autopilotlisted

Use to run the board autonomously — let the AI advance through SEVERAL ready cards on its own, each as an independent PR off main, guided by the blocker graph, without stacked PRs. Trigger when the user says "run the board", "execute the board", "advance on your own", "knock out the ready tasks", "autopilot", or asks the AI to work multiple cards in one go. It maps the dependency graph, settles every ready card's open decisions with the user UP FRONT (so execution never guesses), asks sequential-vs-parallel, then runs each ready card through the `implement` lifecycle + `review` gate on its own branch, opens a PR, and STOPS when only PR-dependent or blocked work remains — never merging to main itself. To execute ONE specific card interactively, use `implement`; to plan a new demand, use `plan`.
fmilioni/claude-organizer · ★ 4 · API & Backend · score 77

Install: claude install-skill fmilioni/claude-organizer

# Running the board autonomously This skill is the **board-level orchestrator**. Where `implement` walks **one** card through its lifecycle with the user validating each step, `autopilot` advances through **many** ready cards in one go — each as an **independent PR off `main`** — using the board's **blocker graph** to know what can run now and what must wait. The methodology is **trunk-based, dependency-aware**: no stacked PRs, no cascade rebases, no coupled merges. It does not replace `implement` and `review` — it **drives** them per card. Its own job is the orchestration around them: the dependency graph, the up-front decision sweep, the two execution modes, and knowing when to stop. <SKILL-GATE> **Load the `claude-organizer` panorama first.** This skill assumes you are oriented on the board. If you have **not** already loaded the **`claude-organizer`** skill in this conversation, invoke it now (Skill tool) and run its start-of-session orientation **before** anything below. If it is already loaded in this conversation, don't reload it — just continue. Don't enter this skill cold. </SKILL-GATE> <HARD-GATE> Four invariants hold for **every** card, in both modes. Breaking one is a defect, not an optimization: 1. **Never stack.** A branch only ever starts from an up-to-date `main`. You do **not** start a card whose blocker isn't **merged into `main`** yet (not merely `done`/PR-open — see _Detecting "merged into main"_). 2. **The AI never merges to `main`.** In autonomous m