copilot) as a background implementer, then\nreview its diff and land it yourself. Use this whenever the user wants to delegate implementation\nwork to Copilot - phrasings like \"have Copilot implement X\", \"delegate this to copilot\", \"run it\nthrough Copilot CLI\", or \"use copilot to implement/fix/refactor\" - or wants to run a queue of\ncoding tasks through Copilot while staying the reviewer. DO NOT USE for tasks small enough to do\ninline, or when the user wants the code written directly without delegating.\nlicense: MIT\ncompatibility: Requires the copilot CLI installed and authenticated (copilot login), Node 18+ to run the relay (the copilot CLI itself requires Node 22+), and git. The orchestrator must be able to run shell commands and read files.\nmetadata:\nversion: 0.5.0You are the orchestrator. Delegate a bounded coding task to a separate implementer — the\nGitHub Copilot CLI — then review what it produced and land it yourself. You write the brief and own\nthe judgment; the implementer makes changes in its own session in a clean working tree; you verify\nand commit.
\nThe loop needs only a shell command and file access, so any comparable orchestrator can drive it.
\ncopilot CLI is not installed or authenticated./sandbox command and settings, disabled by default) — this relay\ndoes not configure them. --read-only only disables edit tools (--mode plan); shell commands\nstill run. If project files must not change at all, dispatch against a clean or isolated worktree.copilot (npm install -g @github/copilot; the CLI requires Node 22+, the relay\nitself runs on Node 18+ — the relay probes copilot version).copilot login (interactive web/device flow), or set\nCOPILOT_GITHUB_TOKEN / GH_TOKEN / GITHUB_TOKEN in the environment.copilot version succeeds.--cd at, the target git repository.Copilot picks a default model (auto). To choose another, pass --model <name>.\nThe relay accepts letters, digits, and . _ : / - only (the value reaches a shell on Windows).
Copilot supports a reasoning effort dial: --effort <level> with values\nlow, medium, high, xhigh, or max. The relay rejects any other value\nbefore dispatch.
Run these five steps per task. Steps 1, 4, and 5 require judgment; 2 and 3 are mechanical.
\nCopilot sees only the text you send. It cannot read your conversation: the brief must stand alone\nwith the goal, current state, what to change, what to leave untouched, the project's real\ngates, and a report contract. Keep each brief to a single task. Write it to a file and pass it as\nthe relay's --brief. See references/writing-the-brief.md.
Use the bundled relay. It runs copilot -p with --output-format json --no-color --stream off,\ncaptures the JSONL event stream, and writes result.json.
node \"<skill-dir>/scripts/relay.mjs\" --brief brief.txt --cd /path/to/repo\n# choose a model: add --model <name>\n# set reasoning effort: add --effort <level>\n# read-only planning pass: add --read-only (forces --mode plan)\n# full tool autonomy: add --allow-all-tools\n# hard time limit (watchdog): add --timeout 2h (the 30m default suits brief runs)\n# resume a session: add --session <id> or --resume-last\n# see all options: node .../relay.mjs --help\nThe child's cwd pins the workspace. The relay writes artifacts under the system temp dir by\ndefault and never commits. See references/dispatch-and-poll.md.
\nThe relay blocks until copilot finishes. Run it with the orchestrator's background-command\nfacility, or background it in the shell and poll for result.json. A pre-run usage error exits 2\nand writes no result; a missing copilot exits 127 and writes status: \"copilot_unavailable\".
Completion means the process exited and result.json exists — trust process state and the\nworking tree, not the progress display. Copilot's final assistant message is the finalMessage\nfield of result.json.
touchedFiles.See references/review-and-land.md.
\nIf the work is good, commit it. The relay never commits — the diff and result.json are the\nrecord; run git status and git diff first to confirm exactly what changed. If the group has a\nPR flow, make the commit and push a branch; let human review happen. If the diff is wrong or\nincomplete, re-dispatch a corrected brief in a fresh run and review again.
Without --allow-all-tools, copilot auto-denies tool calls in headless mode: the process exits 0\nbut the relay detects the denial events and reports status: \"failed\" with the CLI's own error\nmessage and a hint to pass --allow-all-tools. This is the honest default — the orchestrator sees\nthe failure rather than a silent no-op.
--allow-all-tools explicitly grants full tool autonomy. --read-only selects --mode plan,\nwhich disables edit tools so project files can't be changed by direct edits; it works without\n--allow-all-tools. Shell commands still run in plan mode, so it guards against edits, not\nagainst everything. The two flags are mutually exclusive.
Copilot also exposes sandbox controls, but they are upstream-experimental (MXC-based, controlled\nvia the /sandbox command and settings, disabled by default). This relay does not configure them.
Delegation is something the human opts into. Once briefed, copilot works as a tool you approved use\nof. The boundary is: do not accept conclusions from the self-report; verify everything on\ndisk. For anything touching credentials, production data, or irreversible operations, stop and ask\nthe human first instead of encoding it in a brief.
\nresult.json, and failure recovery.