Migrations split across parallel agents

Ask for the split you want, such as one agent per directory or package. Capy drafts a parallel agent for each part with its full prompt, starts them, and each one ships its own pull request. Parallel agents must touch separate parts of the codebase.

An agent that writes code boots a fresh machine with a clean checkout from the upstream branch, so two agents editing code can’t collide with each other or with the thread that started them.

Capy stacks them: the first agent ships its pull request, and the next starts from that branch and builds on top. With GitHub’s native stacks, each pull request shows a stack chip such as 2/3, and you merge from the bottom up.

CI results and conflicts with the base branch wake the thread that owns the pull request, which is the thread that last pushed to it. That thread fixes the failure or resolves the conflict and pushes again.

Yes. You can run 25+ agents on different tasks in the same repo at the same time, and each one pushes its own branch. Capy never pushes to the base branch, main, master, develop or release branches.

Yes. Each agent can run on its own model: a cheaper one for mechanical edits, a stronger one for a hard refactor, or a connected subscription like Codex. By default they use the thread’s model.

Use a Captain. It starts a thread for each part of the work, each with its own parallel agents, reads their reports, and tells you only what needs you.

Set up an automation such as a nightly dependency check: on a schedule like 0 3 * * *, Capy audits dependencies for security advisories and opens a pull request bumping anything vulnerable.