'external'`).
 		WillReturnRows(sqlmock.NewRows([]string{"workspace_id"}))
 }
 
@@ -492,7 +492,7 @@ func TestSweepOnce_StaleTokenRevokeFiresWhenNoContainer(t *testing.T) {
 	// excludes 'external' (2026-05-03 fix — the sweep was incorrectly
 	// targeting external workspaces which have no container by design),
 	// and the staleness predicate appears in the SELECT.
-	mock.ExpectQuery(`(?s)^\s*SELECT DISTINCT t\.workspace_id::text\s+FROM workspace_auth_tokens.*status NOT IN \('removed', 'provisioning'\).*runtime NOT IN \('external', 'mock'\).*COALESCE\(t\.last_used_at, t\.created_at\) < now\(\) - make_interval`).
+	mock.ExpectQuery(`(?s)^\s*SELECT DISTINCT t\.workspace_id::text\s+FROM workspace_auth_tokens.*status NOT IN \('removed', 'provisioning'\).*runtime != 'external'.*COALESCE\(t\.last_used_at, t\.created_at\) < now\(\) - make_interval`).
 		WillReturnRows(sqlmock.NewRows([]string{"workspace_id"}).
 			AddRow(orphanedID))
 
@@ -548,7 +548,7 @@ func TestSweepOnce_StaleTokenRevokeFailureBailsLoop(t *testing.T) {
 
 	// Third-pass returns two stale-token workspaces; the first revoke
 	// errors. Loop must bail without attempting the second.
-	mock.ExpectQuery(`(?s)^\s*SELECT DISTINCT t\.workspace_id::text\s+FROM workspace_auth_tokens.*status NOT IN \('removed', 'provisioning'\).*runtime NOT IN \('external', 'mock'\)`).
+	mock.ExpectQuery(`(?s)^\s*SELECT DISTINCT t\.workspace_id::text\s+FROM workspace_auth_tokens.*status NOT IN \('removed', 'provisioning'\).*runtime != 'external'`).
 		WillReturnRows(sqlmock.NewRows([]string{"workspace_id"}).
 			AddRow("aaaa1111-0000-0000-0000-000000000000").
 			AddRow("bbbb2222-0000-0000-0000-000000000000"))

workspace/build-all.sh

@@ -2,7 +2,7 @@
 # build-all.sh — Rebuild base image and optionally adapter images.
 #
 # NOTE: Adapters have been extracted to standalone template repos:
-#   https://github.com/Molecule-AI/molecule-ai-workspace-template-<runtime>
+#   https://git.moleculesai.app/molecule-ai/molecule-ai-workspace-template-<runtime>
 #
 # This script now only builds the base image from workspace/Dockerfile.
 # Each adapter repo has its own Dockerfile that installs molecule-ai-workspace-runtime