DeepReport Intelligence Briefing — April 10, 2026

[github-actions[bot]]

🔍 Executive Summary

The gh-aw agent ecosystem is in active recovery on April 10, 2026. The dominant story this week was the Copilot CLI v1.0.21 silent startup crash that paralyzed 66% of the fleet (124 Copilot-engine workflows) from April 8–9 UTC. The v1.0.22 fix is now deployed and the recovery signal is clear: Copilot session completion rate climbed from 0% (Apr 7) → 8% (Apr 9) → 18% today, with 6 active coding-agent sessions — the highest count since April 4. The residual damage is 63 open [aw] failed issues, a record high that now dominates the failure backlog and obscures genuinely broken workflows.

The non-Copilot portions of the ecosystem continue to perform well. The Claude engine has grown +30% in 11 days and is increasingly taking on structural analysis tasks (typist, repository-quality, docs-review). Auto-triage is clean (0 unlabeled issues), the firewall block rate is low (1.06%), and token efficiency has improved 55% since April 4 (122.6M → 54.7M Copilot tokens, 99 runs). Two structural failures remain outstanding: Design Decision Gate (100% failure — empty prompt bug) and Test Quality Sentinel (90% failure — stale PR branch checkout), neither of which were caused by the Copilot outage.

Three quick-win issues have been filed: fixing the Design Decision Gate prompt bug, bulk-closing the outage-era [aw] failures, and splitting the largest validation file back under the AGENTS.md 300-line limit.

---

📊 Pattern Analysis

Positive Patterns

Copilot recovery trajectory is strong. The Apr 8–9 v1.0.21 crash was the ecosystem's largest single-day disruption, but recovery is proceeding at the expected pace. The CLI Version Checker successfully identified and filed the update (v1.0.22), the fix was deployed, and Copilot session rates are climbing. The PR merge rate held at 78.4% through the outage — healthy and stable.

Auto-triage operating at peak. Today's report found 0 unlabeled issues (against a 5% target), a perfect run. The DIFC integrity filter caught 7 community-submitted issues that required min-integrity relaxation, correctly protecting the triage pipeline. This demonstrates mature pipeline safety.

Claude engine expanding successfully. Workflows using Claude are growing fastest in the fleet: +30% in 11 days. Today's outputs include go-type-consistency analysis (Typist), validation file governance (repository-quality), user docs noob-testing, UX analysis, and MCP structural analysis — all high-quality structured reports. Smoke Claude continues to exercise the widest safe-output palette in the ecosystem (22 items per run).

Token efficiency improving. Copilot tokens: 122.6M (Apr 4) → 96.5M (Apr 6) → 88.0M (Apr 7) → 54.7M (Apr 9). This 55% reduction with run counts recovering to near-peak (99 runs) strongly suggests individual workflows are becoming more efficient, not just running less.

Concerning Patterns

63 open [aw] failed issues — record backlog. This is a 2.5× spike from 25 on April 9. Most originate from the v1.0.21 outage and are no longer actionable, but they are drowning the signal from genuinely broken workflows. Without bulk-closing these, the agentic-workflows label becomes a noise channel.

Design Decision Gate persistent 100% failure. All 10/10 PR-triggered runs failed on April 9 with an empty --print prompt to Claude Code. The prompt construction step (awk extraction from adr-writer.agent.md) is producing empty PROMPT_TEXT. This has been failing for at least 1 day of full observation and represents a complete gap in automated ADR generation.

Test Quality Sentinel 90% failure rate. The workflow attempts to checkout PR branches that are already merged/deleted by the time it runs. Nine of ten runs failed with git exit code 128. The fix is a simple guard check before checkout — but it has not been applied.

Firewall observability gap persists. Ten firewall-enabled runs are missing access.log (Design Decision Gate × 3, Test Quality Sentinel × 3, Daily Semgrep Scan, GitHub API Consumption, Daily Observability, plus one more). The Observability Coverage report confirmed firewall logging coverage at only 67.7% (21/31 runs). This means roughly one-third of firewall activity is invisible to the daily firewall report.

Emerging Patterns

Validation file governance pressure. The repository-quality agent has now filed or confirmed 9 validation files exceeding the AGENTS.md 300-line hard limit. The largest is mcp_config_validation.go at 462 lines. This is a slow-burn governance issue: the repository is growing (905K LOC, +4.8K from yesterday) but the validator splits haven't kept pace.

Human-reported issues rising. Today saw 4 ai-inspected issues from community users: OIDC env vars not forwarded to MCP Gateway docker run command (#25646), create_pull_request_review_comment missing at runtime (#25656), MCP servers policy blocked on GHE (#25626), and Copilot CLI v1.0.21 COPILOT_MODEL regression (#25593). All 4 represent real integration failures visible to external users — worth prioritizing.

---

📈 Trend Intelligence

Metric	Apr 3	Apr 9	Apr 10	Trend
Workflows compiled	241	~185	187	↑ stable
Open issues (7d window)	56	—	113	↑ (outage spike)
Open [aw] failures	~25	~61	63	↑ critical
Safe-output success rate	100%	100%	85.7%	↓ slight dip
WHM score	72/100	—	65/100	↓ (outage)
Copilot session completion	—	8%	18%	↑ recovering
Copilot tokens (daily)	99.5M	54.7M	—	↓ efficiency ↑
Codebase LOC	—	900.5K	905.3K	↑ steady
Lock file avg size	66.2KB	—	70.4KB	↑ +6.4% (11d)
Claude engine share	—	~34%	~36%	↑ growing

Safe-output dropped to 85.7% today (from 100% five consecutive days). Root causes: upload_artifact staging failure (2 occurrences in Smoke Copilot), update_issue context error (1), and a permission error on resolve_pull_request_review_thread (1). These are not systemic — the first is likely an artifact staging race condition, the others are contextual errors.

