DeepReport Intelligence Briefing - 2026-07-20

[github-actions[bot]]

🔍 Executive Summary

The gh-aw fleet is broadly healthy but running below peak: production success is 91.4% (128/140 runs) with API consumption flat and safe-outputs at 94.6% job success on a write-rich day, yet the ecosystem health score has slipped to ~60/100 (↓ from 63 on Jul 8) on a cluster of workflow-specific failures. The top finding is a HIGH-severity silent-green failure: when the safeoutputs gateway registers zero tools, every safe-output call fails "unknown tool" while the run still reports success (#46826/#46807/#46847) — this masks real breakage as green and is the most urgent fix. Urgent action: merge the 12-day-stale PR #43527 to revive the fully non-functional Q workflow, and switch AI Moderator off the 404ing Codex engine.

Continuity: prior full DeepReport was 2026-07-17 (#46271); repo-memory before that was 3 months stale (2026-04-03) and has now been refreshed. A recurring stale memory/default/ orphan (~21 KB of old data) reappeared and pushed repo-memory over the 12 KB cap again — pruned this run; memory now validates at 4 KB.

🚨 Top 5 Findings

• Silent-green safeoutputs failure (HIGH). Gateway starts late, registers 0 tools → every call fails "unknown tool", but the run is marked success (Copilot MCP tools fail on v0.82.x: Squid denies internal gateway (awmg-mcpg) with 403 TCP_DENIED #46826, cache-memory: update_cache_memory (+ conclusion AIC cache) save fails "token has no writable scopes" — job lacks actions: write #46807, Safeoutputs backend starts after gateway timeout, remains at zero tools, and workflow reports success #46847). Real breakage renders as green. Needs a fail-closed zero-tool guard.
• Q workflow non-functional 12+ days. Quality 10/100, effectiveness 5/100 (4/5 skipped + 1 AR). Single blocker: quality-gate PR Add shared prompt quality gate for plateaued agent-review workflows #43527 unmerged. Escalate the merge.
• AI Moderator dead on Codex 404 ([aw] AI Moderator produced no safe outputs #44241). 3/5 auto-rejected + 2/5 skipped; prompt hardcodes Codex despite persistent 404. Known fix = switch engine to Copilot.
• logs MCP 120s fetch cap degrades daily metrics. Workflow Audit ran on 100/204 runs, Safe-Output Health on 82/131 conclusions, API Consumption morning-only — three audits compute health/cost on partial data daily (fix [deep-report] Fix recurring 60s timeout in gh-aw logs MCP tool degrading audit datasets #45510 closed, symptom persists).
• Positive: pi engine fully recovered (96.3%) after the 07-03 collapse, run.TokenUsage now populated fleet-wide, and Copilot SWE Agent shipped 20 PRs on Jul 20 alone (quality 92). Weekly issues are well-triaged — only 1 of 500 unlabeled, 0 open >7 days.

✅ Actionable Agentic Tasks

Seven quick-win issues were filed this run (data-driven, independent, high-impact):

1. Switch AI Moderator engine Codex→Copilot — Fast; fixes recurring 404 ([aw] AI Moderator produced no safe outputs #44241).
2. Escalate merge of PR Add shared prompt quality gate for plateaued agent-review workflows #43527 — Quick; unblocks the 12-day-dead Q workflow.
3. Add fail-closed guard for safeoutputs zero-tools silent-green — Medium; protects every write workflow (Copilot MCP tools fail on v0.82.x: Squid denies internal gateway (awmg-mcpg) with 403 TCP_DENIED #46826/cache-memory: update_cache_memory (+ conclusion AIC cache) save fails "token has no writable scopes" — job lacks actions: write #46807/Safeoutputs backend starts after gateway timeout, remains at zero tools, and workflow reports success #46847).
4. Fix chroot-node-not-available in Daily Issues Report Generator — Medium; long-recurring copilot-CLI failure.
5. Fix logs MCP 120s fetch cap — Medium; restores full datasets to 3 daily audits.
6. Widen tool allow-lists for Agent of the Day Blog Writer & Layout Specification Maintainer — Quick; both newly hit the tool-denial limit.
7. Scope PR-review workflows with path filters — Medium; cut redundant REST API consumption from the top-burning reviewer family.

View supporting data

Metric	Value
Runs (day)	140 (128 ✅ / 12 ❌) = 91.4%
REST API calls/day	85,565 (flat; stable 75k–105k band)
Safe-output job success	94.6% (35/37), 202 writes
Ecosystem health	~60/100 (↓ from 63 Jul 8)
Open org issues / PRs	61 (60 unassigned) / 54
Discussions updated (7d)	47 (Audits 28, General 13, Announcements 4, Daily News 2)
Weekly issues (500 window)	112 open / 388 closed; 1 unlabeled

Sources analyzed: Agent Performance (#46837), Org Health (#46783), Safe Output Health (#46714), Workflow Audit (#46683), API Consumption (#46808), Auto-Triage (#46831), Issue Arborist (#46727), Lockfile Stats (#46669), prior DeepReport (#46271).

References:

• §29745587435
• §29739077610
• §29754584155

Generated by 🔬 Deep Report · 165.1 AIC · ⌖ 17 AIC · ⊞ 10K · ◷

• expires on Jul 27, 2026, 7:29 AM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Deep Report.

A newer discussion is available at Discussion #47104.