[copilot-cli-research] Copilot CLI Deep Research - 2026-06-27

[github-actions[bot]]

Analysis Date: 2026-06-27
Repository: github/gh-aw
Scope: 41 Copilot workflows / 253 total .md workflows

---

📊 Executive Summary

This first Copilot CLI deep research run found 12 optimization opportunities — 3 high, 5 medium, 4 low priority. The repository makes solid foundational use of Copilot (73% strict: true, 96% gh-proxy mode where github is used), but several Copilot-exclusive capabilities remain completely untapped.

Primary Recommendation: Enable max-continuations: 3 on complex workflows — it's Copilot's most distinctive capability (autopilot mode) and is used by zero of 41 workflows.

Baseline saved to repo-memory (memory/copilot-cli-research/) for future trend tracking.

---

🔴 High Priority Findings

H1 — 33/41 workflows missing copilot-requests: write

The recommended auth path uses the Actions token automatically. Only 8 workflows use it; 33 rely on fallback behavior that may require a manually configured COPILOT_GITHUB_TOKEN secret.

permissions:
  copilot-requests: write   # ← add to all Copilot workflows

Affects all 24 github-tool workflows that lack this permission, plus others.

H2 — max-continuations never used (0/41)

Copilot's autopilot mode (--autopilot --max-autopilot-continues N) is a Copilot-exclusive feature not available in Claude, Codex, or Gemini. Zero workflows use it despite being ideal for: research.md, copilot-cli-deep-research.md, weekly-blog-post-writer.md, repository-quality-improver.md, refactoring-cadence.md.

engine: copilot
max-continuations: 3     # allow up to 3 autopilot continues
timeout-minutes: 60      # increase timeout for multi-session runs

H3 — engine.agent never used in production (0/41)

11 custom agent files exist in .github/agents/ (technical-doc-writer, adr-writer, grumpy-reviewer, ci-cleaner, etc.) but no Copilot workflow links to them via engine.agent. Direct matches: daily-doc-updater → technical-doc-writer, refiner → grumpy-reviewer, super-linter → ci-cleaner.

engine:
  id: copilot
  agent: technical-doc-writer   # .github/agents/technical-doc-writer.agent.md

---

🟡 Medium Priority Findings

M1 — 3 workflows use github: without toolsets: scoping

agent-persona-explorer.md, smoke-copilot-arm.md, weekly-blog-post-writer.md expose all 200+ GitHub MCP tools to the agent. Add toolsets: [default] to reduce token usage and blast radius.

M2 — smoke-copilot-arm.md not using mode: gh-proxy

This is the only github-tool workflow still on legacy Docker MCP mode. All others correctly use gh-proxy.

M3 — max-ai-credits missing on high-cost workflows (40/41)

Only workflow-health-manager.md sets a credit ceiling. craft.md, research.md, copilot-cli-deep-research.md, and weekly-blog-post-writer.md run without limits.

M4 — strict: true missing on 11 production workflows

agent-persona-explorer, craft, refactoring-cadence, release, repo-tree-map, slide-deck-maintainer, super-linter, test-project-url-default, visual-regression-checker, workflow-health-manager (smoke workflows excepted).

M5 — engine.bare never used

The --no-custom-instructions flag (disable AGENTS.md loading) is useful for smoke tests and detection workflows that need isolated, self-contained execution. Not used in any of the 15 smoke/detection workflows.

---

🟢 Low Priority Findings

• web-fetch underused (5/41): Several workflows use bash: [curl *] where tools: web-fetch: would be simpler and more secure.
• BYOK not explored: COPILOT_PROVIDER_BASE_URL can route Copilot CLI through any OpenAI-compatible endpoint — useful for cost optimization with cheaper models on triage-type workflows.
• mcp-scripts not used: Custom MCP CLI mounting is available but unexplored.
• engine.version never pinned for Copilot: release.md in particular would benefit from pinning for reproducibility.

---

Feature Usage Matrix

Feature	Available in Copilot	41-workflow adoption	Notes
copilot-requests: write	✅	8/41 (20%)	⚠️ Underused
github.mode: gh-proxy	✅	23/24 (96%)	✅ Excellent
github.toolsets	✅	21/24 (88%)	✅ Good
strict: true	✅	30/41 (73%)	✅ Good
network.allowed	✅	27/41 (66%)	✅ Good
cache-memory	✅	10/41 (24%)	OK
engine.model	✅	4/41 (10%)	OK
max-ai-credits	✅	1/41 (2%)	⚠️ Missing
max-continuations	✅ Copilot-only	0/41 (0%)	🚨 Unused
engine.agent	✅ Copilot-only	0/41 (0%)	🚨 Unused
engine.bare	✅	0/41 (0%)	🚨 Unused
engine.api-target	✅	0/41 (0%)	Not needed
BYOK (COPILOT_PROVIDER_*)	✅	0/41 (0%)	Not explored
web-search	via MCP	0/41 (0%)	Low demand
mcp-scripts	✅	0/41 (0%)	Not explored
engine.version (pinned)	✅	0/41 (0%)	Intentional

---

7️⃣ Action Items

Immediate (this week):

• Add copilot-requests: write to the 24 github-tool workflows lacking it
• Add toolsets: [default] to agent-persona-explorer, smoke-copilot-arm, weekly-blog-post-writer
• Fix smoke-copilot-arm: add mode: gh-proxy to github tool

Short-term (this month):

• Pilot max-continuations: 3 on research.md and repository-quality-improver.md
• Link engine.agent: technical-doc-writer to daily-doc-updater.md
• Add max-ai-credits: 300-500 to craft, research, copilot-cli-deep-research
• Add strict: true to 11 production workflows missing it

Long-term (this quarter):

• Roll out max-continuations to all complex scheduled workflows
• Document internal engine.agent usage patterns and publish guideline
• Evaluate BYOK routing for cost optimization on triage workflows
• Implement copilot-token-audit tracking to measure AIC costs over time

---

References:

• §28279210901 — this research run
• Engines docs
• copilot_engine_execution.go

Generated by 🔬 Copilot CLI Deep Research Agent · 149.1 AIC · ⌖ 14.1 AIC · ⊞ 7.4K · ◷

• expires on Jun 27, 2026, 9:09 PM UTC-08:00

[francoDeveloper1]

Hello Copilot Engine Execution Agent Here .Yes please I'm freelancer python Bot developer. Who can develop this copilot requirements as python bot client to someone's interested My portfolio skills python bot projects can be reached here at www.upwork.com/freelancers/alfonsof4 Ok Let me know .Thanks

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-06-28T05:09:22.720Z.

Closed by Workflow