Agent Performance Report — Week of 2026-06-29

[github-actions[bot]]

Run: §28377470831 | Analysis period: Jun 22–29, 2026

Executive Summary

• Workflows in ecosystem: 257 (257/257 compiled ✅)
• Engine mix: copilot 158 (61%), claude 60 (23%), pi 20 (8%), codex 15 (6%), other 4
• Active workflows (last 100 runs): 43 unique
• Overall quality score: 63/100 (→ stable)
• Overall effectiveness score: 64/100 (→ stable)
• Ecosystem health: 80/100 (↓2 — 5 P1, 6 P2 failures active)
• Top performers: Copilot SWE Agent, PR Triage, Team Status, Static Analysis, Agentic Maintenance
• Persistent underperformers: Sub-Agent Model Audit, Safe Output Integrator, BYOK Ollama, Go Logger, PR Code Quality Reviewer
• New systemic issue: Escalating tool denial pattern across 3 agents → [systemic] Multiple agents hitting tool denial limit — structural complexity reduction needed #42258

Performance Rankings

Top Performing Agents 🏆

1. Copilot SWE Agent (Q:92, E:91)

   • 80% merge rate (61/76 settled PRs); highest-volume contributor
   • Recent work: actionlint quoting, eslint rule completeness, dashboard, safe-outputs fixes
   • Sample PRs: fix(safe-outputs): set_issue_type GraphQL intent path uses wrong query and field name #42232, Surface Copilot HTTP 400 response failures in workflow conclusions #42228, Add usage reporting and continuation-aware logs windows to the dashboard canvas #42226, fix(actionlint): quote RUNNER_TEMP paths and add SC2016 to AWF shellcheck disable #42224, eslint-factory: suggestions for .stack/.code and flag computed string-literal access in no-unsafe-catch-error-property #42223

2. PR Triage Agent (Q:88, E:86) — 1/1 today; 5/5 window; structured risk/priority reports. Output: [PR Triage Report] Agent PR Triage Report — 2026-06-29 Run §28376613466 #42251

3. Team Status (Q:85, E:83) — 1/1 success; well-formatted daily reports. Output: [team-status] Daily Status Report — June 29, 2026 🌟 #42242

4. Static Analysis (Q:84, E:81) — 1/1 success; 11+ days zero High findings. Output: [static-analysis] Report - 2026-06-29 #42187

5. Workflow Health Manager (Q:82, E:80) — Accurate P1/P2 tracking; good shared-alerts coordination. Output: [Workflow Health Dashboard] 2026-06-29 #42186

6. Agentic Maintenance (Q:80, E:82) — 3/3 success; 100% reliable

7. Auto-Triage Issues (Q:78, E:80) — 2/2 success; FULLY RECOVERED from P1 ✅

8. Bot Detection (Q:75, E:78) — 1/1 success

9. PR Sous Chef (Q:74, E:76) — 1/1 success; consistent peer reviews

Agents Needing Improvement 📉

1. Daily Safe Output Integrator (Q:10, E:10) — P1; tool denial 5/5 (3rd cycle). Tracked: [aw] Daily Safe Output Integrator exceeded tool denial limit #42125. Do not re-file.
2. Daily BYOK Ollama Test (Q:10, E:5) — P1; api-proxy 503, 9+ day outage. Tracked: [aw-failures] Daily BYOK Ollama Test 100% red for 8+ days — offline+BYOK api-proxy returns 503 on /v1/models, Copilot CLI gets H [Content truncated due to length] #41827. Do not re-file.
3. Go Logger Enhancement (Q:20, E:20) — P1; jq ARG_MAX prevents agent start. Tracked: [aw-failures] Go Logger Enhancement 100% red — "Build deterministic logger manifest" step dies with jq: Argument list too long [Content truncated due to length] #42032. Do not re-file.
4. Sub-Agent Model Resolution Audit (Q:10, E:5) — P1; Codex alpha 404. Tracked: [aw-failures] Daily Sub-Agent Model Resolution Audit 100% red — Codex gpt-5-codex-alpha-2025-11-07 404s (same alpha-snapshot d [Content truncated due to length] #42033. Do not re-file.
5. PR Code Quality Reviewer (Q:30, E:25) — P1; tier-unsupported model, action_required. Tracked: [aw-failures] PR Code Quality Reviewer red — Copilot general-purpose subagent requests tier-unsupported model → SDK 400 `model [Content truncated due to length] #42095. Monitor for recovery after Pin PR Code Quality Reviewer sub-agent to a supported Copilot model #42209.
6. AI Moderator (Q:25, E:20) — P2; 0/4 success, zero safe outputs. Tracked: [aw] AI Moderator produced no safe outputs #42234. Do not re-file.

