You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Task Could Not Be Completed: The agent reported that the task could not be performed due to an infrastructure or tool failure.
Reasons:
The mandatory create_discussion safe-output for this run was already consumed by an exploratory CLI test call (title test-title-check, placeholder body) made while learning the tool syntax, before the real audit report could be submitted. create_discussion has a limit of 1 per run for this workflow, so the real findings below could not be posted as a discussion in this run.
REAL SAFE OUTPUT HEALTH AUDIT FINDINGS FOR 2026-08-08 (this is the actual report that should have been posted as a GitHub discussion):
EXECUTIVE SUMMARY
Period: Last 24 hours (2026-08-07 ~21:00 UTC to 2026-08-08 ~05:00 UTC)
Runs Analyzed: 202
Distinct Workflows Active: 72
Safe Output Jobs Executed: 202 (one safe_outputs job per run)
Safe Output Jobs Failed: 6
Overall Safe Output Success Rate: 97.0%
Error Clusters Identified: 3
Architecture note: this repo does not run separate jobs per safe-output type (no create_issue job, add_comment job, etc). Every run has a single consolidated safe_outputs job whose Process Safe Outputs step handles all configured output types (create_issue, create_discussion, add_comment, update_pull_request, push_to_pull_request_branch, send_slack_message, create_check_run, set_issue_type, comment_memory, submit_pull_request_review, upload_artifact, etc) in one step. A failure processing any single item type currently fails the entire job for that run, and there is no per-type success/failure breakdown recorded anywhere in the run artifacts.
Runs: 31220056968, 31222502383, 31231536662 (PRs #51133 and #51174), triggered via the smoke slash command on PRs.
Observation: In every run the agent job succeeded and generated 8-10 safe-output items across ~9 types (create_discussion, create_issue, add_comment x2, send_slack_message, create_check_run, set_issue_type, comment_memory, submit_pull_request_review, upload_artifact). The consolidated Process Safe Outputs step still failed every single time.
Root Cause: Undetermined - no raw step log text was accessible to this audit (see Tooling Gap below). Leading suspects: (a) send_slack_message failing due to a missing/invalid SLACK_WEBHOOK_URL secret when triggered from a PR context, or (b) create_check_run failing due to insufficient checks:write permission on PR-triggered smoke runs.
Impact: Medium. Smoke-test workflow (not user-facing), but a 100% failure rate across all 3 samples suggests a deterministic, reproducible bug rather than a flaky one-off.
ERROR CLUSTER 2: Smoke Copilot - AOAI (Entra) - 2/6 runs failed, correlated with but not fully explained by empty output
Runs: 31222505053 (PR #51133), 31231538240 (PR #51174).
Sample Error Signature: Both failing runs show the agent job completing successfully but producing 0 safe-output items (total_items: 0), yet the safe_outputs job conclusion is failure.
Root Cause: Undetermined and puzzling - a third run of the same workflow (31220058346) also produced 0 items but SUCCEEDED. So empty output alone does not explain the failure; something else differs between the passing and failing zero-item runs.
Impact: Medium.
ERROR CLUSTER 3: PR Sous Chef - isolated single failure in a production workflow
Run: 31230238662, scheduled trigger, pi engine.
Sample Error Signature: Agent produced 5 items (update_pull_request x2, add_comment x2, create_issue x1); the safe_outputs job failed processing them. Exact failing item/error unknown.
Root Cause: Undetermined (same tooling gap). Given PR Sous Chef runs every 15 minutes and only 1 failure was observed in the 24h window, this looks like an isolated/rare event rather than a systemic regression - but it is the only failure in this audit that affects a real production workflow rather than a smoke test.
Impact: Low-frequency but highest real-world relevance of the three clusters, since it is the only non-test workflow affected.
ROOT CAUSE ANALYSIS
API-Related Issues: None identified - no evidence of GitHub API rate-limiting or auth failures in any of the 6 failing runs.
Data Validation / Parsing Issues: Cannot be confirmed without raw logs.
Permission Issues: Possible for Cluster 1 (create_check_run may lack checks:write on PR-triggered smoke runs) - plausible but unconfirmed.
Tooling Gap (most important finding): None of the available tools - the gh-aw MCP logs/audit tools, or the read-only GitHub MCP server (no workflow-run/job-log-fetching method exposed) - can retrieve the raw stdout/stderr of a failed Actions step. The cached run artifacts (run_summary.json, summary.json, aw_info.json) only carry pass/fail status per job/step and aggregate item counts, never the actual error message. All 6 failures could be located but not diagnosed.
RECOMMENDATIONS
Critical Issues: None - no user-facing production breakage detected.
Bug Fixes Required:
Investigate and fix the 100%-reproducible Smoke Copilot safe-outputs failure. File: .github/workflows/smoke-copilot.lock.yml. Pull raw Actions logs for run 31231536662 to identify the specific failing item/step, then fix the underlying handler or workflow permissions/secrets.
Investigate the Smoke Copilot - AOAI (Entra) empty-output inconsistency. File: .github/workflows/smoke-copilot-aoai-entra.lock.yml. Compare failing run 31222505053 against succeeding zero-item run 31220058346 to isolate the differentiating condition.
