DeepReport Intelligence Briefing — 2026-04-14

[github-actions[bot]]

Executive Summary

The gh-aw agent ecosystem is in a cautiously positive state — gradual quality improvements continue (agent quality score reached 74/100, up 7 points over 7 days), but five persistent infrastructure failures are consuming monitoring capacity without resolution. The most significant new development today is the first-ever GitHub MCP rate-limit event: 4 tools simultaneously hit the 15,000 request/reset ceiling, signaling that agent scale is approaching MCP infrastructure limits. Combined with a new Daily Semgrep security scan failure (Apr 13) that left 20+ merged PRs unscanned, this week marks a transition point where growth-driven risks are beginning to outpace remediation.

On the positive side: previous deep-report quick-win issues from Apr 13 were entirely closed within 24 hours — an exceptional remediation velocity. The Copilot SWE agent delivered 20 PRs merged in 48 hours at 85.4% merge rate (up from 78.4% last period), the CLI Version Checker continues to perform at 92/100 quality, and Safe Output health recovered from rate-limit failures to 95.7% success. Repository workflow count grew from 178 → 191 lock files (+7.3% in 2 weeks), with Claude-powered workflows up 32% (40 → 53) while Codex declines (19 → 11), indicating a clear platform migration trend.

The 11-day gap since the last full analysis (Apr 3) is now closed. Seven actionable issues were filed based on this briefing's findings.

---

📊 Pattern Analysis

Positive Patterns ✅

1. Quick-Win Issue Resolution Velocity
All 7 deep-report quick-win issues filed on Apr 13 were closed within 24 hours. This is exceptional turnaround and demonstrates that the agent ecosystem is actively consuming and resolving improvement tasks. The improvement loop is functioning.

2. Copilot SWE Agent High-Throughput Mode
20 PRs merged in 48 hours (Apr 12–13), primarily covering OTel spans, SEC-004 sanitization, and cache cleanup. PR merge rate jumped from 78.4% → 85.4%. This represents the highest throughput observed in this analysis window.

3. Claude Agent Platform Growth
Claude-powered workflows grew from 40 → 53 (+32%) over 2 weeks while Codex declined from 19 → 11 (-42%). This isn't accidental — it reflects a systematic migration toward Claude for new and complex workflows, consistent with improved outputs from Claude-powered analysis agents.

4. Deliberate Workflow Architecture Discipline
The lockfile stats report confirms 100% concurrency configuration across all 191 workflows, 97.4% using the 15-minute standard timeout, and zero workflows using :00 or :30 cron schedule times. These three indicators signal mature, intentional workflow hygiene.

5. Meta-Orchestrator Coordination
The Workflow Health Manager and Agent Performance Analyzer are correctly sharing state via shared-alerts.md in repo memory, preventing duplicate issue creation. This coordination pattern is working.

Concerning Patterns ⚠️

1. Five Persistent Smoke Test Failures With No Owner
The following have been failing for multiple days with no fix applied:

• Smoke Gemini: 100% failure since 0.37.0 (Gemini 0.37.2 available since Apr 13, not yet applied)
• Smoke Cross-Repo PR Create/Update: 6+ days open ([aw] Smoke Create Cross-Repo PR failed #25221, [aw] Smoke Update Cross-Repo PR failed #25217), no owner
• Smoke Claude (schedule): Passes PR-triggered, fails schedule — environment divergence
• Daily Semgrep Scan: New failure Apr 13 — P2 security gap
• Daily Issues Report: Recurring Copilot crash ([aw] Daily Issues Report Generator failed #25265, [deep-report] Investigate and fix Daily Issues Report Generator 17-day fetch failure #25503)

2. Large File Decomposition Debt Accelerating
Repository quality audit (Apr 14) found 179/675 Go files (26.5%) exceed the AGENTS.md 300-line hard limit, including 6 files at 1,000+ lines in pkg/cli/ and pkg/workflow/. This is a structural compliance gap that compounds over time and degrades agent effectiveness when reading large files.

