[health] Metrics Collector — 2 consecutive failures, metrics infrastructure degraded #23344

Overview

The Metrics Collector workflow has failed on 2 consecutive runs:

This is a metrics infrastructure issue — the Metrics Collector feeds shared metrics data to:

• Workflow Health Manager (/tmp/gh-aw/repo-memory/default/metrics/latest.json)
• Agent Performance Analyzer
• Campaign Manager

Without working metrics, trend analysis across all meta-orchestrators is degraded.

Root Cause Analysis

Run #107 failure (primary failing job: push_repo_memory):

Working in repository: /home/runner/work/gh-aw/gh-aw
Checking out branch: memory/meta-orchestrators...
File glob filter enabled: metrics/**
##[warning]Skipping file that does not match allowed patterns: agent-performance-latest.md
  Pattern: "metrics/**" -> ❌ NO MATCH
Scan complete: Found 0 file(s) to copy
No files to copy from artifact
##[error]Can't find 'action.yml', 'action.yaml' or 'Dockerfile' under
'/home/runner/work/gh-aw/gh-aw/actions/setup'.

Two issues:

1. Wrong file location: The agent is writing agent-performance-latest.md to the artifact root instead of under the metrics/ subdirectory. The push_repo_memory job has a metrics/** glob filter that doesn't match root-level files. Result: Zero metrics saved.
2. Post Setup cleanup failure: Same systemic bug as PR Triage Agent ([aw] PR Triage Agent failed #23151) — after checking out memory/meta-orchestrators branch, actions/setup is not found.

Run #106 failure appears to be the same pattern.

Impact

• Latest metrics file (metrics/latest.json) is now stale (last written by run Remove frontmatter concurrency_policy support, implement code-based concurrency rules #105 on 2026-03-25)
• Historical daily metrics are not being appended
• WHM and Agent Performance Analyzer are operating without fresh data
• This affects accuracy of health scores and trend analysis

Possible Fixes

Issue 1 (agent writing wrong path):
The Metrics Collector prompt instructs the agent to write to /tmp/gh-aw/repo-memory/default/metrics/latest.json. If the agent is writing to /tmp/gh-aw/repo-memory/default/agent-performance-latest.md instead, either:

• The agent misunderstood the instructions
• The agent is reading from another memory source and re-writing it
• The glob filter in the push_repo_memory job configuration is too restrictive

Issue 2 (Post Setup cleanup): Shared systemic bug tracked in #23151.

Action Required

1. Investigate run MCP Network Permissions Test Results - Domain Restrictions Validation #107 agent job logs to understand why agent-performance-latest.md was created instead of metrics/latest.json
2. Consider updating the push_repo_memory file-glob configuration to include *.md or fix the agent prompt
3. Verify the metrics format after fixing — downstream consumers depend on the JSON structure

Generated by Workflow Health Manager · run #284

tools:
  github:
    min-integrity: approved  # merged | approved | unapproved | none

Generated by Workflow Health Manager - Meta-Orchestrator · ◷

• expires on Mar 29, 2026, 12:04 PM UTC