Configuration Changes:
Review checks:write permission and Slack webhook availability for PR-triggered smoke workflows - confirm secrets/permissions are actually available in that trigger context, or skip those output types when running from a PR-triggered smoke invocation.
Process Improvements:
Record per-item-type success/failure in run artifacts - emit a structured safe_outputs_errors.json (or extend run_summary.json) capturing which specific item(s) failed and their error message, independent of whether other items in the same run succeeded. This would have let this audit pinpoint exact root causes for all 6 failures instead of stopping at job failed.
Expose raw Actions job/step logs to future audits - add a GitHub MCP tool (or gh-aw MCP tool) for fetching job/step logs by run ID. This is the single biggest blocker to root-causing safe-output failures in future audits, and is also why this current audit-tooling failure happened: no tool could be found to safely compose a large discussion body without hitting shell-escaping and safe-output quota limits either.
WORK ITEM PLANS
Work Item 1: Fix Smoke Copilot safe-outputs 100% failure rate. Priority High. Pull raw logs for run 31231536662; identify failing handler; patch. Estimated effort Small-Medium.
Work Item 2: Diagnose Smoke Copilot - AOAI (Entra) intermittent empty-output failure. Priority Medium. Compare run 31222505053 (fail) vs 31220058346 (success) job logs step-by-step. Estimated effort Small.
Work Item 3: Add per-item-type error capture to safe-outputs processing. Priority Medium. Extend the safe-outputs processing script to catch and record per-item errors instead of failing the whole step. Estimated effort Medium.
HISTORICAL CONTEXT
No prior Safe Output Health audit was found in cache memory (path was empty before this run). This is the first baseline audit. This baseline has been saved to cache memory at /tmp/gh-aw/cache-memory/safe-output-health/2026-08-08.json, index.json, and error-patterns.json for future comparison.
METRICS AND KPIS
Overall Safe Output Success Rate: 97.0% (196/202). Most Reliable Job Type: safe_outputs job across all non-smoke-test workflows, 100% success rate. Most Problematic Workflow: Smoke Copilot, 0% success rate (3/3 failed) in this window. Average Time to Failure: not meaningfully computable, failing steps ran 24-54 seconds before reporting failure, in line with normal step duration.
NEXT STEPS
Pull raw Actions logs for runs 31231536662 (Smoke Copilot) and 31222505053 (Smoke Copilot - AOAI Entra) via an authenticated session to get the actual error text
File/track Work Items 1-3 above
Add raw-log or per-item-error capture to close the tooling gap that limited this audit
Re-run this audit after fixes land to confirm Smoke Copilot returns to a passing state
This is a structured incompletion signal (report_incomplete), not a real task outcome. Any other safe outputs emitted alongside this signal (e.g., comments) describe the failure state, not a completed review or action.
Action Required
Assign this issue to an agent to debug and fix the issue.
Debug with any coding agent
Use this prompt with any coding agent (GitHub Copilot, Claude, Gemini, etc.):
Debug the agentic workflow failure using https://raw.githubusercontent.com/github/gh-aw/main/debug.md
The failed workflow run is at https://github.com/github/gh-aw/actions/runs/31239625370
Manually invoke the agent
Debug this workflow failure using your favorite Agent CLI and the agentic-workflows prompt.
Workflow Failure
Workflow: Safe Output Health Monitor
Branch: main
Run: https://github.com/github/gh-aw/actions/runs/31239625370
Warning
Task Could Not Be Completed: The agent reported that the task could not be performed due to an infrastructure or tool failure.
Reasons:
The mandatory create_discussion safe-output for this run was already consumed by an exploratory CLI test call (title test-title-check, placeholder body) made while learning the tool syntax, before the real audit report could be submitted. create_discussion has a limit of 1 per run for this workflow, so the real findings below could not be posted as a discussion in this run.
REAL SAFE OUTPUT HEALTH AUDIT FINDINGS FOR 2026-08-08 (this is the actual report that should have been posted as a GitHub discussion):
EXECUTIVE SUMMARY
Architecture note: this repo does not run separate jobs per safe-output type (no create_issue job, add_comment job, etc). Every run has a single consolidated safe_outputs job whose Process Safe Outputs step handles all configured output types (create_issue, create_discussion, add_comment, update_pull_request, push_to_pull_request_branch, send_slack_message, create_check_run, set_issue_type, comment_memory, submit_pull_request_review, upload_artifact, etc) in one step. A failure processing any single item type currently fails the entire job for that run, and there is no per-type success/failure breakdown recorded anywhere in the run artifacts.
SAFE OUTPUT JOB STATISTICS
safe_outputs job (all types combined): 202 executions, 6 failures, 97.0% success rate
ERROR CLUSTER 1: Smoke Copilot - 100% failure rate (3/3 runs)
Runs: 31220056968, 31222502383, 31231536662 (PRs #51133 and #51174), triggered via the smoke slash command on PRs.
Observation: In every run the agent job succeeded and generated 8-10 safe-output items across ~9 types (create_discussion, create_issue, add_comment x2, send_slack_message, create_check_run, set_issue_type, comment_memory, submit_pull_request_review, upload_artifact). The consolidated Process Safe Outputs step still failed every single time.