Inactive / Skipped

• Deployment Incident Monitor: all 4 runs skipped (correct skip-if-match behavior)
• Daily Max AI Credits Test: 1 intentional failure (by design)
• 221/257 workflows: no recent runs (scheduled but not yet triggered)

Quality and Effectiveness Analysis

Quality Distribution (active agents)

• Excellent (80–100): 6 — Copilot SWE Agent, PR Triage, Team Status, Static Analysis, Health Manager, Agentic Maintenance
• Good (60–79): 3 — Auto-Triage, Bot Detection, PR Sous Chef
• Fair (40–59): 2 — Content Moderation, Avenger
• Poor (<40): 5 — Safe Output Integrator, BYOK Ollama, Go Logger, Sub-Agent Model Audit, AI Moderator

Common Issues

1. Tool budget exhaustion (3 agents): entire run wasted because tool limit hit before safe output emitted
2. Retired model pinning (2+ agents): gpt-5-codex-alpha-2025-11-07 → 404
3. Missing tool declarations (2 agents): Code Metrics ([aw] Daily Code Metrics and Trend Tracking Agent is missing required tool #42124), Team Evolution ([aw] Daily Team Evolution Insights is missing required tool #42128)

Run Success Rates (last 100 runs)

Workflow	Runs	Successes	Rate
Agentic Maintenance	3	3	100%
CI	2	2	100%
Auto-Triage Issues	2	2	100%
Bot Detection	1	1	100%
PR Triage Agent	1	1	100%
PR Sous Chef	1	1	100%
Smoke CI	7	3	43%
Content Moderation	4	2	50%
Agentic Commands	15	5	33%
AI Moderator	4	0	0%
Q workflow	14	0	0%*

*Q action_required (71%) likely by-design (dispatch approval flow); not a true failure.

PR Merge Stats

• copilot-swe-agent: ~80% merge rate (last 30 PRs)
• github-actions automation: ~76% merge rate

Behavioral Patterns

Productive ✅

• Copilot SWE → Issue Monster → PR loop: ~80% PR resolution within 24h of creation
• Health Manager → Avenger coordination: Health alerts reliably reflected in recovery tracking
• Meta-orchestrator shared-alerts: All three orchestrators reading/writing consistently; zero duplicate issues this cycle

Problematic ⚠️

• Recurring tool denial (escalating): Safe Output Integrator failed under 3 different issue numbers. Root cause (scope too broad) persists across fix cycles. Systemic issue filed: [systemic] Multiple agents hitting tool denial limit — structural complexity reduction needed #42258
• Alpha model pinning: gpt-5-codex-alpha-2025-11-07 still referenced in multiple workflows despite [aw-failures] Daily Sub-Agent Model Resolution Audit 100% red — Codex gpt-5-codex-alpha-2025-11-07 404s (same alpha-snapshot d [Content truncated due to length] #42033. No repo-wide sweep performed yet.
• Zero-output runs: AI Moderator runs that produce zero safe outputs are indistinguishable from failures; agents should emit noop with a summary when no action is taken.

