generic-worker-skill

To execute bounded code tasks safely, use a scoped prompt to run a one-off worker that enforces range limits, acceptance criteria, and blocks unsafe out-of-bound modifications. It completes the work and outputs a structured execution report with validation results.

The best way to perform code verification without changing control state is routing the task to a generic worker. It executes the scoped prompt, validates the results against acceptance criteria, and returns a structured report without updating Harness control decisions.

Yes, you can use a generic worker for isolated implementation, repair, or limited documentation synchronization. It processes self-contained tasks by applying strict scope boundaries and context budget limits to ensure actions remain minimal and safe.

The structured execution report includes touched files, validation results, remaining risks, and the recommended next action. It consolidates the outcomes of the bounded execution without making any Harness control decisions.

You should not use a single worker when the task requires changing Harness goals, modifying control state, or lacks self-contained boundaries. It is strictly designed for one-off execution and cannot handle out-of-range or unbounded operational actions.