DeepReport Intelligence Briefing — April 9, 2026

[github-actions[bot]]

🔍 Executive Summary

The gh-aw agent ecosystem is in a partial outage state on April 9, 2026. A P1 silent startup crash affecting all Copilot CLI v1.0.21 workflows has been active for 35+ hours (since Apr 8 01:02 UTC), driving open [aw] failure issues to an all-time high of 61 — a 2.5× single-day jump from 25 yesterday. The crash affects an estimated 124 Copilot-engine workflows (66% of the fleet), producing exit code 1 with zero output and no retries. Issue #25468 tracks the outage; a companion MCP gateway wildcard bug (#25497) may be a contributing root cause.

Despite the outage, the Claude and Codex portions of the ecosystem remain operational. Safe-output health holds at 100% for the fifth consecutive day. The active Copilot coding agent is achieving 100% merge success on its current fix-discussion-label-limit task — back-to-back perfect days (Apr 8–9). The PR merge rate trajectory (82%→90%→100%) remains the most positive trend in the data set. Schema analysis (Typist, Schema Consistency) has surfaced two new actionable Go code quality findings that are independent of the Copilot outage.

Three actionable quick-win tasks were identified and filed as GitHub issues: adding the missing proxy-args field to the MCP stdio schema, wiring up the dead MCP config schema validator, and investigating the 17-day Daily Issues Report Generator failure.

---

📊 Pattern Analysis

Positive Patterns

Safe-output infrastructure remains bulletproof. Five consecutive days at 100%. Today's run processed 10 write operations (across push_to_pull_request_branch, update_pull_request, add_comment types) with zero failures. The April 2 rate-limit incident appears fully resolved with no recurrence.

Active Copilot coding agent achieving peak performance. The fix-discussion-label-limit task is producing flawless results: 100% merge success on Apr 8 and Apr 9. Session depth correlates with quality — 10.24 min average for successful Copilot sessions today vs 0.2 min for review-bot passes.

Stale lock files remain zero. The batch-recompile that cleared 19 stale files on Apr 3 has held; no new stale locks have accumulated in 6 days despite the fleet growing by +5 files to 187 total.

PR merge rate improvement continues. 76.0% (Mar 31) → 78.4% (Apr 3) → 82% (Apr 7) → 90% (Apr 8) — a +14 point improvement over 9 days. The underlying quality of agent-generated PRs is improving even as failure count rises (the failures are startup crashes, not merge rejections).

Concerning Patterns

Copilot CLI v1.0.21 silent crash cascade. The dominant story. All Copilot workflows fail within 1–2 seconds at startup with exit code 1, zero output, and no retry. isCAPIError400=false rules out API 400 errors. The v1.0.21 release merged Apr 8; the concurrent MCP gateway tools wildcard bug (tools: ["*"] filtering all tools instead of allowing them) may compound the failure. Affected workflows: Auto-Triage Issues, Architecture Guardian, Daily Documentation Healer, Contribution Check, Refactoring Cadence, Daily Go Function Namer, Smoke Multi PR, and 15+ others confirmed today.

Workflow failure issues at all-time high. 61 open [aw] failure issues as of Apr 9, vs 25 on Apr 8 and 15–18 on Apr 6–7. This is a 4× increase in 3 days, almost entirely attributable to the Copilot cascade. The failure issues are accumulating without triage — most carry only the agentic-workflows label with no severity classification.

Daily Issues Report Generator: 17-day failure. First flagged Apr 3 at 11 days; now at 17+ consecutive days with no fix applied despite an issue being filed. This represents the longest-running unresolved workflow failure in recent memory. The step failing is "Fetch issues data" — distinct from the current Copilot startup crash pattern — suggesting a separate, persistent data access issue.

Dead MCP config schema validation. Schema Consistency Check identified that mcp_config_schema.json (14 properties, maintained) is compiled but never invoked for runtime validation. schema_validation.go:102 has a comment stub with no function body. Invalid MCP server configurations silently bypass validation. This has been present for an unknown duration and represents a latent security/reliability risk.

Emerging Patterns

Schema technical debt accumulating. Two schema findings from the Apr 9 Schema Consistency Check (missing proxy-args in stdio_mcp_tool, dead MCP config validation) join existing gaps. The schema infrastructure appears to be growing faster than its test/validation coverage.

Go type naming conflicts emerging at scale. Typist analysis of 666 Go source files found a HIGH-impact naming conflict: two completely unrelated structs named RepoConfig exist in different packages (pkg/cli/trial_types.go and at least one other). With 654 total type definitions across 20 packages, type namespace collisions will increase as the codebase grows without a naming convention audit.

Shared component adoption plateau at 81%. 36 of 187 workflows (19%) have zero shared component imports. The top shared components (shared/reporting.md at 96 uses, shared/observability-otlp.md at 55) are near-universal, but 70+ available components have single-digit adoption. The 19% floor appears sticky.

---

📈 Trend Intelligence

Metric	Apr 3	Apr 7	Apr 8	Apr 9	Direction
Open [aw] failures	~15	~18	25	61	↑↑↑ (P1 cascade)
Copilot PR merge rate	78.4%	82%	90%	100%	↑↑
Safe-output failures	0	0	0	0	→ (stable)
Stale lock files	19	0	0	0	→ (stable)
Lock file count	241	~185	182	187	↑ (steady)
Firewall block rate	—	—	2.9%	—	Low (healthy)

Token telemetry remains a coverage gap — the daily audit reports 0 tokens and $0 cost for all runs, which is a data collection issue rather than true zero usage.

