Guide · Agent skill
How to use the Autopilot agent skill: idea to working code
Updated · written with AI from the project's own documentation and checked against our catalogue data
Autopilot is an agent skill that tells an AI coding agent to take a short description of a software task and carry it through the whole job on its own: requirements, design, planning, implementation, QA and validation. This guide explains what the skill does, how to install it from the ZIP on this page, how to trigger it with plain requests, and what to check before trusting the output. By the end you will know whether it fits a small-to-medium build you want to hand off.
Autopilot: Idea to Working CodeYeachan-Heo/oh-my-claudecode · MIT Download the skill Quick start
- 1Download the ZIP. Download the autopilot ZIP from this page. It contains SKILL.md.
- 2Unzip into a skills folder. Unzip the package into ~/.claude/skills/ for all projects, or into a .claude/skills/ folder inside one project.
- 3Check the requirements. Confirm you have Claude Code or a SKILL.md-compatible agent with the oh-my-claudecode plugin installed.
- 4Describe the task. Write two or three lines about what you want built, including any concrete file or function names you know.
- 5Run the skill. Invoke /autopilot with your description, or make a natural request such as "build me" followed by the task.
- 6Review the output. Read the generated spec, plan and code and run your own checks before using the result.
What the autopilot skill makes the agent do
Autopilot is designed for hands-off runs. You describe a product idea or task in two or three lines, and the agent works through phases: expanding the idea into a detailed spec, technical design, planning, parallel implementation where possible, QA cycles, and multi-perspective validation before it reports back.
The skill is built to execute, not to discuss. If you want to explore options, the source says to use a planning skill instead.
- Phase 0: turn the idea into a spec (written to .omc/autopilot/spec.md)
- Phase 1: create an implementation plan (written to .omc/plans/autopilot-impl.md)
- Phase 2: implement with executor agents, in parallel where possible
- QA cycles repeat up to 5 times; if the same error persists 3 times, the agent stops and reports the fundamental issue
- Validation needs approval from all reviewers; rejected items are fixed and re-validated
Install the autopilot skill
Download the ZIP from this page. The package contains a single file, SKILL.md. Unzip it so the skill folder contains SKILL.md, then place the folder in one of these locations:
You can also install it in the Claude apps by uploading it under Settings → Capabilities → Skills.
- For all your projects: ~/.claude/skills/
- For one project only: a .claude/skills/ folder inside the project
Requirements before you run it
The skill needs Claude Code or another SKILL.md-compatible agent, with the oh-my-claudecode plugin installed. Without that plugin, the phase workflow it describes is not available.
Some features are optional. Named --workflow profiles need Linux with the flock utility. A companyContext MCP tool and the OMC_RUN_BUDGET_TOKENS environment variable only matter if you set them up. The skill runs scripts and does not use the network, according to its listing.
- Required: Claude Code or a SKILL.md-compatible agent, plus the oh-my-claudecode plugin
- Optional: Linux with flock, for named --workflow profiles
- Optional: a companyContext MCP tool and OMC_RUN_BUDGET_TOKENS, if you use those features
Trigger autopilot with a natural request
You can invoke the skill with its command and a description, using the argument hint format: /autopilot followed by your task. The skill also activates on phrases such as "autopilot", "autonomous", "build me", "create me", "make me", "full auto", "handle it all", or "I want a/an...".
- Good: "Build me a small habit tracker web app with a login page and a weekly chart."
- Good: "Autopilot: a CLI tool that converts CSV files to JSON, with tests."
- Not a fit: "What are my options for storing habit data?" (use a planning skill)
- Not a fit: "Fix the typo on line 12." (a direct fix is faster)
/autopilot <product idea or task description>Use a named workflow profile (optional)
To run a configured stage schedule, pass --workflow with a profile name. Profiles are defined in project or user JSONC under autopilot.workflows. In v1 the admitted stage sequences are limited to the four listed in the source, such as [ralplan, execution, qa]. Without --workflow, autopilot uses its legacy lifecycle.
If you are not on Linux with flock, explicit --workflow activation is rejected before any state changes. Use the legacy autopilot in that case.
/autopilot --workflow plan-build-qa <task>Safety notes: scripts and review
The skill runs scripts, so read SKILL.md and the project's configuration before running it on code you care about. Autonomous runs can produce code that still needs careful checking, so the skill is not suited to production-critical code without human review.
You can cancel at any time with /oh-my-claudecode:cancel. Before cleanup, the agent runs the Phase 5 closeout and keeps resumable progress artifacts. Budget stops, when you have set OMC_RUN_BUDGET_TOKENS, are a stop condition rather than a failure and keep state for resume.
- Review every generated file before merging or deploying
- Set OMC_RUN_BUDGET_TOKENS if you want a token cap on the run
- Use /oh-my-claudecode:cancel to stop a run
Tips for better results
Give a clear, bounded brief. The skill is aimed at small-to-medium builds, and concrete anchors such as file paths or function names help. If the input is vague, the skill may offer to redirect you to /deep-interview for clarifying questions first.
- Describe the goal, the users and the must-have features in two or three lines
- Mention existing file paths or function names when you have them
- Use a planning skill first if you are still deciding what to build
What people use it for
- Prototyping a small web app or tool from a short description and reviewing the result afterwards
- Handing off a small-to-medium build so the agent implements, tests and validates it
- Teams already using the oh-my-claudecode plugin who want a repeatable automated workflow with a defined stage order
Frequently asked questions
How do I install the autopilot skill?
Download the ZIP from this page and unzip it so the folder contains SKILL.md. Place it in ~/.claude/skills/ for all projects or in a .claude/skills/ folder inside one project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Is autopilot free to use?
The source material for this listing does not state pricing or licence terms beyond the MIT licence shown on the item. Check the repository's licence file for the terms that apply to the code.
What do I need to run autopilot?
You need Claude Code or another SKILL.md-compatible agent with the oh-my-claudecode plugin installed. Named --workflow profiles also need Linux with the flock utility, and the companyContext tool and budget variable are optional.
Does autopilot work on Windows or macOS?
Standard autopilot uses its legacy lifecycle without --workflow. The named workflow profiles require Linux with flock, and the source says unsupported environments reject explicit --workflow activation, so use legacy autopilot there.
Is autopilot the same as a planning skill?
No. Autopilot executes the whole task, while the source says to use the plan skill when you want to explore options or brainstorm, and plan --review to critique an existing plan.
Why did autopilot stop before finishing?
A run can stop when the same QA error persists three times, when you cancel it, or when a token budget set through OMC_RUN_BUDGET_TOKENS is reached. The source treats a budget stop as a stop condition rather than a failure, and progress is kept so the run can resume.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Autopilot suits developers and solo builders who want to hand off a small-to-medium build and review the result afterwards, especially if they already use the oh-my-claudecode plugin. Skip it for brainstorming, one-line fixes, or production-critical code that will not get careful human review.
Download the skill