The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted codebase data through automated agents with significant file system and shell capabilities. This creates a risk surface where malicious instructions embedded in source code could influence agent behavior during cleanup tasks.\n * Ingestion points: The skill accesses codebase files and directory structures via Read, Glob, and Grep tools.\n * Boundary markers: Lacks explicit delimiters or specific instructions for the agent to ignore embedded prompts within the source code being analyzed.\n * Capability inventory: Possesses significant capabilities including file modification via Write and Edit, and scoped command execution through Bash for tools like git, npm, npx, pnpm, python3, tsc, wc, and ls.\n * Sanitization: Codebase content is processed without explicit filtering for potential adversarial instructions.\n- [DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file executes shell commands at load time to gather repository state and project configuration.\n * Evidence: The skill uses git rev-parse --show-toplevel, ls package.json, and cat package.json to detect the environment and display context to the agent.\n- [REMOTE_CODE_EXECUTION]: The skill uses npx to dynamically download and execute various static analysis and cleanup tools from the NPM registry.\n * Evidence: The documentation and tools.md reference executing npx knip, npx madge, npx jscpd, and npx depcruise for code analysis.

cleanup-code