V2 preview · open for review

Give direction.
Keep moving.

A reusable mission for your AI agent. Carry code, research, and documents forward with clear scope, checked outcomes, and a handoff you can inspect.

01 / MISSION

Why add Autopilot?

Native goals already keep an agent working. Autopilot packages the mission and its working record so you can reuse the brief, inspect progress, and pick up where you left off.

01

A brief that stays useful

Keep the outcome, allowed scope, and completion checks together.

02

Evidence at each checkpoint

Record what was actually checked, against which version, and what remains unknown.

03

A handoff that preserves work

Inspect existing edits and partial results before continuing an interrupted mission.

02 / IN PRACTICE

One mission. Your kind of work.

Use your existing agent and tools. Choose the outcome; let the agent work within the agreed boundaries.

$autopilot Fix the import failure without changing the public API. Reproduce it, run the integration tests, and preserve my edits. Use native goal mode to continue until complete.

03 / EVIDENCE

Evidence before promises.

We compare actual artifacts and execution cost against native goals. Short fixtures can establish basic behavior; they cannot establish long-running reliability or time saved.

No comparative advantage established

Read the pilot results, exact conditions, and limitations before choosing the workflow.

Read the evidence →

Use it where the record matters.

For a clear one-off task, a native goal may be enough. The skill is not a scheduler, a sandbox, or an independent verifier. Continued execution depends on the host.

04 / START

Install the v2 preview

The default branch remains v1.3 while v2 is reviewed. Install this branch to try the behavior described here.

Terminal
git clone --branch improve/codex-missions-v2 https://github.com/PresentJay/autopilot-skills.git autopilot-skills-v2
npx skills add ./autopilot-skills-v2 --skill autopilot

Codex $autopilot · Claude Code /autopilot

05 / BUILD IN PUBLIC

Help make it earn its place.

Bring a mission that failed, a better verification method, or a measured simplification. Code, research, and document contributions are equally welcome.