[copilot-cli-research] Copilot CLI Deep Research - 2026-06-12

[github-actions[bot]]

Analysis Date: 2026-06-12 | Scope: 246 total workflows, 133 Copilot (54%) | Run: 27396054543

---

Executive Summary

This first comprehensive analysis of Copilot CLI usage in gh-aw reveals strong SDK mode adoption (47%) and toolsets coverage (62%), but significant gaps in safety, autopilot, and persona features. Nine opportunity areas identified.

Key finding: 9 custom .agent.md files exist in .github/agents/ (grumpy-reviewer, technical-doc-writer, adr-writer, ci-cleaner, contribution-checker, w3c-specification-writer, create-safe-output-type, interactive-agent-designer, custom-engine-implementation) yet none are used via engine.agent: in any Copilot workflow. bare mode is used in only 11% of workflows, max-continuations in just 5%, and token-weights/api-target are never used.

---

Critical Findings

High Priority

54 workflows missing strict: true (41%) - Several have write permissions. strict enables fork PR injection protections. Affected: craft.md, release.md, refactoring-cadence.md, slide-deck-maintainer.md, super-linter.md.

13 daily scheduled workflows lack concurrency guards - Without deduplication, simultaneous triggers double AI credit usage. Affected: copilot-cli-deep-research.md, daily-safe-output-integrator.md, repository-quality-improver.md, spec-extractor.md, spec-librarian.md, sub-issue-closer.md, update-astro.md, others.

Medium Priority

9 custom agent files unused - grumpy-reviewer, technical-doc-writer, adr-writer, ci-cleaner, contribution-checker and 4 more are available but never referenced via engine.agent: in production Copilot workflows.

max-continuations at 5% (7/133) - This Copilot-exclusive autopilot feature is ideal for long-running research/refactoring workflows (30-45 min timeouts). Only 7 of 133 use it.

---

Feature Usage Matrix

Feature	Used	Rate	Status
copilot-sdk	63/133	47%	Good
toolsets	83/133	62%	Good
strict	79/133	59%	54 missing
network	68/133	51%	Good
max-daily-ai-credits	39/133	29%
model	34/133	26%
cache-memory	34/133	26%
version pinning	23/133	17%
engine.agent	21/133	16%	9 files unused
experiments	17/133	13%	Low A/B
bare	14/133	11%	Underused
web-fetch	10/133	8%
max-continuations	7/133	5%	Very underused
concurrency	6/133	5%	13 daily at risk
BYOK	3/133	2%	Smoke only
token-weights	0/133	0%	Never used
api-target	0/133	0%	Never used

---

Recommendations

Immediate:

• Add strict: true to craft.md, release.md, refactoring-cadence.md, slide-deck-maintainer.md, super-linter.md
• Add concurrency guards to 13 daily-scheduled Copilot workflows

Short-term:

• Pilot engine.agent: grumpy-reviewer in refiner.md
• Add max-continuations: 2 to repository-quality-improver.md and refactoring-cadence.md
• Enable bare: true on smoke test and pure-report Copilot workflows

Long-term:

• Map all 9 agent files to appropriate workflows
• Add A/B experiments to top-5 highest-credit Copilot workflows
• Define token-weights for workflows using non-standard models

---

Best Practice Guidelines

1. strict: true on all write-permission workflows - No exceptions; prevents injection.
2. concurrency: on all scheduled workflows - Prevents credit-wasting duplicate runs.
3. Use custom agent files - Match the 9 available personas to workflow purposes.
4. max-continuations: 2-4 for complex iterative tasks with 30+ min timeouts.
5. bare: true for report-only workflows - Suppress AGENTS.md for deterministic prompts.
6. experiments: for model cost/quality tuning - Use the built-in A/B framework.
7. token-weights: when using custom models - For accurate gh aw audit reporting.

---

Baseline saved to repo-memory at copilot-cli-research/latest.json for future trend analysis.

References: 27396054543

Generated by 🔬 Copilot CLI Deep Research Agent · 739.6 AIC · ⌖ 27.2 AIC · ⊞ 19.9K · ◷

• expires on Jun 12, 2026, 9:27 PM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-06-13T05:27:11.670Z.

Closed by Workflow