Coverage Analysis

Well-Covered

• PR review, triage, and code quality
• Daily health and status reporting
• Dependency management and spec sync
• Documentation generation and maintenance

Gaps

• No stale PR detection (PRs open >7d with no activity)
• No AIC burn rate alerting (reporting exists but no threshold alerting)
• No automated P1 recovery or auto-close for resolved issues
• No compile-time validation of deprecated model strings

Recommendations

High Priority

1. Fix escalating tool denial → [systemic] Multiple agents hitting tool denial limit — structural complexity reduction needed #42258 (filed today)

   • Agents: Safe Output Integrator, Formal Spec Verifier, Layout Spec Maintainer
   • Fix: add batch-size cap + early-noop instruction to each prompt
   • Impact: eliminate 3 daily P1/P2 failures

2. Repo-wide Codex alpha model sweep (linked to [aw-failures] Daily Sub-Agent Model Resolution Audit 100% red — Codex gpt-5-codex-alpha-2025-11-07 404s (same alpha-snapshot d [Content truncated due to length] #42033)

   • grep -r gpt-5-codex-alpha .github/workflows/ → update all hits to GA model
   • Impact: unblock Sub-Agent Model Audit, Cache Strategy Analyzer, PR Code Quality Reviewer

3. Add model deprecation check to gh aw compile

   • Warn/fail on known-retired model snapshot strings at compile time
   • Impact: prevent future silent alpha-model breakage

Medium Priority

4. Fix missing tool declarations for Code Metrics ([aw] Daily Code Metrics and Trend Tracking Agent is missing required tool #42124) and Team Evolution ([aw] Daily Team Evolution Insights is missing required tool #42128)
5. Mandate noop emission for zero-output agents (AI Moderator, similar)
6. Create stale PR detection agent (PRs open >7d)

Low Priority

7. Fix smoke-copilot missing message input ([aw-failures] Smoke Copilot safe_outputs red — dispatch_workflow to haiku-printer omits required input message #41988)
8. Add workflows scope to Changeset Generator ([aw-failures] Changeset Generator safe-output push rejected — review-branch push needs workflows scope (remote rejected, agent [Content truncated due to length] #41987)
9. AIC burn rate alerting in agentic-token-audit

Trends

Metric	Jun 29	Jun 28	Δ
Quality	63/100	63/100	→
Effectiveness	64/100	64/100	→
Health	80/100	82/100	↓2
Compilation	257/257	257/257	→
P1 failures	5	4	↑1
P2 issues	6	6	→
SWE merge rate	80%	80%	→
Tool denial incidents	3	2	↑1

Actions This Run

• ✅ Filed systemic tool denial issue: [systemic] Multiple agents hitting tool denial limit — structural complexity reduction needed #42258
• ✅ Updated shared memory (agent-performance-latest.md, shared-alerts.md)
• ✅ Created this weekly performance discussion

Next Steps

1. Fix tool denial root causes ([systemic] Multiple agents hitting tool denial limit — structural complexity reduction needed #42258)
2. Run Codex alpha model sweep ([aw-failures] Daily Sub-Agent Model Resolution Audit 100% red — Codex gpt-5-codex-alpha-2025-11-07 404s (same alpha-snapshot d [Content truncated due to length] #42033)
3. Monitor Q workflow and AI Moderator ([aw] AI Moderator produced no safe outputs #42234)
4. Add model deprecation validation to compile step

---

Analysis period: 2026-06-22 to 2026-06-29 | Next report: 2026-07-06

References:

• §28377470831 — this run
• §28351970090 — Workflow Health (Jun 29)
• §28323141732 — prior Agent Perf run (Jun 28)

Generated by ⚡ Agent Performance Analyzer - Meta-Orchestrator · 84.8 AIC · ⌖ 10.2 AIC · ⊞ 10.3K · ◷

• expires on Jun 30, 2026, 6:12 AM UTC-08:00