[aw-failures] [P0] safeoutputs MCP gateway registered 0 tools — fleet-wide failure, fix merged in #51870, verify recovery

[github-actions]

Executive summary

Ship the fix — done. PR #51870 (merged 2026-08-11T01:03 UTC) patches the gh-aw-mcpg v0.4.9 mount-policy rejection that was killing the safeoutputs MCP gateway fleet-wide. Every one of the 50 failed runs in this 6h window carries the identical annotation safeoutputs MCP gateway registered 0 tools during setup. Zero of today's failures were covered by any open agentic-workflows issue. One P0 cluster, no P1/P2 gaps. What's left is verification, not more code — confirm the next scheduled run of each affected workflow actually recovers post-merge (sub-issue below).

Failure cluster table

Sev	Cluster	Runs (confirmed / sampled)	Representative	Comparator	Status
P0	safeoutputs MCP gateway registers 0 tools at startup → every workflow using safe-outputs fails at "Ingest agent output" (or earlier, at "Execute GitHub Copilot CLI")	8/8 sampled runs match; signature explains all 50 failed_run_ids in window	§31423026452 (Code Scanning Fixer)	§31344892463 (Documentation Unbloat, last success, 2026-08-10)	Fixed upstream (#51870), unverified in prod

Evidence

Fix the shared symptom, not per-workflow logic — the annotation is identical across every workflow sampled:

##[error]safeoutputs MCP gateway registered 0 tools during setup; the agent could not emit any safe outputs. This is a gateway infrastructure failure, not a normal no-op. Check the MCP gateway startup logs for ECONNRESET errors or delayed backend registration and re-run the workflow.

Sampled and confirmed identical across 8 distinct workflows, spread across the full 6h window:

Workflow	Run	Failed step(s)
PR Sous Chef	§31447900905	Ingest agent output
Issue Monster	§31447732072	Ingest agent output
Documentation Unbloat	§31446561055	Ingest agent output
Auto-Triage Issues	§31446362728	Ingest agent output
PR Triage Agent	§31446329055	Ingest agent output
Code Scanning Fixer	§31423026452	Execute GitHub Copilot CLI + Ingest agent output
Detection Analysis Report	§31441806294	Ingest agent output
Daily Observability Report for AWF Firewall and MCP Gateway	§31443961838	Ingest agent output

Root cause, from #51870 directly: gh-aw-mcpg v0.4.9 introduced a trusted host-path mount policy that only permits read-write access to specific roots. The compiler mounts $GITHUB_WORKSPACE read-write into the built-in safeoutputs server, so the gateway rejected it at startup and registered 0 tools — breaking every safe-output call fleet-wide, not a bug in any individual workflow.

audit-diff comparison was skipped — root cause is already definitive from annotations + the merging PR's own description, and gh run view --log-failed returns error connecting to api.github.com in this sandbox (same known log-storage restriction noted in #51791), so a deeper log pull wouldn't add signal here.

All 50 failed run IDs in window (2026-08-10T19:13–2026-08-11T01:00 UTC)

31447900905, 31447732072, 31446561055, 31446362728, 31446329055, 31446300276, 31446278001, 31444637406, 31444636467, 31444594175, 31443961838, 31443748964, 31443224684, 31443174861, 31442005150, 31441806294, 31441778852, 31440996820, 31440825781, 31439994130, 31439145860, 31437852016, 31437730226, 31436817805, 31436460859, 31435941725, 31435771037, 31433644029, 31433607305, 31433561146, 31433495358, 31432268253, 31432075060, 31431466514, 31431380654, 31431216362, 31431178194, 31430911362, 31430908774, 31430906259, 31428898075, 31428667503, 31428542931, 31427423614, 31425441705, 31425440399, 31424342221, 31424274298, 31423110186, 31423026452

Existing issue correlation

No open agentic-workflows issue tracked this signature — close the gap now, don't wait for it to recur. Checked issue search for "registered 0 tools" and "MCP gateway": only unrelated matches (#39692 network-host mount flag, #51799 single unrelated workflow failure). Confirmed not the same as the other open P0/P1 trackers, which stay untouched because none of today's evidence bears on them:

• [aw-failures] [P0] Fix Copilot CLI harness segfault (exit 139) killing Agent Performance Analyzer #51789 — Copilot CLI harness segfault (exit 139)
• [aw-failures] [P0] Step Name Alignment sandbox blocks its own input file — 100% failure #51547 — Step Name Alignment sandbox file-lockout
• [aw-failures] [P1] Fix Copilot billing 403 masking Daily Max Ai Credits Test — 5+ consecutive daily failures #51589 — Copilot billing 403
• [aw-failures] Fix Copilot CLI Task-tool subagent calls — 100% failure with No model available #51033 — Copilot CLI Task-tool "No model available"

No existing issue closed this pass — nothing in scope today was fixed, stale, or contradicted by fresh evidence.

Fix roadmap

• P0 — done, verify next: Root cause fixed in Fix MCP gateway mount policy rejection for safeoutputs backend server #51870 (merged 01:03 UTC): added MCP_GATEWAY_ALLOWED_MOUNT_ROOTS allowlist, recompiled all .lock.yml. Track verification in the sub-issue below — do not close until a post-merge success is confirmed.
• P1: none new this pass.
• P2: none new this pass.

Sub-issues created

• [aw-failures] [P0] Verify safeoutputs MCP gateway recovery after #51870 #51897 — Verify safeoutputs MCP gateway recovery after Fix MCP gateway mount policy rejection for safeoutputs backend server #51870

[github-actions]

🎉 Automatically closed by Sub-Issue Closer

All sub-issues have been completed. This parent issue is now closed automatically.

Sub-issues status: 1/1 closed (100%)

Generated by 🔧 Sub-Issue Closer · auto · 51.3 AIC · ⌖ 2.08 AIC · ⊞ 7.8K · ◷