The first two days of GitHub API call tracking show 1,479 REST API calls on Apr 8 with an overall 12.7% success rate (29.6% non-skipped). Top consumers: Go Logger Enhancement, Test Quality Sentinel, Design Decision Gate, Auto-Triage Issues — these 4 account for 99.7% of API consumption.

---

🚨 Notable Findings

P1 outage expiry timer is ticking. Issue #25468 was filed with an expiry of 2026-04-10T12:22:51Z — approximately 22 hours from now. If the Copilot crash isn't resolved and the issue expires without a fix, the 61 accumulated failure issues will persist without a parent reference.

Copilot v1.0.21 + MCP gateway wildcard = dual failure vector. The plan issue #25497 identifies that the new v0.67.3 compilation may produce tools: ["*"] which the MCP gateway v0.2.16 incorrectly filters as a literal string, keeping 0 tools. A pending bump to DefaultMCPGatewayVersion=v0.2.17 (#25500) may resolve the gateway side of the issue.

Copilot fix-discussion-label-limit task delivers. PR #25430 addresses update_discussion silently truncating labels to 3 due to validateLabels defaulting maxCount — a real user-facing bug. This is the active Copilot agent task with 100% success rate.

DIFC integrity filtering is blocking 11 items per run. The Workflow Health Manager run shows 11 items filtered for integrity below "approved" level. This pattern is stable and expected but creates persistent manual review overhead.

---

🔮 Predictions and Recommendations

Near-term: Resolving the Copilot v1.0.21 crash (#25468) will auto-recover ~124 workflows and close/expire the majority of the 61 open [aw] failure issues. The MCP gateway version bump (#25500) should be prioritized alongside the CLI fix. Once resolved, expect open [aw] failures to drop back to the 10–15 range.

Medium-term: The schema technical debt pattern (dead validation, missing properties, type conflicts) suggests a dedicated "schema health" sprint would be high value. Three independent findings in one day (Typist + Schema Consistency × 2) indicate more gaps likely exist.

Recommendation: Consider adding a priority label (P1, P2, P3) to [aw] failure issues at creation time based on failure pattern matching. The current 61 open failures are all labeled only agentic-workflows, making it impossible to distinguish P1 cascades from isolated failures at a glance.

Recommendation: The Codex chatgpt.com telemetry pattern (13/18 blocked requests = 72% of all blocks) is low-risk but noisy in firewall reports. Consider either adding chatgpt.com to an explicit blocklist with a descriptive comment, or filtering it from default firewall alert thresholds.

---

✅ Actionable Agentic Tasks (Quick Wins)

Three GitHub issues were created for the following tasks:

[deep-report] Add proxy-args to stdio_mcp_tool schema — Fast (< 30 min). The $defs.stdio_mcp_tool in main_workflow_schema.json uses additionalProperties: false but omits proxy-args, causing valid workflows to fail schema validation. Simple schema edit.

[deep-report] Wire up dead MCP config schema validation — Medium (1–4 hours). ValidateMCPConfigWithSchema at schema_validation.go:102 is a comment stub with no body. The compiled schema is maintained but never invoked, providing zero runtime protection against invalid MCP configs. Implement the function and wire up the caller.

[deep-report] Investigate and fix Daily Issues Report Generator 17-day fetch failure — Quick (< 1 hour). The "Fetch issues data" step has been failing for 17+ consecutive days. An issue was filed Apr 3 but not resolved. Investigation to determine if this is a Copilot-engine issue (would self-resolve with P1 fix) or an independent data-access regression.

---

📚 Source Attribution

Discussions Analyzed

• #25477 — Safe Output Health Report (2026-04-09)
• #25463 — Daily Copilot Agent Session Analysis (2026-04-09)
• #25449 — Daily Copilot Agent Analysis (2026-04-09)
• #25455 — Typist Go Type Consistency Analysis (2026-04-09)
• #25454 — Workflow Skill Extractor Report (2026-04-09)
• #25427 — Schema Consistency Check (2026-04-09)
• #25413 — Lock File Statistics (2026-04-09)
• #25409 — GitHub API Consumption Report (2026-04-08)
• #25388 — Daily Audit (2026-04-08)
• #25389 — Regulatory Report (2026-04-08)
• #25286 — Daily Firewall Report (2026-04-08)
• #25328 — DeepReport Briefing (2026-04-08) [prior report]

Issues Referenced

• #25468 — P1: Copilot silent startup crash
• #25497 — MCP gateway tools wildcard bug
• #25500 — DefaultMCPGatewayVersion bump
• #25453 — Expand shared/pr-code-review-config.md adoption

Repo Memory

• Previous analysis: memory/deep-report/ branch (2026-04-03T15:00:00Z)
• Trend data, known patterns, and flagged items loaded and updated

Data Range

• Discussions: Available dataset (100 discussions, categories: announcements/audits/general/q-a/dev/artifacts)
• Issues: 7-day rolling window (500 issues, Apr 2–9)
• Workflow logs: MCP server timed out; analysis based on available discussion reports

References:

• §24197600442 — This DeepReport run
• §24172367990 — Schema Consistency Check (Apr 9)
• §24189259135 — Workflow Health Manager (P1 report)

Generated by DeepReport - Intelligence Gathering Agent · ● 418.5K · ◷

• expires on Apr 16, 2026, 3:19 PM UTC

[github-actions[bot]]

🤖 Beep boop! The smoke test agent was here! 🎉

Run §24201784352 says hello. All tests green. The robots are operational. 🦾✨

(Fun fact: This comment was generated by a Copilot smoke test validating that discussion interaction works — and it does! 🎊)

📰 BREAKING: Report filed by Smoke Copilot · ● 869.2K · ◷

[github-actions[bot]]

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

A newer discussion is available at Discussion #25673.