메뉴
Umbrella for the three Copilot delegation skills (/copilot-cli JSONL, /copilot-acp ACP, /copilot-sdk Go SDK). Default routes to /copilot-cli. Use --compare to fan out across all three for evaluation. README.md in this folder explains the trade-offs.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | copilot |
| description | Umbrella for the three Copilot delegation skills (/copilot-cli JSONL, /copilot-acp ACP, /copilot-sdk Go SDK). Default routes to /copilot-cli. Use --compare to fan out across all three for evaluation. README.md in this folder explains the trade-offs. |
| allowed-tools | Bash, Read, Write, Glob |
You are the /copilot umbrella skill. There are three backend skills that all delegate to GitHub Copilot via different mechanisms:
/copilot-cli — copilot -p --output-format json. Python runner parses JSONL. This is the default workhorse./copilot-acp — copilot --acp. Python client speaks JSON-RPC over stdio (Agent Client Protocol)./copilot-sdk — github.com/github/copilot-sdk/go. Go program built from scripts/main.go.For routine work, use /copilot (which routes to /copilot-cli). For evaluation or comparison, use /copilot --compare. To force a specific backend, type the backend skill directly (/copilot-acp ...).
| User input | What to do |
|---|---|
/copilot <task> | Forward to /copilot-cli. Same flags pass through. |
/copilot --compare <task> | Run the task through all three backends in parallel and print a comparison table (final answer + usage per backend). |
/copilot --backend acp <task> | Force a specific backend (cli / acp / sdk). |
/copilot --usage | Print a cost table across all .copilot-runs/*.usage.json. |
Run bash .agents/skills/copilot/route.sh <user-args...>. route.sh does the dispatching:
/copilot-cli--backend acp|sdk|cli <task> → that backend--compare <task> → all three in parallel via compare.sh--usage → cost table via usage.sh--bootstrap → run both shared installer + Go buildPass stdout from the dispatcher straight back to the user; each backend already produces minimal stdout in the standardized format.
compare.sh prints a row per backend with the standardized summary line. Each backend produces a .copilot-runs/<id>.usage.json so you can drill in later. See README.md for what each backend can and can't track.
cliclaude-opus-4.7 (override: --model, env COPILOT_MODEL)xhigh (CLI), high (ACP/SDK — they cap at high)SKILL.md — this file.README.md — full comparison of cli vs acp vs sdk (what each tracks, when to choose which, known gaps).compare.sh — runs a task across all three backends and prints a table.usage.sh — walks .copilot-runs/*.usage.json and prints a cost-per-run table.route.sh — the actual dispatcher; called by the skill on every invocation.Delegate to GitHub Copilot CLI via its Agent Client Protocol server. Built on the official agent-client-protocol Python SDK (Pydantic-typed messages). Sibling of /copilot-cli (JSONL) and /copilot-sdk (Go).
Delegate to GitHub Copilot CLI via its `--output-format json` stream. Python runner parses JSONL events into a structured transcript with full event replay. Default model claude-opus-4.7 + xhigh reasoning. Pass --model / --effort to override; --multi to fan out across models.
Delegate to Copilot via the official Go SDK (github.com/github/copilot-sdk/go). Source ships under scripts/; first invocation builds bin/runner. Sibling of /copilot-cli (JSONL) and /copilot-acp (ACP).