[copilot-cli-research] Copilot CLI Deep Research - 2026-05-29

[github-actions[bot]]

Analysis Date: 2026-05-29 · Scope: 236 workflows, 126 Copilot (53%) · Previous Run: §26491933777

---

📊 Executive Summary

16th consecutive run. The Copilot CLI ecosystem is mature and growing — strict mode (161/236 workflows), imports (233/236), and network restrictions (123/236) show strong adoption. Model overrides are popular (68 workflows). However, 7 safety/productivity features have never been used across all 16 runs. The most actionable gap: startup-timeout and tool-timeout — zero workflows configure these protective guards despite availability.

---

🔴 Critical Findings

1. Zero workflows use startup-timeout or tool-timeout (16 runs)

A hung Copilot CLI call silently burns minutes in workflows with timeout-minutes: 30–180. These fields directly map to GH_AW_STARTUP_TIMEOUT / GH_AW_TOOL_TIMEOUT.

timeout-minutes: 30
startup-timeout: 120    # fail fast if agent doesn't start (seconds)
tool-timeout: 300       # kill stuck tool calls after 5 min (seconds)

Affected: agent-persona-explorer.md (180 min), api-consumption-report.md (45 min), all other long-running workflows.

2. max-continuations used by only 5/126 Copilot workflows

Autopilot mode (--autopilot --max-autopilot-continues N) enables multi-turn autonomous runs. Only contribution-check.md uses it meaningfully (max: 20). Candidates: architecture-guardian.md, dev-hawk.md, dead-code-remover.md, docs-noob-tester.md.

---

🟡 Medium Priority

3. Zero engine.version pinning (16 runs)

All 126 Copilot workflows always install latest. One breaking CLI release affects everything. Pin high-value workflows: hourly-ci-cleaner.md, contribution-check.md.

4. mcp-scripts adoption: 1 workflow (down from 3 on May 25)

Only security-review.md uses this. Workflows with large bash: allowlists could use mcp-scripts instead.

5. max-runs used by only 2 workflows

No cost cap on 27+ scheduled experiments. Add max-runs: 30 to daily/hourly workflows.

---

📉 Feature Usage Matrix

Feature	Count	%	Status
strict mode	161	68%	✅ strong
engine.agent	17	14%	✅ growing
engine.bare	10	8%	✅ in use
max-continuations	5	4%	🟡 low
mcp-scripts	1	0.8%	🟡 low
BYOK	1	0.8%	🟡 new
startup-timeout	0	0%	🔴 never
tool-timeout	0	0%	🔴 never
engine.args	0	0%	🔴 16 runs
engine.api-target	0	0%	🔴 16 runs
engine.harness	0	0%	�� never
engine.version	0	0%	🔴 16 runs

---

7️⃣ Action Items

Immediate:

• Add startup-timeout: 120 + tool-timeout: 300 to agent-persona-explorer.md
• Add timeout guard template to shared/ imports

Short-term:

• Enable max-continuations: 3-5 for architecture-guardian.md, dev-hawk.md
• Pin engine.version in contribution-check.md, hourly-ci-cleaner.md
• Add max-runs: 30 to all scheduled workflows lacking cost caps

Long-term:

• Document concrete engine.args use cases or mark as advanced/internal
• Evaluate surfacing engine.harness with examples for custom retry logic

---

References:

• §26619051030 (this run)
• §26491933777 (previous run)

Generated by 🔬 Copilot CLI Deep Research Agent · sonnet46 5M · ◷

• expires on May 30, 2026, 5:12 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-05-30T05:12:42.120Z.

Closed by Workflow