DeepReport Intelligence Briefing - 2026-07-24

[github-actions[bot]]

🔍 Executive Summary

The gh-aw fleet remains healthy — 96.1% raw success (256 runs, 07-24) / 97.2% raw · 97.6% ex-intentional (07-23 audit), zero missing-tools/data/MCP errors, all failures mapping to known families. The top finding is execution debt, not health: all 7 of yesterday's DeepReport tasks (#47603–#47609) are still open and unactioned, the Code Quality agent has independently filed duplicates of three of them (#47652, #47678, #47653/#47679), and the PR Sous Chef scheduled target-resolution hard-fail (fix #47607 unmerged) recurred 3 times in 2 days. Most urgent watch: REST API consumption is up +46% over 30 days (169.8K on 07-24), concentrated in four PR-review workflows that re-fetch diffs every run — the clearest untracked optimization lever.

🚨 Top 5 Findings

• Fix backlog is stalling (HIGH). Yesterday's 7 tasks ([deep-report] Add recover() to WASM compile goroutine to prevent browser runtime hang #47603–[deep-report] Join the CheckForUpdatesAsync goroutine and drop its time.After timer leak #47609) all remain OPEN; two agents are now filing duplicate goroutine/timeout fixes ([deep-report] Add recover() to WASM compile goroutine to prevent browser runtime hang #47603↔[Code Quality] Add panic recovery to WASM goroutine in cmd/gh-aw-wasm/main.go #47652, [deep-report] Join the CheckForUpdatesAsync goroutine and drop its time.After timer leak #47609↔[Code Quality] Fix timer leak in CheckForUpdatesAsync — use time.NewTimer + defer Stop() #47678, [deep-report] Harden HTTP server timeouts (bootstrap GitHub App + MCP server) against slow-client resource exhaustion #47604↔[Code Quality] Add HTTP server timeouts to bootstrap_profile_github_app.go #47653/[Code Quality] Add WriteTimeout and IdleTimeout to MCP HTTP server #47679) — split effort, noisy backlog. Consolidation filed this run.
• PR Sous Chef scheduled target-resolution recurring (HIGH). update_pull_request/add_comment hard-fail on scheduled runs with no PR context (vs review-comment's soft-skip) — 3 hard-fails in 2 days (07-23, 07-24×2). Tracked [deep-report] Soft-skip add_comment/update_pull_request when no PR/issue context exists on scheduled runs #47607 + [aw-failures] [aw-fix] P1: PR Sous Chef — update_pull_request (branch-from-base) rejected by "verified signatures" rule, failing [Content truncated due to length] #47753; fix unmerged. DO NOT RE-FILE.
• REST API +46%/30d, untracked (MEDIUM). 169.8K on 07-24; top-4 reviewers (PR CQR 15.2K, Test Quality Sentinel 11.5K, Matt Pocock 11.4K, Impeccable 10.2K) = ~54%. No open issue covers PR diff/file-list caching — filed.
• CWI + CGO regressed to 100% AR (MEDIUM). Both were stabilized Jul 8, both now 100% activation-refused (correlated CI signal). CGO tracked [CGO] Workflow failure on main - Run #11644 #38777; CWI has no issue — filed. Broader drift: agent quality 61 / effectiveness 62 (↓ from 63/64 Jul 8).
• Positive core + [WIP] waste. Engine health strong (claude 100%, codex 100%, copilot 96.8%); token reporting recovered (~10.97M/day); PR Sous Chef/Avenger/Auto-Triage top performers; backlog clean (0 open >7d). But prompt clustering shows [WIP] PRs merge at 26% vs 76% — repeated same-task retries (robots.txt ~5×, js-integration-live-api 3×) dominate wasted cycles.

✅ Actionable Agentic Tasks

Seven data-driven quick-win issues filed this run, each verified non-duplicate against the open backlog. Deliberately NOT re-filed (already tracked): #47603–#47609 (yesterday's tasks), #47607/#47753 (PR Sous Chef target), #47606 (observability), #47605 (pr_review_buffer 422), #38777 (CGO), #43527 (Q), #44241 (AI Moderator), #47614 (Matt Pocock optimize), #47641/#47369 (robots.txt).

