[audit-workflows] Daily Agentic Workflow Audit — 2026-08-07 (31-day audit gap found; PR Code Quality Reviewer now dominant failure cluster)

[github-actions[bot]]

Summary

Window: 2026-08-07, 11:23–21:07 UTC (~10h; the -1d lookback returned no earlier activity). 219 runs, 204 success / 15 failure → 93.2% raw, 93.6% excluding intentional-failure tests. main-branch only (excl. intentional): 102/106 = 96%. Fleet is healthy on baseline, but two findings stand out below.

⚠️ Audit cadence gap: no audit ran 2026-07-07 → 2026-08-06 (31 days)

The last recorded audit entry before this one was dated 2026-07-06. This audit's -1d window only sees the last ~10h of activity, so any fleet regressions, recoveries, or new failure modes during the 31-day gap are unverified — we have zero visibility into that period. This is a meta-monitoring gap in the audit workflow itself, not a fleet-workflow issue.

Recommendation: add an out-of-band check (e.g. a separate lightweight monitor, or an alert condition) that fires if this audit workflow itself fails to run for >48h, since a silent audit-schedule gap defeats the entire point of daily auditing. Logged as audit-cadence-gap-2026-07 (known-issue, MEDIUM) and anom-audit-cadence-gap-2026-07 (anomaly, NEW-WATCH).

🔴 "PR Code Quality Reviewer" now the dominant single failure concentration (7/20 = 35%)

All 7 failures are the chronic driver-exit signature: the agent does real work (30K–88K tokens, 8–21 minutes, non-zero AIC) but the run is still marked conclusion: failure with 0 reported turns — no missing-tools, missing-data, or MCP failures involved. This is the same long-standing copilot-sdk-driver-failures family tracked since June, but the concentration has shifted: this one workflow was 10/34=29% of the family on 2026-07-06 and is now 7/20=35%, while sibling workflows that used to share the pattern (Matt Pocock Skills Reviewer, Impeccable Skills Reviewer) are now fully clean (0 fails / 20+ runs each) this window.

Sample failing runs: 31210616628 (88K tok, 21m), 31210517576 (49K tok, 10m), 31208532093 (30K tok, 12m).

Recommendation: root-cause the specific Copilot CLI invocation path used by this workflow (18 of its 20 runs are copilot-engine, 2 are pi) rather than treating it as generic fleet noise — it's now the single largest offender in a chronic family that's otherwise resolving elsewhere. Logged as rec-2026-08-07-pr-cqr-rootcause (HIGH) and anom-pr-cqr-dominant-concentration-2026-08-07 (ESCALATE).

Other signals

• token-usage-reporting-gap resolved (watch): the previously-open issue (open since 2026-06-19, near-total gap in per-run token reporting) looks fixed — 98% of runs this window report real non-zero token_usage. Downgraded to RESOLVED-WATCH pending one more clean window.
• Design Decision Gate 🏗️ had 2/20 fails (same driver-exit signature): 31216594533, 31210669962.
• Recovery watch (small sample, not yet confirmed fixed): Avenger 0/3 fail (was chronic 100%-fail err-config-no-structured-logs), Smoke CI had zero runs this window (unconfirmed whether its prior 100%-fail-on-push issue is resolved or just unobserved), chroot-node and codex-binary-not-found issues also unobserved.
• Zero missing-tools, missing-data, or MCP failures fleet-wide this window.
• One non-agentic failure excluded from fleet stats: "Super Linter Report" (engine_id/agent null — not an agentic workflow run, out of audit scope, but note it failed on main).

Trend charts (last 30 days, incl. today)

Workflow health — daily run counts (success/failure) with success-rate line:

History is sparse before 06-13 and has a hard 31-day data gap (07-07→08-06, shown as a dashed/faded segment on the rate line — not real interpolated data). Within observed windows, the fleet oscillates between ~68–94% depending on which chronic driver-exit cluster is active that day; today's 93.2% sits at the healthy end of that range.

Token usage — daily token volume with 7-day moving average:

Several days have no token data (null, pre-dating the reporting-gap fix) so the moving average is computed only over days with real values. Today's 6.28M tokens over a partial ~10h window is consistent with, or above, the recent daily volumes recorded once reporting resumed (07-03 onward), supporting the read that the reporting gap — not actual token usage — was the anomaly.

References:

• Run 31210616628 — PR Code Quality Reviewer, 88K tok/21m driver-exit
• Run 31216594533 — Design Decision Gate, driver-exit fail
• Run 31198445994 — PR Code Quality Reviewer, 50K tok driver-exit

Generated by 🔍 Agentic Workflow Audit Agent · agent · 264.6 AIC · ⊞ 8.4K · ◷

• expires on Aug 8, 2026, 1:42 PM UTC-08:00

[github-actions[bot]]

🔍 Smoke test bot dropping by — thanks for the thorough audit! Adding a friendly comment here as part of automated Copilot engine validation (run 31222502383).

Warning

Firewall blocked 6 domains

The following domains were blocked by the firewall during workflow execution:

• accounts.google.com
• android.clients.google.com
• clients2.google.com
• contentautofill.googleapis.com
• safebrowsingohttpgateway.googleapis.com
• www.google.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "accounts.google.com"
    - "android.clients.google.com"
    - "clients2.google.com"
    - "contentautofill.googleapis.com"
    - "safebrowsingohttpgateway.googleapis.com"
    - "www.google.com"

See Network Configuration for more information.

📰 BREAKING: Report filed by Smoke Copilot · auto · 50.9 AIC · ⌖ 2.5 AIC · ⊞ 8.6K · ◷
_{Comment /smoke-copilot to run again
Add label smoke to run again}