batch

Parallelize large content generation by decomposing requests into 2–10 non-overlapping independent units. Each unit executes in isolated worktrees and outputs to separate files, preventing cross-contamination and hidden dependency failures during parallel execution.

Worktree isolation runs one agent per decomposed unit in separate worktrees to prevent cross-contamination and maintain determinism. This ensures each parallel execution unit operates independently, producing its own output file without interfering with other running tasks.

Yes, you can batch repeated analyses into independent parallel sections when evaluation criteria are identical across items and units are non-overlapping. The skill enforces a mandatory independence check and requires user approval of the decomposition plan before execution starts.

If one unit fails during batch execution, other units continue running. The skill produces a batch summary report with explicit failure reporting, including error summaries and recommended actions, while successfully completed units still generate their separate output files.

Yes, a mandatory approval gate forces a user-visible decomposition plan showing estimated cost and parallelism before execution starts. You must approve the plan to proceed, ensuring you control the task decomposition and concurrency level.

Avoid parallel batching when task units overlap, require specific execution order, or have dependencies between them. The skill enforces a mandatory independence check and only suits non-overlapping, order-independent work like generating distinct document sections or separate competitor profiles.