Skip to content

[workflow-health] Metrics Collector: 0/0 success-fail anomaly masks skip-dominated workflows (squad, squad-implement-worker, dep #54724

Description

@github-actions

Problem

Four workflows show a 0-successful / 0-failed anomaly in metrics/latest.json (2026-08-22 collection), each with hundreds of "runs" but ~2 second average duration and a success_rate of 0.0:

Workflow Total runs Successful Failed Avg duration
squad 423 0 0 2.1s
squad-implement-worker 106 0 0 2.2s
deployment-incident-monitor 93 0 0 2.2s
workflow-generator 8 0 0 2.4s

None of these are genuine failures — a 2-second "run" with no successful/failed conclusion strongly suggests the runs are being recorded as skipped (e.g. slash-command/bot-gated triggers like squad's issue_comment/slash_command config that exit immediately when the trigger condition doesn't match) but the metrics collector's GitHub-API-fallback path (used because agentic-workflows logs timed out on every attempt this window, per latest.json's collection_note) isn't classifying skipped conclusions separately from success/failed, producing a 0.0 success rate that looks identical to total failure.

This degrades the reliability of automated health scoring (Workflow Health Manager, Agent Performance Analyzer) for these workflows and risks false P0/P1 escalation.

Evidence

  • /tmp/gh-aw/repo-memory/default/metrics/latest.jsoncollection_status: "complete", data_source: "github_api_fallback", note explicitly states the primary logs-based collection timed out on every attempt (count=80/20/5/3) and the API fallback pagination doesn't recover per-workflow success/failure/skipped breakdown for these four.
  • .github/workflows/squad.md — triggers on slash_command (issues, issue_comment, pull_request_review_comment) plus bots: ["github-actions[bot]"] gate; most comment events won't match, producing near-instant skips.
  • .github/workflows/deployment-incident-monitor.md, squad-implement-worker.md, workflow-generator.md — similarly event/dispatch-gated, consistent with high skip volume.

Suggested Fix

  1. In the Metrics Collector workflow, when using the GitHub API fallback path, explicitly classify conclusion: "skipped" and conclusion: "action_required" runs separately from success/failure in the per-workflow rollup (instead of leaving successful/failed both at 0, which reads as 100% failure).
  2. Surface a skipped count field per workflow in metrics/latest.json so downstream consumers (Workflow Health Manager, Agent Performance Analyzer) can exclude skip-dominated workflows from failure-rate scoring.
  3. Investigate why the primary agentic-workflows logs collection path is timing out on every attempt (60s context deadline exceeded) — restoring it would give richer duration/cost data for all workflows, not just these four.

Priority

P2 — not a functional workflow failure, but a data-quality gap in the shared metrics pipeline that risks false alarms in downstream health/performance reports.

Generated by 🏥 Workflow Health Manager - Meta-Orchestrator · auto · 79.7 AIC · ⌖ 10.8 AIC · ⊞ 12K ·

  • expires on Aug 22, 2026, 8:47 PM UTC-08:00

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

cookieIssue Monster Loves Cookies!

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions