[cli-tools-test] audit tool: failure_analysis does not surface actual errors from agent logs + opaque error on invalid run IDs

[github-actions]

Summary

Daily exploratory testing (run #22531620317) identified two bugs in the audit MCP tool related to error reporting quality.

---

Bug 1: failure_analysis.error_summary always reports "No specific errors identified" for failed runs

Steps to Reproduce

1. Audit a failed workflow run (e.g., AI Moderator run #22531456203):

   audit(run_id_or_url: "22531456203")
   

Expected Behavior

failure_analysis.error_summary should surface the actual error message extracted from agent-stdio.log.

Actual Behavior

"failure_analysis": {
  "primary_failure": "failure",
  "failed_jobs": ["agent"],
  "error_summary": "No specific errors identified"
}
```

Yet the actual error is clearly present in `agent-stdio.log`:
```
ERROR: stream disconnected before completion: This user's access to gpt-5.3-codex has been
temporarily limited for potentially suspicious activity related to cybersecurity.
[WARN] Command completed with exit code: 1
```

The `run_summary.json` also has an empty errors array `[]` despite the agent-stdio.log containing a clear `ERROR:` line.

### Impact

- **Severity**: High
- **Frequency**: Always (reproducible on both AI Moderator run 22531456203 and Terminal Stylist run 22506113327)
- The audit tool is the primary debugging interface for failed runs; returning "No specific errors identified" makes it useless for diagnosing failures.

---

## Bug 2: Invalid/non-existent run ID returns opaque `exit status 1`

### Steps to Reproduce

```
audit(run_id_or_url: "99999999999")
```

### Expected Behavior

A clear error message like: `"Run ID 99999999999 not found in github/gh-aw"` with guidance to check the run ID.

### Actual Behavior

```
McpError: MCP error -32603: calling "tools/call": failed to audit workflow run: exit status 1

No indication that the run doesn't exist. The user has no actionable information.

Impact

• Severity: Medium
• Frequency: Always
• Degrades developer experience when there's a typo in the run ID

---

Environment

• Repository: github/gh-aw
• Run ID: 22531620317
• Date: 2026-02-28
• Test runs audited: 22531177735 (success), 22531456203 (failure), 22506113327 (failure)

---

Additional Observations (non-blocking)

• compile with actionlint: true fails with "docker images not ready" instead of a message explaining actionlint requires Docker — the error is misleading when Docker isn't available in the environment.
• workflow_path field is empty in unfiltered logs results but populated when using the engine filter — minor inconsistency in response data.

Generated by Daily CLI Tools Exploratory Tester

• expires on Mar 7, 2026, 11:54 PM UTC

[pelikhan]

@copilot remove failure analysis , let the agents do that