The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard command-line tools to manage the refactoring process. It uses git for creating atomic commits and managing rollbacks, npm (or yarn/pnpm) for running test suites, and tsc for TypeScript type-checking. It also relies on an "oracle" command—a user-defined project-specific command—to verify code correctness throughout the process.
[REMOTE_CODE_EXECUTION]: To handle codebases without existing test coverage, the skill generates a temporary TypeScript scratchpad (characterize.ts) based on a provided template. It then executes this script locally using npx --no-install tsx to establish a behavioral baseline and verify that refactoring does not change outputs. This is a localized dynamic execution of code derived from the repository itself.
[PROMPT_INJECTION]: The skill processes repository source code, which represents an ingestion surface for indirect prompt injection. While malicious instructions could theoretically be embedded in code comments or test data, the skill's reliance on strict verification gates (automated tests and type-checking) and its prescriptive operational workflow serve as effective mitigations against such vectors.

refactor