メニュー
Full autonomous engineering workflow
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Full autonomous engineering workflow
Session kickoff protocol - the opposite of "land the plane". Use this skill whenever the user says "takeoff", "take off", "/takeoff", "starting a new session", "what should I work on", "kickoff", "what's next", or wants a prioritized view of the backlog at the start of work. Surfaces the top-priority backlog tasks as a table with status, dependencies, and blockers pulled from Backlog.md (CLI + MCP). Use this at the start of a coding session to orient on what to work on.
Produce two independent agent reviews on the current PR (an existing Copilot review when present, otherwise two distinct subagent reviewers), adjudicate findings with a third independent subagent, post inline PR comments for verified issues, run a tight inline fix-and-reply-and-resolve loop, then deliver a verdict — APPROVE if clean, APPROVE-WITH-CHANGES if findings were verified (and fixed or noted), CLOSE if the PR is fundamentally unsound, or WITHHELD if CI is red post-fix and the regression cannot be resolved. The skill is read-only with respect to GitHub reviewer assignment — it never adds reviewers. Use whenever the user invokes /ghcp-review-resolve, asks to "run copilot review and resolve", asks to "review and fix my PR with copilot", asks for a "dual review and fix pass", or wants automated review triage, remediation, and verdict on a pull request they just opened. Submits APPROVE or CLOSE; never submits REQUEST_CHANGES and never merges.
Session completion protocol - the opposite of "takeoff". Use this skill whenever the user says "land", "/land", "land the plane", "land it", "let's land", "land this", "bring it in", "wrap it up", "land the plan", "land plane", "time to land", "ok land", "go ahead and land", or any variation that signals they want to finish, close out, ship, or wrap up the current session's work. Executes the full commit → push → PR → handoff checklist without asking. If the user's message contains "land" in the context of finishing work, invoke this skill — it is a keyword trigger, not an exact match.
Full autonomous engineering workflow using swarm mode for parallel execution
Full autonomous engineering workflow using swarm mode for parallel execution
SOC 職業分類に基づく
| name | lfg |
| description | Full autonomous engineering workflow |
| argument-hint | [feature description] |
| disable-model-invocation | true |
CRITICAL: You MUST execute every step below IN ORDER. Do NOT skip any required step. Do NOT jump ahead to coding or implementation. The plan phase (step 2) MUST be completed and verified BEFORE any work begins. Violating this order produces bad output.
This workflow is resumable. A tiny helper tracks which phases are done and where each phase's output lives, so re-invoking /lfg continues from the first unfinished phase instead of restarting.
node .github/hooks/scripts/lfg-state.js — commands init, bind-plan, done <phase> --run-id <id> [--artifact <repo-relative-path>], status --run-id <id>, run-id-from-plan --plan <path>. It writes .atv/runs/<run-id>/ (gitignored, local-only).docs/plans/, derive RUN_ID from it: node .github/hooks/scripts/lfg-state.js run-id-from-plan --plan <plan-path>. Otherwise create a provisional id: run node .github/hooks/scripts/lfg-state.js init --skill lfg --feature "$ARGUMENTS" --repo <repo> --branch <branch> and read run_id.node .github/hooks/scripts/lfg-state.js status --run-id <RUN_ID> and skip every phase whose sentinel is already done; resume at the first not-done phase.node .github/hooks/scripts/lfg-state.js done <phase> --run-id <RUN_ID> [--artifact <path>] (pass the artifact path the phase produced; omit when it produced none).run:<RUN_ID> to every sub-skill so artifacts co-locate and downstream phases read paths, not full content..atv/runs/ is absent, infer progress from the plan, branch, and any open PR. Non-done phases are re-entered from the start, so ce-work MUST be called with mode:orchestrated to reconcile existing work instead of duplicating commits/PRs.Optional: If the ralph-loop skill is available, run /ralph-loop-ralph-loop "finish all slash commands" --completion-promise "DONE". If not available or it fails, skip and continue to step 2 immediately.
/ce-plan $ARGUMENTS run:<RUN_ID>
GATE: STOP. Verify that the ce-plan workflow produced a plan file in docs/plans/. If no plan file was created, run /ce-plan $ARGUMENTS run:<RUN_ID> again. Do NOT proceed to step 3 until a written plan exists. Record the plan file path, then bind it and mark the phase done: node .github/hooks/scripts/lfg-state.js bind-plan --run-id <RUN_ID> --plan <plan-path> and node .github/hooks/scripts/lfg-state.js done ce-plan --run-id <RUN_ID> --artifact <plan-path>.
/ce-work mode:orchestrated plan:<plan-path-from-step-2> run:<RUN_ID>
GATE: STOP. Verify that implementation work was performed - files were created or modified beyond the plan. Do NOT proceed to step 4 if no code changes were made. Then node .github/hooks/scripts/lfg-state.js done ce-work --run-id <RUN_ID>.
/ce-review mode:autofix plan:<plan-path-from-step-2> run:<RUN_ID>
Pass the plan file path from step 2 so ce-review can verify requirements completeness. Then node .github/hooks/scripts/lfg-state.js done ce-review --run-id <RUN_ID> --artifact <review-artifact-path>.
/unslop fix
Strip AI slop after review fixes land — removes commented-out code, filler comments, stale TODOs. Then node .github/hooks/scripts/lfg-state.js done unslop --run-id <RUN_ID>.
/observe on the areas of code that were changed — analyze patterns in the modified files to capture what was done and how.
/learn — extract reusable patterns from this session into project instincts.
/compound-engineering-todo-resolve — then node .github/hooks/scripts/lfg-state.js done todo-resolve --run-id <RUN_ID>
/compound-engineering-test-browser — then node .github/hooks/scripts/lfg-state.js done test-browser --run-id <RUN_ID> --artifact <report-path>
/compound-engineering-feature-video — then node .github/hooks/scripts/lfg-state.js done feature-video --run-id <RUN_ID>
Output <promise>DONE</promise> when video is in PR
Start with step 2 now (or step 1 if ralph-loop is available). Remember: plan FIRST, then work. Never skip the plan.