The Agent Skills Directory

[SAFE]: The skill implements a structured reasoning protocol designed to improve the reliability of code refactoring tasks by requiring the agent to attempt to refute its own completion claims. This acts as a defensive measure against accidental bugs.\n- [COMMAND_EXECUTION]: Examples and methodology suggest using standard tools like grep, node, and curl for verifying local repository state and running test code. These operations are restricted to the context of the refactoring task and do not pose an external security risk.\n- [SAFE]: The included Node.js script scripts/checklist.mjs is a simple, zero-dependency utility used only to output the skill's checklist in JSON format. It contains no harmful logic.

refactor-adversarial-verify