chore(template): Security Auditor DAST must clean up its own test artifacts

[HongmingWang-Rabbit]

Summary

The Security Auditor's hourly DAST cron was creating test workspaces during security probes and never tearing them down, leaking rows into the live workspaces table. Adds an explicit 4a. DAST TEARDOWN step to the cron prompt with a track-and-cleanup pattern for workspaces, secrets, and plugins.

Root cause

Investigated under #17 (see comment 2026-04-14 02:14 UTC). The cron prompt already cleaned up secrets and plugins it created, but workspace-create probes had no teardown. Over four hourly runs we got four leaked rows with sequential IDs (aaaaaaaa-…/bbbbbbbb-…/cccccccc-…/dddddddd-…), each trapped in a restart loop against the missing config.yaml.

Platform log evidence at 2026-04-14 04:27:17 UTC (one sample, representative of every cycle):

POST  /admin/secrets                                           — create test secret
DELETE /admin/secrets/SEC-TEST-C5                              — cleanup (good)
POST  /workspaces/<PM>/plugins (404 for non-existent plugin)
DELETE /workspaces/<PM>/plugins/nonexistent-sec-test           — cleanup (good)
ProxyA2A: access denied <PM> → <BE>                            — CanCommunicate bypass probe
[workspace row `dddddddd-1234-5678-abcd-000000000004` appears in DB]
[no corresponding DELETE]

Changes

Single edit to org-templates/molecule-dev/org.yaml — adds step 4a to the Hourly security audit cron prompt:

• Declare three tracking lists at the start of step 4: TESTS_WORKSPACES, TESTS_SECRETS, TESTS_PLUGINS.
• Every probe that creates an artifact appends its ID to the right list.
• At the end of step 4 (unconditional, runs even on partial audit failure), iterate each list and issue DELETE. Uses || true so a failed cleanup doesn't abort the audit.
• Includes the prior-incident narrative inline so future editors understand why the step exists.

Why inline vs. platform-layer

A platform-level circuit-breaker (reject workspace creates from DAST, or TTL test workspaces) is still valid (proposed in #17 options B/C/D). But this template-layer fix is orthogonal and the cheapest immediate stop — it doesn't depend on any code change, and it closes the specific incident class we've already seen four times today.

Test plan

• python -c "import yaml; yaml.safe_load(open('org-templates/molecule-dev/org.yaml'))" — YAML valid.
• Next hourly cron run (after merge + template re-import): no new eeeeeeee-… row appears in workspaces.
• docker ps | grep -c 'ws-[a-z]\{8\}-123' stays at 0 across successive runs.
• Secret/plugin cleanup still works as before (no regression on existing teardown).

Related

• Provisioner: rogue test workspace in restart loop — no config.yaml, restarts forever #17 — root cause captured; this PR closes the template-layer side. Platform-level circuit-breaker remains in scope for that issue.
• chore(template): audit crons require PM-routing + GH-issue filing; add UIUX schedule #26 — audit cron routing (merged). This change sits alongside that structure without conflict.

🤖 Generated with Claude Code

[HongmingWang-Rabbit]

7-gate verification PASS:

• Gate 1 CI: 6/6 green
• Gate 2 Build: YAML-only, skipped
• Gate 3 Tests: test_api.sh 62/62 pass
• Gate 4 Security: cleanup scope bounded to self-created IDs; uses legacy /admin/secrets alias (valid); no unbounded deletion; no committed secrets
• Gate 5 Design fit: Security Auditor role unchanged; PM-routing unaffected; uses standard REST endpoints
• Gate 6 Line review: correct bash parameter expansion for ws:plugin split; || true guards against missing artifacts; quoting safe for UUIDs
• Gate 7 Canvas: N/A

Merging.