Kill the Copilot CLI once the agent has emitted its final safe output — it currently hangs until the step timeout, turning healthy no-op runs into red failures.
On main, the Issue Triage Agent completed its logic correctly — it emitted a valid noop ("No unlabeled open issues found ... no triage action needed") — but the Execute GitHub Copilot CLI process then hung and was killed by the 5-minute step timeout, marking a fully successful run as failed.
Affected workflow / runs
- Issue Triage Agent (
.github/workflows/issue-triage.lock.yml / Copilot engine, branch main)
- Failed run: §28954445421
Evidence
15:26:21 [safeoutputs] recv tools/call name=noop arguments.reason="No unlabeled open issues found ... No triage action was needed."
15:31:22 [safeoutputs] Final processed config ... (backend still alive, ~5 min later)
15:31:34 ##[error]The action 'Execute GitHub Copilot CLI' has timed out after 5 minutes.
The agent's work finished at 15:26:21; the process did not exit and was force-timed-out ~5 minutes later. No sandbox-firewall permission error is present in this run (distinguishing it from #44242).
Probable root cause
The Copilot CLI harness does not terminate after the agent produces its terminal safe output. The process idles (open MCP/backend connections) until the workflow-level GH_AW_TIMEOUT_MINUTES fires, which the runner records as a failure.
Proposed remediation
- After the agent emits a terminal result/safe output, have the harness send an explicit shutdown to the Copilot CLI (and close MCP client connections) so the process exits with code 0.
- Add a post-result watchdog: if no further activity within N seconds of the final result, terminate the CLI cleanly rather than waiting for the hard step timeout.
- Ensure a clean post-result exit is scored as success, not timeout-failure.
Success criteria
- A run whose agent emits
noop/final output exits promptly (well under the step timeout) and is marked success.
- No
Execute GitHub Copilot CLI ... timed out failures on runs that already produced a terminal safe output.
Note: feature-branch smoke-test timeouts this window (Smoke Copilot AOAI 28966796489, Smoke Claude 28966636432) are a different mechanism — post-run "Rootless artifact permission repair" retries on root-owned dirs, tracked under #44242.
Analyzed run: 28954445421.
Generated by 🔍 [aw] Failure Investigator (6h) · 211.3 AIC · ⌖ 34.6 AIC · ⊞ 5.2K · ◷
Kill the Copilot CLI once the agent has emitted its final safe output — it currently hangs until the step timeout, turning healthy no-op runs into red failures.
On
main, the Issue Triage Agent completed its logic correctly — it emitted a validnoop("No unlabeled open issues found ... no triage action needed") — but theExecute GitHub Copilot CLIprocess then hung and was killed by the 5-minute step timeout, marking a fully successful run as failed.Affected workflow / runs
.github/workflows/issue-triage.lock.yml/ Copilot engine, branchmain)Evidence
The agent's work finished at 15:26:21; the process did not exit and was force-timed-out ~5 minutes later. No sandbox-firewall permission error is present in this run (distinguishing it from #44242).
Probable root cause
The Copilot CLI harness does not terminate after the agent produces its terminal safe output. The process idles (open MCP/backend connections) until the workflow-level
GH_AW_TIMEOUT_MINUTESfires, which the runner records as a failure.Proposed remediation
Success criteria
noop/final output exits promptly (well under the step timeout) and is marked success.Execute GitHub Copilot CLI ... timed outfailures on runs that already produced a terminal safe output.Note: feature-branch smoke-test timeouts this window (Smoke Copilot AOAI 28966796489, Smoke Claude 28966636432) are a different mechanism — post-run "Rootless artifact permission repair" retries on root-owned dirs, tracked under #44242.
Analyzed run: 28954445421.