auto-pilot — Claude Code and Codex plugin: what it does
auto-pilot turns a short Claude request into a one-question-at-a-time intake, saves the answers as project files, and lets a later session pick up where the last one stopped.
MIT ★ 1,596 updated 2026-10-29
/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install auto-pilot@awesome-codex-pluginsWhat auto-pilot does
Auto Pilot adds an intake stage to Claude. A one-line request becomes a list of questions asked one at a time, and the answers turn into a written project contract. Claude then executes from that contract instead of re-deriving context every session.
At the start it writes autopilot/state.json, autopilot/blockers.json, autopilot/secrets-status.json and docs/next.md. Once intake finishes you get docs/spec.md and docs/progress.md, plus docs/design.md for anything with a user interface. Missing credentials for auth or a managed service stop the run in a setup-secrets phase: it asks for every value in one payload and leaves a .env.example behind.
The stable install is a shell one-liner that needs curl, tar and python3 on the machine; without them it stops. Restart Claude once afterwards or the slash command stays invisible. Claude gets the plugin from the marketplace instead. Interrupting work is fine, because the next session reads the saved files.
What it helps with
- turn a one-line project idea into a written brief
- resume a Claude session from saved state
- gather missing env values in a single pass
- fix the design direction before UI work starts
- mark the blockers that need a human
Example prompts for the agent
/auto-pilot:autopilot Build a budgeting app for freelancers- Build a diary app for a friend who writes every day ap
- Pick up the project in
/tmp/my-projectfromautopilot/state.json
How to install auto-pilot
Plugin marketplace
/plugin marketplace add hashgraph-online/awesome-codex-plugins/plugin install auto-pilot@awesome-codex-pluginsStep by step, with commands for each platform: Claude Code plugins: marketplaces, installing, and building your own
Questions about auto-pilot
How do I install auto-pilot in Claude Code?
Run /plugin marketplace add hashgraph-online/awesome-codex-plugins, then /plugin install auto-pilot@awesome-codex-plugins.
What does auto-pilot do?
It interviews you one question at a time, turns the answers into a project contract, and keeps state files so the next session continues from them. Blockers and missing credentials go into their own files.
What does the Claude install require?
The stable v0.1.1 installs from a shell one-liner that expects curl, tar and python3. Restart Claude once after installing so /auto-pilot:autopilot loads.
Source text
auto-pilot · MIT
# Auto Pilot <p align="center"> <img src="./assets/auto-pilot.png" alt="Auto Pilot mascot" width="260" /> </p> > Ask once. Lock the brief. Keep shipping. The mascot was created with inspiration from the hero icon style introduced by the creator of [oh-my-codex](https://github.com/Yeachan-Heo/oh-my-codex).     English | [한국어](./README.ko.md) | [日本語](./README.ja.md) ## Install in One Line Stable `v0.1.1` currently expects `curl`, `tar`, and `python3` and keeps the published slash-command entry point: ```bash curl -fsSL https://raw.githubusercontent.com/minsu0707/auto-pilot/v0.1.1/install.sh | bash ``` To preview the next release line on `develop`, install it explicitly: ```bash curl -fsSL https://raw.githubusercontent.com/minsu0707/auto-pilot/develop/install.sh | env -u NO_COLOR bash ``` Use `v0.1.1` for the currently published release. Use `develop` only when you intentionally want unreleased behavior for testing. When the next stable tag is created, update the stable install URL, run examples, and baked-in stable ref together. ## Run in Codex ### Stable `v0.1.1` ```text /auto-pilot:autopilot Build a budgeting app for freelancers ``` Use `/auto-pilot:autopilot` as the stable public entry point. Restart Codex once after installation so the slash command is loaded. Install once, then route to intake for new projects and resume automatically for existing ones. After restart, you can also use the natural-language shortcut `Build a budgeting app for freelancers ap`. ### Preview on `develop` ```text $auto-pilot Build a diary app my friend Dohyeon would love ``` Use `$auto-pilot` only if you installed from `develop`. After restart, you can also use the natural-language shortcut `Build a diary app my friend Dohyeon would love ap`. The sections below describe the current repository workflow on `main`, including features that will ship in the next stable release. ## What It Is Auto Pilot is a Codex plugin for turning a short product request into a resumable execution workflow. - Ask for the minimum important inputs once - Save a reusable project contract - Keep shipping