DeepReport Intelligence Briefing - 2026-07-14

[github-actions[bot]]

🔍 Executive Summary

The gh-aw fleet is healthy and holding its 82–95% band (83.2% over 101 runs; detection-enabled runs at 93.1% vs 82.4% regular), safe-output jobs are 100% clean for a 2nd consecutive fully read-only day, and every non-copilot engine (claude 91.7%, pi, codex, gemini, antigravity) is reliable. The top finding is that the failure surface remains almost entirely Smoke CI's copilot-CLI 0-turn streak (9/9 fail, 14+ days, tracked #45429) — which is simultaneously the #1 REST API burner at 20.7% of daily quota — while a separate antigravity engine-catalog change keeps getting re-filed by agents despite 3 maintainer rejections (#45017 / PR #44649). Urgent action: root-cause the copilot-CLI driver (one fix retires both the largest failure family and a fifth of API spend) and stop the antigravity re-filing loop rather than merging it.

Continuity: last full DeepReport was #45291 (2026-07-13), which filed 7 issues (PR Sous Chef fetch, detection-config #45284, engine-catalog #45285/#45286, engine docs, safe-output actuation, PR-cluster API caching). This run avoids duplicating those — the antigravity/detection items are confirmed already open and unactuated, so no re-files. Also pruned a stale 32 KB orphan memory/default/ folder (dated 2025-12-27) that had reappeared in repo memory.

🚨 Top 5 Findings

• Smoke CI is the whole story — failure AND cost. 9/9 copilot-CLI 0-turn failures (recur Weekly Research Report: AI Workflow Automation Landscape and Market Opportunities - August 2025 #7+, unfixed since 06-30) and 20.7% of the last-24h REST quota (15,175 calls / 13 runs). It single-handedly drags regular-run success from ~93.8% down to 82.4%. Tracked [aw] Smoke CI failed #45429; fix the driver.
• antigravity engine-catalog re-filing loop = agent churn. The same catalog-sync change has been filed 4+ times ([Code Quality] Document antigravity engine in engines.md and update schema/code comments #45403/[deep-report] Add antigravity engine to main_workflow_schema.json and engines.md #45286/[deep-report] Fix stale built-in engine catalog comment in engine_definition.go (add pi, antigravity) #45285) and rejected 3 times by maintainers ([doc-healer] DDUw improvement: stop re-attempting the rejected antigravity engine catalog sync (3rd rejection: PR #44649) #45017, PR Sync built-in engine catalog references for antigravity and pi #44649). Agents keep re-proposing a change the humans have declined — a governance gap, not a code gap.
• logs MCP tool 60s timeout is degrading audits (3+ days). 07-11/07-13/07-14 runs of Safe Output Health ([safe-output-health] 🏥 Safe Output Health Report - 2026-07-14 #45381) and API Consumption ([api-consumption] 📊 GitHub API Consumption Report — 2026-07-14 #45454) both fell back to partial datasets (context deadline exceeded mid-download). Fleet metrics are being reported on incomplete data.
• Token telemetry is half-dark (recur Add workflow: githubnext/agentics/weekly-research #4). The audit's token-usage-reporting-gap persists — only ~10 of 22 days carry a fleet token total, so every report substitutes AIC as a cost proxy and the token-trend chart is unreliable.
• Detection is winning but 5 workflows still misconfigured. Detection-enabled runs beat regular ones (93.1% vs 82.4%) at ~1.06× token cost, yet 5 audit workflows still lack gh-aw-detection: true ([deep-report] Add gh-aw-detection: true to 5 flagged audit/analysis workflows #45284, filed 07-13, still open and unactuated 07-14).

