[audit-workflows] Agentic Workflow Audit — 2026-07-28: log collection outage (0 runs retrieved, 22-day visibility gap)

[github-actions[bot]]

Agentic Workflow Audit — 2026-07-28

Collection failed — this audit could not retrieve any fresh run data. Everything below is a transparent incident report plus the last known-good trend data (through 2026-07-06), not a live 24h audit.

What happened

The agenticworkflows MCP logs tool failed on every one of 5 attempts over ~25 minutes of wall time:

#	Query	Result
1	start_date=-1d (default)	Timed out ~300s → "No runs found. Timeout reached before any runs could be downloaded."
2	start_date=-1d, timeout=280	Completed in ~319s, but 0 runs → "No runs found matching the specified criteria."
3	start_date=-3d, count=10	Same cached empty result, 0 runs
4	workflow_name=smoke-ci (kebab id), start_date=-6h	Fast (1.1s) hard error: could not find any workflows named Smoke CI
5	workflow_name="Smoke CI" (title case)	Same error, 3.7s

The repo is confirmed active during this window — recent commits as late as 2026-07-28T21:46Z were visible via the GitHub MCP server — so this is a tooling failure, not a quiet repository. The status tool (reads locally-compiled workflow metadata) worked fine and listed smoke-ci normally; only the live log-download path is broken. The fast, deterministic failure on workflow-name lookup (#4/#5) is the most actionable lead: something in the run-listing path can't resolve a workflow that the metadata path sees fine.

This is the first total collection failure in this audit's history. The closest precedent was a partial degradation on 2026-07-02 ("count-limited 20-run slice, bridge 60s cap"). Combined with a pre-existing 22-day gap in audit history (last successful audit: 2026-07-06), the fleet has now gone 22+ days without verified visibility.

Filed as a new CRITICAL known-issue (agenticworkflows-logs-tool-total-outage) and P0 recommendation in repo memory, with reproduction details, exact error strings, and a proposed fix path (canary health-check + investigate workflow-name resolution divergence between the metadata and log-download paths).

Trend charts (historical — last available data, NOT current)

Workflow Health — Run Counts and Success Rate (2026-06-13 to 2026-07-06)

Daily run volume grew roughly 5x over the window (from ~70-100 runs/day in mid-June to 300-400/day by early July) while the failure count stayed roughly flat (single digits to ~50/day), so the success-rate line holds a healthy 80-95% band for most of the period, with two dips (06-17 ~70%, 07-02 ~60%) that prior audits traced to isolated driver-failure incidents rather than sustained regressions. No inference should be drawn about the 22 days since — that's exactly the gap this incident report exists to flag.

Token Usage — Daily Volume and 7-Day Trend (2026-06-13 to 2026-07-05, sparse)

Token data is sparse — several days in this window have a null token field (tracked separately as known-issue token-usage-reporting-gap), and there's a 13-day recording hole between 06-20 and 07-03 that the chart deliberately breaks the moving-average line across rather than smoothing over. Where data exists, daily volume fell from ~40-55M tokens/day in mid-June to ~1-4M tokens/day in early July, consistent with the same period's run-count/success-rate improvements (fewer expensive retries), but the gap means this trend cannot be confirmed as sustained.

Repo memory updated

• known-issues.json: +1 (agenticworkflows-logs-tool-total-outage, CRITICAL, new)
• recommendations.json: +1 (P0 fix + canary-check proposal)
• anomalies.json: +1 (audit-gap-2026-07-07-to-2026-07-27, monitoring gap)
• audit-history.jsonl: +1 entry for 2026-07-28 marked incomplete/failed collection

Next steps

1. Escalate the logs tool workflow-name-resolution failure to gh-aw MCP maintainers — it's fast, deterministic, and reproducible (see known-issue for exact error strings).
2. If the next scheduled audit also returns zero runs, treat this as a confirmed ongoing outage and escalate further.
3. Once the tool recovers, prioritize backfilling or at least characterizing the 07-07→07-27 gap if any partial data exists in raw Actions logs.

Generated by 🔍 Agentic Workflow Audit Agent · age00 · 160.8 AIC · ⌖ 28 AIC · ⊞ 8K · ◷

• expires on Jul 29, 2026, 1:59 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Agentic Workflow Audit Agent.

A newer discussion is available at Discussion #48985.