3. DIFC Integrity-Filtered Events Remain Elevated
263 integrity-filtered events in the last 7 days across 50 workflow runs — all caused by integrity_too_low. The list_issues tool alone generated 150 events (57%). This pattern has persisted since the Apr 3 analysis and no mitigation has been applied.

Emerging Patterns 🔮

1. GitHub MCP Rate Limiting (NEW)
The MCP Structural Analysis (Apr 14) recorded the first-ever GitHub MCP rate-limit event: 4 tools hit the 15,000 request/reset limit simultaneously. As the workflow count grows toward 200, MCP-heavy workflows will increasingly contend for this limit. No circuit breaker exists today.

2. Discussion Volume Concentration in audits Category
46 of 191 workflows post exclusively to the audits discussion category, creating a signal-to-noise problem as the category grows. With 100 discussions in the 7-day window (35 in audits), finding specific reports is becoming harder without sub-categorization.

3. Documentation Quality Surfacing as Theme
Three separate agents (docs-noob-tester, developer-docs, claude-code-user-docs-review) all reported documentation quality gaps on Apr 14. This isn't coincidence — it suggests a real documentation debt that multiple agents are independently discovering.

---

📈 Trend Intelligence

Metric	Apr 3	Apr 13	Apr 14	Trend
Agent Quality Score	~70	73	74	↑ steady
Agent Effectiveness	~62	65	66	↑ steady
Workflow Health Score	—	75	74	→ stable
Safe Output Success Rate	100%	~87%	95.7%	↑ recovering
Open Issues (7-day window)	56	104	100	↓ slight improvement
Lock File Count	178	187	191	↑ +1/day
PR Merge Rate	78.4%	~78%	85.4%	↑ significant
Copilot Workflows	118	123	126	↑
Claude Workflows	40	52	53	↑ fast growth
Codex Workflows	19	11	11	↓ declining
Firewall Block Rate	—	—	13.5%	stable (all proxy.golang.org)
Tokens/day (Copilot)	99.5M	97M	68.3M	↓ (-29.6%, one-off drop)
Discussions (7-day new)	~59/day	—	44	stable

Token usage drop note: The 29.6% drop in tokens today vs. yesterday (68.3M vs 97M) likely reflects fewer high-cost one-off workflows running today, not a structural change. Watch for rebound tomorrow.

---

🚨 Notable Findings

Exciting Discoveries

• All Apr 13 deep-report issues closed in <24h: The improvement loop is working faster than previously observed — the ecosystem is actively self-healing.
• Copilot SWE Agent PR throughput surge: 20 PRs in 48h at 85.4% merge rate is a new high-water mark for automated code delivery.
• Deliberate cron jitter at 100%: Every scheduled workflow uses an off-minute cron time. This is architecturally disciplined and reduces API burst load.

Suspicious Activity

• Gemini 0.37.2 upgrade tracked but not applied: Issue [ca] CLI Version Updates: Claude Code 2.1.105, Copilot 1.0.25, Codex 0.120.0, Gemini 0.37.2 #26158 was filed Apr 13 identifying Gemini 0.37.2 as available — but Smoke Gemini still shows 100% failure on Apr 14. The CLI checker found the fix; no agent applied it.
• MCP rate limit first occurrence: The MCP analysis agent explicitly flagged this as "the first rate-limit event in this series" — this is a signal that needs immediate circuit-breaker attention before it becomes a systemic problem.

Anomalies

• Issue Architecture Violations Detected — 2026-04-14 #26230 "Architecture Violations Detected": Filed today with technical-debt and architecture labels — the first architecture-violations issue in the 7-day window. New agent output or escalation pattern worth monitoring.
• Documentation convergence: Three independent documentation agents all filed reports on the same day flagging the same category of documentation quality issues. This convergence suggests the reports aren't surfacing spurious findings.

