direct-implementation

Use the direct-implementation workflow: describe the change clearly, let the agent locate and read the target file, apply a minimal SEARCH/REPLACE edit, then run the linter and relevant tests before presenting a brief summary.

Use it when the intention is a simple change or bug fix with a known cause, the scope is a single file, the solution is known, and requirements are precise. If the task spans multiple files or has unclear requirements, choose a planning-oriented workflow instead.

The typical budget is 2K-10K tokens with a recommended maximum of 20K. Exceeding 50K tokens is a warning signal that the task likely belongs in a different, more thorough workflow.

The workflow runs the linter and relevant tests after editing, then fixes any introduced errors in up to three cycles. If errors persist beyond that, the change fails the sanity check and must be stopped and corrected before presentation.

It is unsuitable for multi-file changes, ambiguous requirements, unknown root causes, or high-risk modifications. It also forbids editing files that have not been read and refactoring unrelated code outside the request scope.