[audit-workflows] π Agentic Workflow Audit β 2026-06-22 (94.5% healthy; Avenger & Skillet quiet; copilot-sdk long-run now #1) #40878
Closed
Replies: 2 comments
-
|
Smoke test ping from run 27990513949. Warning Firewall blocked 6 domainsThe following domains were blocked by the firewall during workflow execution:
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.
|
Beta Was this translation helpful? Give feedback.
0 replies
-
|
This discussion has been marked as outdated by Agentic Workflow Audit Agent. A newer discussion is available at Discussion #41112. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
π Agentic Workflow Audit β 2026-06-22
Window: 2026-06-21T23:23Z β 2026-06-22T21:35Z (~22.2h, 10 paginated batches) Β· Repo: github/gh-aw
A healthy, on-baseline day. Overall success 94.5% (259/274), prod-main 91.7% (92.9% excluding 2 intentional test workflows), PR/dev branches 98.9%. Zero new failure classes, zero missing-tools / missing-data / MCP failures, and no shared incident β all 31 observability episodes were standalone. Two multi-day chronic offenders went quiet: Avenger (5-day 100%-fail streak) and Skillet (06-20 fleet-wide incident) both produced 0 failures this window.
π Trend Charts (30 days)
Workflow health: Today's 94.5% sits comfortably above the 90% reference line and continues the rebound that began 06-21 (91.7%) after the 06-20 Skillet incident (69.6%). Run volume (274) is back to normal after the inflated 06-20/06-21 counts that were dominated by Skillet's failing push runs. The 30-day trend is stable in the high-80s to mid-90s with two clear dips (06-13 Avenger/PR-Sous-Chef clusters, 06-20 Skillet).
Token usage: The 7-day moving average is trending down toward ~37M/day from the mid-50s, with the 06-12 spike (127M) being a clear outlier. Today's point is absent because the token artifact was empty β a persistent observability gap that has now affected several recent windows and should be fixed so cost trends remain reliable.
β Failures (15) β all known classes, no cluster
Every one of the 15 failures is a distinct workflow β no single-workflow cluster like Avenger or Skillet on prior days. 2 are intentional self-tests.
Failure breakdown by class
Dominant pattern shift: the copilot-sdk-driver long-run variant is now the #1 failure family (5 fails). The agent does real work for 20β34 minutes, then the job fails (
session_starts=0+ failed-tool-execution signature) β distinct from the classic 0-token / 25-min permission lockout. This burns ~130 minutes of compute/day with no output and is the highest-ROI fix on the board. It only displaced Avenger because Avenger didn't run today.π’ Notable resolutions / watch items
Chronic offenders that went quiet (verify root cause)
slash_command(centralized)trigger no longer fires on arbitrary push.π οΈ Top recommendations
nodenot on PATH in AWF chroot). Bind-mount node + add anode --versionpre-flight check.token_usageartifact has recurred across multiple windows; restore it so cost trends stay reliable.Data caveats
countβ€80to avoid the logs-tool 120s timeout.nullfor 40 runs (aw_info not fetched); failure engines resolved via thestatustool's authoritativeengine_id.References: Β§27966232293 Β· Β§27970148606 Β· Β§27979436784
Beta Was this translation helpful? Give feedback.
All reactions