/batch

Orchestrates large changes in parallel across the entire codebase: researches, splits the work into units, and has subagents implement them.

Orchestrates large changes in parallel across the entire codebase: researches, splits the work into units, and has subagents implement them. [Skill] Breaks the work into 5-30 independent units and presents a plan; once approved, a background subagent starts per unit in an isolated git worktree, implements, tests, and opens a pull request. Requires a git repository.

✅ WHEN TO USE IT?

A large change that splits into many independent units across the whole codebase

/batch rename the legacy HTTP client to the new SDK across all 15 services

A migration where each unit should get its own PR and be tested in parallel

/batch migrate every module from CommonJS to ESM

⛔ WHEN NOT TO?

A small task touching one or two files

the planning phase plus isolated worktrees and per-unit PRs are pure overhead for a small task

Better: use a normal prompt or a single subagent instead of /batch

No git repository present

/batch requires a git repository for the isolated per-unit worktrees

Better: run 'git init' first, or make the change manually without /batch

SOURCES

Claude Code Doku (offiziell) ↗ code.claude.com