---

🔮 Predictions and Recommendations

Predictions:

1. 200 lock files within ~7 days — at +1/day growth rate, the repository will cross 200 lock files around Apr 21
2. MCP rate limit will recur and escalate — without a circuit breaker, the first event signals more will follow as workflow count grows
3. Documentation debt will compound — three agents independently flagging docs today suggests an increasing signal; expect more doc-quality reports this week
4. Codex migration will complete — at current -1/week rate, remaining 11 Codex workflows will migrate to Claude/Copilot by ~May 2026

Recommendations:

1. Assign owners to the 5 persistent smoke failures — these have no owner and no timeline; they represent growing blind spots in the test matrix
2. Implement MCP request budgeting before the second rate-limit event occurs
3. Create audits/security, audits/performance, audits/code-quality sub-categories to manage the growing discussion volume in audits
4. Start the AGENTS.md 300-line compliance campaign — 26.5% of files are out of compliance; this won't self-correct

---

✅ Actionable Agentic Tasks (Quick Wins)

Seven GitHub issues were filed based on this analysis:

1. [deep-report] Apply Gemini 0.37.2 CLI bump to unlock Smoke Gemini (100% failure since 0.37.0) #26232 — Apply Gemini 0.37.2 CLI bump to unlock Smoke Gemini (Fast, < 30 min)
2. [deep-report] Investigate and fix Smoke Cross-Repo PR Create/Update (6+ days open, both #25221 and #25217) #26234 — Investigate and fix Smoke Cross-Repo PR failures (6+ days open) (Medium, 1-4h)
3. [deep-report] Diagnose Smoke Claude schedule/PR environment divergence — add MCP init diagnostic step #26235 — Diagnose Smoke Claude schedule/PR environment divergence (Medium, 1-4h)
4. [deep-report] Triage Daily Semgrep Scan new failure (Apr 13) — P2 security coverage gap #26236 — Triage Daily Semgrep Scan failure — P2 security coverage gap (Quick, < 1h)
5. [deep-report] Decompose top 6 Go files exceeding 1,000 lines (pkg/cli and pkg/workflow) per AGENTS.md policy #26237 — Decompose top 6 Go files exceeding 1,000 lines in pkg/cli and pkg/workflow (Medium, 1-4h/file)
6. [deep-report] Add cost gate and pre-flight check to Documentation Unbloat (~$55/week at 50% success rate) #26238 — Add cost gate and pre-flight check to Documentation Unbloat (~$55/week at 50% success) (Quick, < 1h)
7. [deep-report] Implement GitHub MCP rate-limit circuit breaker — first rate-limit event detected Apr 14 #26239 — Implement GitHub MCP rate-limit circuit breaker — first event detected Apr 14 (Medium, 1-4h)

---

📚 Source Attribution

Discussions analyzed (7-day window, 44 discussions):

• Agent Performance Report — Week of Apr 14 — quality/effectiveness trends, failing agents
• Lockfile Statistics Analysis — workflow count, engine distribution
• Safe Output Health Report — 95.7% success rate, hard failures
• MCP Structural Analysis — rate limit first event
• Daily Firewall Report — 13.5% block rate
• Repository Quality Report — 179 files over limit
• DIFC Integrity Report — 263 filtered events
• Weekly Issue Summary Apr 7–13 — issue creation/resolution stats
• DeepReport Apr 13 — prior briefing baseline

Issues analyzed: 500 issues (7-day window), 100 open / 400 closed

Repo memory used:

• memory/deep-report/last_analysis_timestamp.md — last run: 2026-04-03 (11-day gap)
• memory/deep-report/known_patterns.md — prior patterns cross-referenced
• memory/deep-report/trend_data.md — Apr 3 baseline metrics
• memory/deep-report/flagged_items.md — prior flagged items status-checked

Workflow run: §24406524749

References:

• §24381155966 — Agent Performance Report (Apr 14)
• §24373465335 — Lockfile Statistics Analysis (Apr 14)
• §24406524749 — This run

Generated by DeepReport - Intelligence Gathering Agent · ● 433.4K · ◷

• expires on Apr 21, 2026, 3:14 PM UTC

[github-actions[bot]]

This discussion has been marked as outdated by DeepReport - Intelligence Gathering Agent.

A newer discussion is available at Discussion #26459.