[agent-job-health] MCP Gateway readiness timeout: 5 PR-reviewer workflows fail simultaneously at Start MCP Gateway

Summary

During the 2026-10-21 Agent Job Health Monitor run (window: 2026-10-20 23:18 UTC → 2026-10-21 23:18 UTC), a novel failure cluster was identified: 5 PR-reviewer workflows failed simultaneously at the Start MCP Gateway step, all triggered by the same pull_request event.

This is distinct from the already-known/accepted "Checkout PR branch" obsolete-PR race (where a PR's branch is deleted after merge before checkout runs) — this cluster fails at gateway startup instead, after the checkout step itself succeeded.

Cluster details

Count: 5 runs across 5 distinct workflows: Design Decision Gate 🏗️, Matt Pocock Skills Reviewer, PR Code Quality Reviewer, Ponytail Reviewer, Test Quality Sentinel
Failing step: Start MCP Gateway
Engines involved: pi, codex, copilot (mixed — rules out an engine-specific cause)
All 5 fired on the same pull_request event at 2026-10-21T01:50:41Z
Representative run: https://github.com/github/gh-aw/actions/runs/39529635821 (Ponytail Reviewer)
Sibling runs: 39529635814 (Design Decision Gate), 39529635864 (PR Code Quality Reviewer), 39529636024 (Matt Pocock Skills Reviewer), 39529636084 (Test Quality Sentinel)

Error

[info] ⏱️  TIMING: Safe outputs readiness wait took 137170ms
[error] Error: MCP gateway safeoutputs readiness failed after 137 attempt(s): no tools registered or backend unavailable. The agent will not start. Check MCP gateway startup logs for connection resets (ECONNRESET) or delayed backend registration. Last probe: insufficient startup budget for a readiness probe (requires at least 750ms).
[error] FATAL: MCP gateway safeoutputs readiness failed after 137 attempt(s): no tools registered or backend unavailable. The agent will not start. ...
##[error]Process completed with exit code 1.

All 5 jobs waited ~137 seconds (137 readiness-probe attempts) before giving up, then failed the job outright rather than retrying with a longer budget or backing off.

Suspected root cause

These 5 PR-reviewer workflows all trigger on the same pull_request event and start their MCP gateways concurrently on separate runners for the same commit. In this instance, all 5 timed out waiting for the safeoutputs MCP gateway backend to register its tools — consistent with resource contention or a shared-backend hiccup when 5 gateway instances start at (or near) the same moment for the same commit/PR.

Suggested next steps

Check MCP gateway startup/backend logs around 2026-10-21T01:48–01:51 UTC for connection resets (ECONNRESET) or delayed tool registration that would explain a shared-backend bottleneck.
Consider whether the readiness probe's startup budget (currently requiring ≥750ms per probe, 137 attempts ≈ 137s total) should back off or extend further under concurrent-start load, or whether concurrent gateway starts for the same commit should be staggered/rate-limited.
A related but distinct "Start MCP Gateway step failure (codex)" was noted in this report's own trend history on 2026-10-12 (2 runs / 2 workflows, untracked) — worth checking whether that is the same underlying contention issue recurring, or unrelated.

Scope note

This is reported by the Agent Job Health Monitor, which owns fleet-wide agent-job failure rate tracking (see .github/workflows/agent-job-health.md). The sibling "Checkout PR branch" failure affecting the same 5 workflows (30 runs in the same 24h window) is a separate, already-understood obsolete-PR race and is intentionally not covered by this issue.

Generated by 🩺 Agent Job Health Monitor · claude · agent · 797.1 AIC · ⌖ 0.94 AIC · ⊞ 6.9K · ◷

expires on Oct 28, 2026, 3:55 PM UTC-08:00

🍪 Issue Monster selected this for Copilot

I've identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.