until a real human-only blocker appears - Resume later from saved state instead of rediscovering context ## Why It Exists Short prompts are convenient, but they are usually too weak for long autonomous work. Auto Pilot closes that gap by adding: - one-question-at-a-time intake - explicit definition of done - a default blocker policy - resumable project state The goal is simple: less babysitting, more forward motion. ## Core Features - Converts a short project request into a structured intake session - Uses a `1. Question` / `Questions remaining: N` interaction pattern - Inserts an upfront integration setup phase when auth or managed services need env values - Writes `autopilot/state.json`, `autopilot/blockers.json`, `autopilot/secrets-status.json`, and `docs/next.md` as the initial runtime contract - Writes `docs/spec.md` and `docs/progress.md` once setup is complete - Adds `docs/design.md` for user-facing projects so UI work starts from a concrete design brief instead of generic defaults - Keeps enough state for the next Codex session to resume from where it stopped - Keeps design research scoped to a brief and a curated reference stack instead of pretending broad web research already happened - Keeps canonical plugin code, docs, and installer logic at the repository root ## Quick Start These CLI examples are for plugin development or local debugging. Run them from the Auto Pilot repository root or the installed plugin directory. If you are using Auto Pilot inside Codex, prefer `$auto-pilot` or `/auto-pilot:autopilot` and ignore this CLI section. Start a new intake session: ```bash ./scripts/autopilot start \ --workspace /tmp/my-project \ --prompt "Build a diary app my friend Dohyeon would love" ``` Answer the current question: ```bash ./scripts/autopilot answer \ --workspace /tmp/my-project \ --text "Teenagers who want a cozy, private diary they can decorate every day" ``` Check the current mode and status: ```bash ./scripts/autopilot status \ --workspace /tmp/my-project ``` If the project needs upfront integration env values, submit them in one payload: ```bash ./scripts/autopilot secrets \ --workspace /tmp/my-project \ --text 'GOOGLE_CLIENT_ID=...' ``` After the final answer, Auto Pilot follows one of two paths. If setup is already complete, or no upfront env values are needed, it generates: - `docs/spec.md` - `docs/progress.md` - `docs/next.md` - `docs/design.md` for user-facing projects - `autopilot/state.json` - `autopilot/blockers.json` - `autopilot/secrets-status.json` If required env values are still missing, it moves into `setup-secrets` and writes: - `docs/next.md` - `autopilot/state.json` - `autopilot/blockers.json` - `autopilot/secrets-status.json` - `.env.example` If setup is still pending, that project still counts as an existing Auto Pilot project because `autopilot/state.json` already exists. ## How It Works 1. A short prompt starts intake. 2. Auto Pilot asks one question at a time. 3. The answers are normalized into a project contract. 4. If required env values are missing, Auto Pilot pauses in a single `setup-secrets` phase and asks for them in one consolidated payload. 5. User-facing projects also get a `docs/design.md` brief built from the selected theme, vibe, and design direction. 6. Runtime state is created for future execution and resume. 7. Codex can continue from saved files instead of rediscovering context. ## Repository Layout - `docs/01-product-brief.md`: product overview - `docs/02-prd.md`: functional requirements and operating model - `docs/03-plugin-spec.md`: plugin structure and state model - `docs/04-mvp-roadmap.md`: MVP implementation sequence - `docs/05-how-it-works.md`: visual walkthrough of intake, state, and resume flow - `docs/06-usage-guide.md`: detailed install, run, resume, CLI, and team-usage guide - `docs/i18n/*`: localized README files - `assets/auto-pilot.png`: mascot image - `.codex-plugin/plugin.json`: Codex plugin manifest - `commands/autopilot.md`: secondary plugin command entry point - `skills/auto-pilot/SKILL.md`: main orchestration skill - `skills/autopilot-intake/SKILL.md`: one-question-at-a-time intake skill - `skills/autopilot-resume/SKILL.md`: resume skill - `scripts/autopilot`: recommended CLI entry point - `scripts/autopilot.py`: Python backend for the CLI wrapper - `scripts/*.py`: intake, answer recording, and status scripts - `templates/*.json`: state templates - `install.sh`: canonical installer - `uninstall.sh`: canonical uninstaller ## Current Status This repository root is the canonical plugin root, docs home, and installer home. For end users, the recommended flow is the one-line installer, one Codex restart, then `$auto-pilot` inside Codex. For a fuller walkthrough, start with `docs/06-usage-guide.md`.
© Minsu · MIT · hashgraph-online/awesome-codex-plugins