Quickstart
From a fresh install to a running build loop. New to patok? Read how it works first.
Before you start
- Install patok.
- Install an agent CLI and log in once. By default patok uses Claude Code
(
claude);codexandvibework too. - Use a git repository with at least one commit, on a branch you're happy to have commits on. Commit or stash your own changes first: the agents edit your working tree directly.
- Consider adding
SPEC.mdandTASKS.mdto your.gitignore, or patok will commit them with everything else.
1. Start patok and write a spec
cd my-project
patok
Write a SPEC.md in the project root with the business rules and architecture
every task should respect.
Not sure where to start? Press I (Shift+i) to open your agent CLI in the project
and ask it to draft the spec. Close the session with Ctrl+], then read the result
and fix anything wrong: every task will be held to it.
2. Create the task list
Press Enter at [ Enter ] create queue. The research agent
reads the spec and the project and writes the first TASKS.md. Review it and edit
anything you like, it's plain Markdown.
3. Start the loop
Press Enter again. patok works through the tasks one by one while the output
pane shows the agents live and the task list tracks progress. Press Esc to
stop; run patok again to resume where it left off.
If a task fails
The work is committed as WIP: …, the task stays unchecked and the loop stops.
Look at the last commit, fix the task text or the code, and start again. Pushing is off by
default, so nothing leaves your machine until you push.