Root Cause: Undetermined - no raw step log text was accessible to this audit (see Tooling Gap below). Leading suspects: (a) send_slack_message failing due to a missing/invalid SLACK_WEBHOOK_URL secret when triggered from a PR context, or (b) create_check_run failing due to insufficient checks:write permission on PR-triggered smoke runs.
Impact: Medium. Smoke-test workflow (not user-facing), but a 100% failure rate across all 3 samples suggests a deterministic, reproducible bug rather than a flaky one-off.
ERROR CLUSTER 2: Smoke Copilot - AOAI (Entra) - 2/6 runs failed, correlated with but not fully explained by empty output
Runs: 31222505053 (PR #51133), 31231538240 (PR #51174).
Sample Error Signature: Both failing runs show the agent job completing successfully but producing 0 safe-output items (total_items: 0), yet the safe_outputs job conclusion is failure.
Root Cause: Undetermined and puzzling - a third run of the same workflow (31220058346) also produced 0 items but SUCCEEDED. So empty output alone does not explain the failure; something else differs between the passing and failing zero-item runs.
Impact: Medium.
ERROR CLUSTER 3: PR Sous Chef - isolated single failure in a production workflow
Run: 31230238662, scheduled trigger, pi engine.
Sample Error Signature: Agent produced 5 items (update_pull_request x2, add_comment x2, create_issue x1); the safe_outputs job failed processing them. Exact failing item/error unknown.
Root Cause: Undetermined (same tooling gap). Given PR Sous Chef runs every 15 minutes and only 1 failure was observed in the 24h window, this looks like an isolated/rare event rather than a systemic regression - but it is the only failure in this audit that affects a real production workflow rather than a smoke test.
Impact: Low-frequency but highest real-world relevance of the three clusters, since it is the only non-test workflow affected.
ROOT CAUSE ANALYSIS
API-Related Issues: None identified - no evidence of GitHub API rate-limiting or auth failures in any of the 6 failing runs.
Data Validation / Parsing Issues: Cannot be confirmed without raw logs.
Permission Issues: Possible for Cluster 1 (create_check_run may lack checks:write on PR-triggered smoke runs) - plausible but unconfirmed.
Tooling Gap (most important finding): None of the available tools - the gh-aw MCP logs/audit tools, or the read-only GitHub MCP server (no workflow-run/job-log-fetching method exposed) - can retrieve the raw stdout/stderr of a failed Actions step. The cached run artifacts (run_summary.json, summary.json, aw_info.json) only carry pass/fail status per job/step and aggregate item counts, never the actual error message. All 6 failures could be located but not diagnosed.
RECOMMENDATIONS
Critical Issues: None - no user-facing production breakage detected.
Bug Fixes Required:
Configuration Changes:
Process Improvements:
WORK ITEM PLANS
Work Item 1: Fix Smoke Copilot safe-outputs 100% failure rate. Priority High. Pull raw logs for run 31231536662; identify failing handler; patch. Estimated effort Small-Medium.
Work Item 2: Diagnose Smoke Copilot - AOAI (Entra) intermittent empty-output failure. Priority Medium. Compare run 31222505053 (fail) vs 31220058346 (success) job logs step-by-step. Estimated effort Small.
Work Item 3: Add per-item-type error capture to safe-outputs processing. Priority Medium. Extend the safe-outputs processing script to catch and record per-item errors instead of failing the whole step. Estimated effort Medium.
HISTORICAL CONTEXT
No prior Safe Output Health audit was found in cache memory (path was empty before this run). This is the first baseline audit. This baseline has been saved to cache memory at /tmp/gh-aw/cache-memory/safe-output-health/2026-08-08.json, index.json, and error-patterns.json for future comparison.
METRICS AND KPIS
Overall Safe Output Success Rate: 97.0% (196/202). Most Reliable Job Type: safe_outputs job across all non-smoke-test workflows, 100% success rate. Most Problematic Workflow: Smoke Copilot, 0% success rate (3/3 failed) in this window. Average Time to Failure: not meaningfully computable, failing steps ran 24-54 seconds before reporting failure, in line with normal step duration.
NEXT STEPS
REFERENCES:
This is a structured incompletion signal (
report_incomplete), not a real task outcome. Any other safe outputs emitted alongside this signal (e.g., comments) describe the failure state, not a completed review or action.Action Required
Assign this issue to an agent to debug and fix the issue.
Debug with any coding agent
Use this prompt with any coding agent (GitHub Copilot, Claude, Gemini, etc.):
Manually invoke the agent
Debug this workflow failure using your favorite Agent CLI and the
agentic-workflowsprompt.agentic-workflowsskill from.github/skills/agentic-workflows/SKILL.mdor https://github.com/github/gh-aw/blob/main/.github/skills/agentic-workflows/SKILL.mddebug the agentic workflow safe-output-health failure in https://github.com/github/gh-aw/actions/runs/31239625370Tip
Stop reporting this workflow as a failure
To stop a workflow from creating failure issues, set
report-failure-as-issue: falsein its frontmatter: