[copilot-cli-research] Copilot CLI Deep Research - 2026-05-31

[github-actions[bot]]

Analysis Date: 2026-05-31 · Scope: 236 total workflows, 97 Copilot (41%) · Run: §26703913319
Previous Analysis: §26491933777 (2026-05-27)

---

📊 Executive Summary

This is the 14th consecutive deep-research run. Persistent gaps remain (engine.args, engine.api-target, engine.harness at zero for 14+ runs). First-time BYOK usage observed (daily-byok-ollama-test). Cache-memory adoption grew to 116 workflows (up from 95). AWF sandbox expanded to 23 workflows (up from 16).

Primary Recommendation: Enable max-continuations on multi-step code-change workflows and add network: defaults to the 48 Copilot workflows running without network restrictions.

---

Critical Findings

🔴 High Priority

1. 48/97 Copilot workflows lack network restrictions — code-change workflows running without egress guardrails
2. max-continuations used in only 5% of workflows — 5/97; Copilot-exclusive feature for autonomous multi-step runs
3. 5 custom agent files never used — grumpy-reviewer, interactive-agent-designer, w3c-specification-writer, create-safe-output-type, custom-engine-implementation

🟡 Medium Priority

4. engine.args at zero — 14th consecutive run with zero usage; may indicate docs gap
5. 86/97 Copilot workflows use default model — lightweight tasks could use model: small for 2-5x speed/cost improvement

---

2️⃣ Feature Usage Matrix

Feature	Available	Used	Rate
max-continuations	✅ Copilot-exclusive	5 wfs	5%
engine.agent (custom persona)	✅	20 wfs	21%
engine.bare	✅	12 wfs	12%
model: override	✅	12 wfs	12%
Network restrictions	✅	49 wfs	50%
Sandbox AWF	✅	23 wfs	24%
mcp-scripts	✅	12 wfs	12%
BYOK (engine.env)	✅	1 wf	1% (NEW)
engine.version pinning	✅	0	0%
engine.args custom	✅	0	0%
engine.api-target	✅	0	0%
engine.harness	✅	0	0%

---

3️⃣ Top Recommendations

Network restrictions — add to 48 unguarded workflows:

network: defaults

max-continuations — add to complex code-change workflows:

max-continuations: 5
timeout-minutes: 60

Model right-sizing — add to lightweight report/triage workflows:

model: small

Custom agent activation — use committed but unused agent files:

engine:
  id: copilot
  agent: grumpy-reviewer

---

5️⃣ Historical Trends

Metric	05-14	05-21	05-25	05-27	05-31	Trend
sandbox AWF	16	16	22	16	23	↑
max-continuations	5	5	12	5	5	↔
cache-memory (all engines)	92	73	30*	95	116	↑
engine.args	0	0	0	0	0	❌
BYOK	0	0	0	0	1	✅ NEW

*05-25 counted copilot-only

---

7️⃣ Action Items

This week: Add network: defaults to 48 unguarded workflows; enable max-continuations: 5 on dead-code-remover.md, code-scanning-fixer.md; decide fate of 5 unused agent files.

This month: Add model: small to 15+ lightweight workflows; create PR review workflow using engine.agent: grumpy-reviewer.

This quarter: Evaluate version pinning for stability-critical workflows; expand BYOK pattern; document engine.args use cases.

---

Research methodology: Static analysis of all .github/workflows/*.md files. Historical data in memory/copilot-cli-research branch.

Generated by 🔬 Copilot CLI Deep Research Agent · sonnet46 3.8M · ◷

• expires on Jun 1, 2026, 5:17 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-06-01T05:17:40.787Z.

Closed by Workflow