✅ Actionable Agentic Tasks

Seven GitHub issues were filed from this analysis. All are new — none duplicate the open antigravity (#45403/#45286/#45285), detection-config (#45284), or 07-13 #45291 batch:

1. Fix cli:update_check duplicate logger namespace collision — Fast — from [repository-quality] [repository-quality] Repository Quality Improvement Report — Logger Name Collision & Instantiation Hygiene #45489.
2. Un-shadow stdlib log in 6 files so rawloginlib can fire (linters + modelsdev) — Quick — from [repository-quality] Repository Quality Improvement Report — Logger Name Collision & Instantiation Hygiene #45489.
3. Make inline/dynamic logger.New calls greppable (close_entity_helpers.go + newValidationLogger) — Quick — from [repository-quality] Repository Quality Improvement Report — Logger Name Collision & Instantiation Hygiene #45489.
4. Fix recurring 60s timeout in gh-aw logs MCP tool degrading audit datasets — Medium — from [safe-output-health] 🏥 Safe Output Health Report - 2026-07-14 #45381/[api-consumption] 📊 GitHub API Consumption Report — 2026-07-14 #45454.
5. Surface fleet-wide token total (token-usage-reporting-gap recur Add workflow: githubnext/agentics/weekly-research #4) — Medium — from [audit-workflows] [audit-workflows] 🔍 Agentic Workflow Audit — 2026-07-13: 86% clean baseline, one 16-workflow startup burst #45337.
6. Cut Smoke CI REST API footprint (20.7% of quota) via caching repo-content reads — Medium — from [api-consumption] [api-consumption] 📊 GitHub API Consumption Report — 2026-07-14 #45454.
7. Add title-only/bounded mode for list_issues & search_issues MCP triage to cut token bloat — Medium — from [mcp-analysis] [mcp-analysis] MCP Structural Analysis - 2026-07-14 #45467.

View Full Details — metrics snapshot & non-refiled watch items

Metric	Value (2026-07-14)
Runs/day	101 sampled; success 83.2% (82–95% band)
07-13 anomaly	Raw 64% = one 16-workflow startup burst (14:47Z); clean baseline 86.4%
Engine health	claude 91.7%, pi 81.8%, copilot 71.9% (85.4% ex-Smoke-CI), codex/gemini/antigravity 100%
Detection	93.1% (enabled) vs 82.4% (regular), ~1.06× tokens; 5 workflows misconfigured
Safe-output jobs	35/35 = 100%; 2nd consecutive read-only day (0 messages actuated)
REST API	73.4K calls/day (+22.0% WoW); Smoke CI 20.7%, top-3 = 42.2%
Issues (7d, 500)	118 open / 382 closed; 0 unlabeled; 0 open >7d; 470/500 bot-authored
MCP	17,586 tok/day, usefulness 3.55/5; list_workflows pagination broken (~3,880 tok)

Watch items (already tracked — deliberately NOT re-filed):

• Smoke CI copilot-CLI 0-turn ([aw] Smoke CI failed #45429) — root-cause the driver.
• antigravity catalog ([Code Quality] Document antigravity engine in engines.md and update schema/code comments #45403/[deep-report] Add antigravity engine to main_workflow_schema.json and engines.md #45286/[deep-report] Fix stale built-in engine catalog comment in engine_definition.go (add pi, antigravity) #45285/[doc-healer] DDUw improvement: stop re-attempting the rejected antigravity engine catalog sync (3rd rejection: PR #44649) #45017) — needs a re-filing stop, not another issue.
• detection misconfig ([deep-report] Add gh-aw-detection: true to 5 flagged audit/analysis workflows #45284) — open + unactuated since 07-13.
• pi PR Sous Chef 0-turn re-surface — early-warning (preceded the 07-03 pi collapse).
• 0% safe-output actuation, 2nd read-only day — reconcile intentional-vs-regression.

References:

• §29328025217 — GitHub API Consumption run (Smoke CI 20.7%)
• §29304143346 — Safe Output Health run (logs MCP timeout)
• §29343934320 — this DeepReport run

Generated by 🔬 Deep Report · 260.7 AIC · ⌖ 17.6 AIC · ⊞ 10K · ◷

• expires on Jul 21, 2026, 7:23 AM UTC-08:00