---

🚨 Notable Findings

Safe-output tool reported missing at runtime (#25656). User JanKrivanek filed a confirmed bug: create_pull_request_review_comment is declared in the workflow but not found at runtime by the safe-outputs job. This is a production safe-output infrastructure bug that could affect multiple workflows silently.

OIDC env vars not forwarded to MCP Gateway (#25646). The compiler's docker run command does not pass ACTIONS_ID_TOKEN_REQUEST_URL and ACTIONS_ID_TOKEN_REQUEST_TOKEN environment variables. This blocks any MCP server that requires OIDC token-based authentication. Filed by bbonafed — affects any workflow using OIDC-auth MCP servers.

MCP get_me returns 403 on every run. The MCP analysis agent rated this tool 1/5 — "completely unusable in this environment." This has been a persistent failure for 7+ days. No fix has been filed. The tool is being called in workflows and consuming turns before failing.

Copilot POTD / constraint-solving label now active. The "🧩 Constraint Solving POTD: Sports League Scheduling" announcement appeared today, confirming that the constraint-solving and problem-of-the-day labels previously noted as missing have been created and are operational.

Codebase adding new CLI modules. Today's code metrics show pkg/cli/drain3_integration.go, pkg/cli/runner_guard.go, and pkg/cli/logs_github_rate_limit_usage.go as new files — indicating active CLI feature development around log rate limiting and runner guards.

---

🔮 Predictions and Recommendations

1. Copilot [aw] failure backlog will keep growing unless explicitly bulk-closed. With 63 open and 6+ new ones per day, the issue list will exceed 100 within a week. The filed quick-win issue addresses this.

2. Design Decision Gate will remain broken until the awk/prompt construction bug is fixed manually or by agent. This is a deterministic failure with no self-healing mechanism. Every PR going forward creates one more failed run.

3. Firewall observability gap will persist until the 10 affected workflows have their firewall logging wired correctly. Recommend the Observability agent file explicit issues for each affected workflow.

4. MCP get_me should be removed from all workflows until the 403 is resolved. Each call wastes turns. Consider creating a noop or skip-if-forbidden wrapper.

5. Claude engine share will continue growing. At +30% in 11 days, Claude is taking over structural analysis tasks. Workflows should be reviewed to ensure prompt sizes are optimized for Claude's pricing model (unlike Copilot subscription).

---

✅ Actionable Agentic Tasks (Quick Wins)

Three GitHub issues have been filed based on this analysis:

1. Fix Design Decision Gate 100% failure — empty prompt when --print flag used ([deep-report] prefix)

• Root cause identified: awk extraction from adr-writer.agent.md produces empty PROMPT_TEXT
• Fast fix (< 30 min) — inspect and correct prompt construction

2. Bulk-close resolved [aw] failure issues from Apr 8–9 Copilot v1.0.21 outage ([deep-report] prefix)

• 63 open failures cluttering the agentic-workflows backlog
• Medium effort — check recent workflow run status, close outage-era issues

3. Split mcp_config_validation.go (462 lines) to restore AGENTS.md 300-line compliance ([deep-report] prefix)

• Biggest violator of 9 files exceeding the hard limit
• Medium effort — split into tools-section and MCP schema/requirements files

---

📚 Source Attribution

Discussions Analyzed (past 7 days, 38 total)

• [safe-output-health] Safe Output Health Report - 2026-04-10
• [lockfile-stats] Agentic Workflow Lock File Statistics — 2026-04-10
• Agent Performance Report — Week of 2026-04-10
• [audit-workflows] Agentic Workflow Audit - April 9, 2026
• [copilot-token-audit] Daily Copilot Token Usage Audit — 2026-04-09
• [daily-firewall-report] Daily Firewall Report - 2026-04-10
• [observability] Observability Coverage Report - 2026-04-09
• [observability] Agentic Observability Report — 2026-04-06
• [repository-quality] Repository Quality Improvement Report – Validation File Size Governance
• [daily-code-metrics] Daily Code Metrics Report - 2026-04-09
• [mcp-analysis] MCP Structural Analysis - 2026-04-10
• [copilot-session-insights] Daily Copilot Agent Session Analysis — 2026-04-10
• [Auto-Triage] Auto-Triage Report - 2026-04-10
• [typist] Go Type Consistency Analysis Report
• [claude-code-user-docs-review] Claude Code User Documentation Review - 2026-04-10
• DeepReport Intelligence Briefing — April 9, 2026

Issues Analyzed

• 500 issues from past 7 days (113 open, 387 closed)
• 63 open [aw] failed issues identified
• 4 human-reported ai-inspected issues from external users

Repo Memory

• Previous analysis: memory/deep-report/ from 2026-04-03 (7 days ago — full analysis required)

References:

• §24243815112 — Smoke Copilot (upload_artifact failure)
• §24194682643 — Design Decision Gate (empty prompt failure)
• §24241366040 — Daily Firewall Report

Generated by DeepReport - Intelligence Gathering Agent · ● 368.7K · ◷

• expires on Apr 17, 2026, 3:04 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #26054.