The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands including git, ast-grep, and project-specific build/check scripts (e.g., npm install, lint:comments). These are used for project analysis and verification in a temporary worktree. Resource usage is managed using nice -n 15 to maintain system responsiveness.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes existing code comments to classify them and plan their removal. Maliciously crafted comments within the project's source code could potentially attempt to influence the agent's behavior or decision-making during the classification phase.
Ingestion points: Project source code files (comments extracted via ast-grep).
Boundary markers: No specific delimiters or "ignore instructions" warnings are mentioned for the natural language content of individual comments being processed.
Capability inventory: File system write access (plans, patches), command execution (git operations, project-specific build and test scripts).
Sanitization: The skill relies on AST parsing and a custom verifier script (verify-strip-js.mjs) to ensure code logic is unchanged, but does not explicitly sanitize the natural language content of comments before processing.

cleanup-comments