Fire and steer

Most work does not need you watching it. In CodeAF you hand a job off with /task and a sentence, and it starts at once while your conversation stays free. The work splits into parts that run side by side, each finished part is checked, and you step in only to change its course.

Try it in CodeAF

How a run works

A run works in one copy of your folder, cut as it stands, uncommitted edits included, and every part shares that copy. A worker that finds its share wide splits it again. Ready parts run at once: there is no limit on how many unless you set one, and new parts wait only while the machine is too busy. When a part lands, a separate check reads it against its acceptance, and a check that does not hold becomes a fix: task. At the end the work is merged into your folder, or kept on its own branch when you are on a branch such as main that tasks never merge into, and the chat gets one line saying which.

Steering a running part

Every part in the Tasks column opens its room: click its row, or press Alt+T and walk to it. The box there reads a note for this task. What you type goes to that part's worker as a note, never to the chat's model, and the worker reads it at its next step. Over an empty box, X stops the part after asking first.

Things to build

• A test file for every module: one /task that says "one part per module" writes each module's tests side by side, a check reads each part, and you keep talking in the same chat.
• Twenty issues before lunch: one brief that lists every open issue, one part per issue, lands a single branch in about two minutes. Close twenty issues at once
• A correction for one part: say "note for the Wrap part only: ..." in the chat, and only that worker reads it, at its next step.
• Merge on green: when the run lands on its own branch, say "if go vet and go test pass on that branch, merge it into main", and the chat checks the branch and merges it.

Go further