[integrity-audit] Integrity Filtering Audit — 2026-04-06

[github-actions]

Integrity Filtering Audit — github/gh-aw

Audit period: 2026-04-05T22:34Z – 2026-04-06T22:34Z (last 24 hours)
Runs analyzed: 30 completed runs in github/gh-aw
Runs with mcp-logs artifacts: 0
Actual agent (MCP Gateway) executions: 0

---

Findings Summary

Severity	Count	Description
🔴 Critical	0	—
🟡 Warning	1	Zero DIFC events — all agentic workflows gated by environment protection
🟢 Info	2	Normal command-mismatch skips; environment protection operating as expected

---

Warnings

W-1 · Zero DIFC events — no agentic workflows executed

In the last 24 hours, every .lock.yml agentic workflow run completed with action_required — i.e., the workflow reached the environment protection gate and stopped. No agent job reached execution, and therefore the MCP Gateway was never started, meaning zero DIFC integrity/secrecy labeling events occurred.

• Workflows affected: Scout, Q, Archie, Grumpy Code Reviewer, Security Review Agent, Plan Command, Mergefest, /cloclo, ACE Editor Session, Documentation Unbloat, PR Nitpick Reviewer, Resource Summarizer Agent, AI Moderator, Content Moderation (lock)
• Run count: 28 out of 30 recent runs
• Sample run IDs (all action_required): 24054341149 (Scout), 24054341181 (/cloclo), 24054341172 (Q), 24054341165 (Archie), 24054308113 (Archie), and 23+ more
• Observed over pages 1–25 of issue_comment and pull_request event runs

Impact: The DIFC pipeline cannot be validated during periods where the environment gate blocks all agent execution. Any guard misconfiguration or silent failure would go undetected.

Likely cause: All .lock.yml workflows reference a protected environment (likely agent or production) that requires a human reviewer to approve before the agent job runs. No reviewer approved any pending run in this 24-hour window.

Note: This environment protection is itself a security feature, not a vulnerability. However, sustained zero-execution periods create an observability gap for the DIFC audit trail.

Informational

I-1 · Two Grumpy Code Reviewer runs completed with agent job skipped

• Run 24054539534 and 24054548005
• Both triggered by pull_request events on branches copilot/add-artifact-download-option and copilot/add-assignees-to-create-pull-request
• pre_activation job ran and produced warning: "None of the commands [/grumpy] matched the first word (found: 'Add'). Workflow will be skipped."
• agent job was correctly skipped — command detection logic working as intended
• No artifacts, no MCP gateway invocation

I-2 · Content Moderation and AI Moderator ran successfully

• These are simple github-script moderation workflows (not MCP Gateway workflows)
• No DIFC involvement; not relevant to integrity filtering

I-3 · Historical pattern: Scout workflow shows no completed agent runs back to at least March 25

• Checked pages 1–100 of scout.lock.yml runs (~10,000 runs over ~2 weeks): all action_required or skipped
• Suggests the environment protection gate has been blocking agent execution for an extended period

---

Runs Analyzed

Run	Workflow	Branch	Agent Invoked	DIFC Events	Firewall Blocks	Status
24054626980	Grumpy Code Reviewer	main	❌ skipped	0	0	⚠️
24054548005	Grumpy Code Reviewer	copilot/add-assignees-to-create-pull-request	❌ skipped	0	0	⚠️
24054539534	Grumpy Code Reviewer	copilot/add-artifact-download-option	❌ skipped	0	0	⚠️
24054341149	Scout	copilot/add-assignees-to-create-pull-request	❌ early-exit	0	0	⚠️
24054341181	/cloclo	copilot/add-assignees-to-create-pull-request	❌ early-exit	0	0	⚠️
24054341172	Q	copilot/add-assignees-to-create-pull-request	❌ early-exit	0	0	⚠️
24054341165	Archie	copilot/add-assignees-to-create-pull-request	❌ early-exit	0	0	⚠️
24054209197	Security Review Agent	main	❌ early-exit	0	0	⚠️
24054209201	ACE Editor Session	main	❌ early-exit	0	0	⚠️
24054209203	Mergefest	main	❌ early-exit	0	0	⚠️
24054209205	Plan Command	main	❌ early-exit	0	0	⚠️
22 more	(various .lock.yml)	—	❌ early-exit	0	0	⚠️

All 30 runs: 28 action_required (environment gate), 2 success (agent skipped due to command mismatch). Zero actual agent executions.

---

Recommendations

1. Monitor environment approval latency: The current environment protection gate is blocking all agent runs. Consider tracking how long runs wait in action_required state and alerting if agent workflows go more than N hours without any approved execution — sustained gaps create blind spots in the DIFC audit trail.

2. Add DIFC smoke-test workflow: Consider a scheduled workflow (e.g., nightly) that exercises the MCP Gateway with a known-safe prompt and validates that DIFC events appear in rpc-messages.jsonl. This would provide a baseline heartbeat independent of human-approval gates.

3. Confirm environment protection is intentional at this frequency: If the intent is that agents should regularly execute (e.g., when Copilot SWE agent opens PRs), verify that the environment protection reviewers are being notified and approving in a timely manner. The historical pattern (zero completed agent runs observed across ~2 weeks of Scout history) may indicate the approval workflow is stalled.

4. No direct API bypass detected: No network firewall blocks or direct api.github.com / external AI service bypass attempts were found — consistent with zero actual agent execution.

Generated by Integrity Filtering Audit · ● 4.9M · ◷

• expires on Apr 13, 2026, 10:44 PM UTC