Delegate bounded coding tasks to GitHub Copilot CLI and review

Copilot Delegate hands a bounded coding task to the GitHub Copilot CLI (copilot) as a separate implementer running in its own session in a clean working tree, while you stay the orchestrator writing the brief and reviewing the diff. Without --allow-all-tools, Copilot auto-denies tool calls in headless mode: the process exits 0 but the relay detects the denial events and reports status: "failed" with the CLI's own error message and a hint to pass --allow-all-tools, rather than silently no-op'ing. --allow-all-tools grants full tool autonomy and needs explicit human authorization for that run; --read-only selects --mode plan, which disables edit tools but still lets shell commands run, so it guards against edits, not against everything - the two flags are mutually exclusive. A relay runs copilot -p with --output-format json --no-color --stream off, captures the JSONL event stream, and writes a result.json; a pre-run usage error exits 2 with no result file, and a missing copilot binary exits 127 with status: "copilot_unavailable". Completion means the process exited and result.json exists with the implementer's final message in its finalMessage field - not the progress display. The relay never commits.

Use it to delegate a bounded coding task you want implemented and then review yourself, or when the user explicitly asks for delegation to Copilot. Do not use it when the task is small enough to do inline, when the copilot CLI is not installed or authenticated, or when you need a hard sandbox - Copilot's sandbox controls are upstream-experimental (MXC-based, via the /sandbox command and settings, disabled by default) and this relay does not configure them; dispatch against a clean or isolated worktree instead if project files must not change at all.

Input is a self-contained brief - Copilot cannot read your conversation - with the goal, current state, what to change, what to leave untouched, the project's real gate commands, and a report contract, written to a file and passed as --brief; one task per brief. Prerequisites checked once: install copilot (npm install -g @github/copilot, CLI needs Node 22+, the relay itself runs on Node 18+), authenticate via copilot login or a COPILOT_GITHUB_TOKEN/GH_TOKEN/GITHUB_TOKEN environment variable, and confirm copilot version succeeds. Output is a result.json with the implementer's finalMessage, touchedFiles, and a status field (failed, copilot_unavailable, etc).

Wraps the copilot CLI through a scripts/relay.mjs helper; that executable script is not bundled here - see the upstream project for the runtime. Optional flags select the model (--model <name>, restricted to letters, digits, and . _ : / -) and reasoning effort (--effort low|medium|high|xhigh|max), resume a prior run (--session <id> or --resume-last), or cap the run with --timeout (30-minute default). Adapted from amElnagdy/delegate-skills (MIT).

Anyone orchestrating with an agent that can run a shell command and access files, who wants to hand a bounded coding task to GitHub Copilot CLI as implementer, verify its diff and re-run the project's own gates, and land the reviewed result themselves rather than trusting Copilot's self-report - especially for anything touching credentials, production data, or irreversible operations, where the model calls for stopping to ask a human rather than encoding it in a brief.