Worktrees & workers - tiny-brain

go

Creates a typed git worktree for a work item and launches Claude Code in it — one isolated working copy per fix, feature, or spike, branching from the remote default branch by default.

tiny-brain go --fix my-bug
tiny-brain go --prd checkout --feature guest-checkout
tiny-brain go --spike codex-headless --local     # branch off local HEAD instead

Worktrees land at .claude/worktrees/<type>/<id> on branch <type>/<id>. --skip-install skips dependency installation.

worktree

Manage the worktrees go (and workers) create:

tiny-brain worktree list
tiny-brain worktree remove <name>

For an ad-hoc worktree with no tracked work item: tiny-brain worktree create <type> <id>.

worker

A worker definition is a named recipe — adapter + target + model + effort — that run start consumes:

tiny-brain worker create fast-local --adapter claude-code \
  --target local-worktree --model claude-sonnet-4-6 --effort medium
tiny-brain worker list
tiny-brain worker probe fast-local    # the same preflight gate run start uses

--global writes to ~/.tiny-brain/workers.yaml instead of the repo; --timeout-ms sets the per-invocation deadline.

run

Launches autonomous worker runs against a work item, in isolated worktrees (or a cloud sandbox), and manages their lifecycle:

tiny-brain run start --fix my-bug --worker fast-local
tiny-brain run list --status running
tiny-brain run show <runId> --all          # full event history
tiny-brain run accept <runId> --strategy rebase
tiny-brain run start --fix my-bug --worker fast-local --resume   # pick up a failed run

accept merges a run’s output back with a chosen strategy (ff · merge · cherry-pick · rebase, default auto). Runs, their events, and accept/reject decisions are all visible on the dashboard.

​go

​worktree

​worker

​run

go

worktree

worker

run