1. Add a join point to StartDockerImagePull (docker_images.go:155) — Fast; the only HIGH goroutine-leak gap not covered by yesterday's tasks.
2. Consolidate duplicate goroutine/HTTP-timeout fix issues ([deep-report] Add recover() to WASM compile goroutine to prevent browser runtime hang #47603↔[Code Quality] Add panic recovery to WASM goroutine in cmd/gh-aw-wasm/main.go #47652, [deep-report] Join the CheckForUpdatesAsync goroutine and drop its time.After timer leak #47609↔[Code Quality] Fix timer leak in CheckForUpdatesAsync — use time.NewTimer + defer Stop() #47678, [deep-report] Harden HTTP server timeouts (bootstrap GitHub App + MCP server) against slow-client resource exhaustion #47604↔[Code Quality] Add HTTP server timeouts to bootstrap_profile_github_app.go #47653/[Code Quality] Add WriteTimeout and IdleTimeout to MCP HTTP server #47679) + add dedup guard — Quick; stops two agents doing the same work.
3. Investigate CWI 100% AR regression — Quick; CWI has no issue while its twin CGO does ([CGO] Workflow failure on main - Run #11644 #38777); likely shared root cause.
4. Cache PR file-lists/diffs in the top-4 review workflows — Medium; the named lever against +46%/30d REST growth.
5. Add goleak to goroutine-heavy test packages — Medium; zero goleak usage today; locks in the in-flight goroutine fixes.
6. Label the 6-7 unlabeled open issues; extend Auto-Triage to non-report bugs — Fast; ARC/DinD env bugs (legacy-security: enable emits sudo -E awf but installs awf with --rootless, so the agent fails at startup with sudo: awf: command not found #47804, copilot/arc-dind: XDG_CONFIG_HOME resolves to unwritable /home/runner #47802, [ARC/DinD] safe-outputs: dynamic base-branch unresolved — GH_AW_INPUT_* not passed to MCP container #47795) sit unlabeled.
7. Investigate/retire dormant Changeset Generator (~28d no actuation) — Quick; highest-priority UNVALIDATED safe-output signature.

View supporting data

Metric	Value
Runs (07-24 API)	256 (246 ✅ / 10 ❌) = 96.1% raw
Runs (07-23 audit)	251 → 97.2% raw · 97.6% ex-intentional · prod-main 96.3%
Per-engine (07-23)	claude 100% · codex 100% · copilot 96.8% · pi 94.9%
Fleet tokens	~10.97M/day · AI credits ~12.3K/day
Agent scores (wk)	quality 61 (↓63) · effectiveness 62 (↓64) · AR ~42%
Safe-output in-scope (07-24)	97.44% (76/78) — both fails PR Sous Chef scheduled
REST API (07-24)	169.8K · 30d +46% · 7d +4.5% · top-4 reviewers ~54%
Prompt clustering	1000 PRs · 76.5% merge · [WIP] 26% vs 76%
Weekly issues (500/7d)	147 open / 353 closed · 0 open >7d · 6-7 unlabeled · 469/500 bot-authored
Workflows	~261

Sources analyzed: Agentic Workflow Audit #47668, Safe Output Health #47712, API Consumption #47776, Agent Performance #47788, Auto-Triage #47785, Repository Quality #47797, Prompt Clustering #47770, Daily Status #47760, weekly issues (500), prior DeepReport #47610.

Continuity: prior DeepReport 2026-07-23 (#47610). Repo-memory branch was stale (Apr-3 snapshot; a ~21 KB Dec-2025 legacy memory/default/ set had reappeared) — removed it and refreshed patterns/trends/flagged-items for 07-24 (validated 4 KB, within limit).

References:

• §30089058566
• §30057263862
• §30094134312

Generated by 🔬 Deep Report · age00 · 232.9 AIC · ⌖ 17.8 AIC · ⊞ 10.3K · ◷

• expires on Jul 31, 2026, 7:23 AM UTC-08:00