The Agent Skills Directory

[COMMAND_EXECUTION]: Employs standard CLI utilities such as grep, rg (ripgrep), and npm to audit and modify project files locally. It also provides commands for vercel CLI to manage CDN caches, which require user authorization before execution.
[DYNAMIC_EXECUTION]: Includes a local utility script (scripts/write-transparent-favicon.js) to generate a placeholder icon. The script implements security best practices, including verifying that the target directory is not a symbolic link and ensuring paths do not resolve outside the physical project root.
[INDIRECT_PROMPT_INJECTION]: The skill processes project source code and configuration files, which constitutes a potential attack surface for indirect prompt injection. However, the risk is minimized as the skill primarily uses pattern-matching tools rather than passing raw content to the LLM for reasoning. \n
Ingestion points: Reads package.json, vite.config.ts, index.html, .env, and source files in src/ via grep and ripgrep.\n
Boundary markers: None explicitly defined for file content processing.\n
Capability inventory: Dependency removal (npm uninstall), file deletion, local script execution, and Vercel CDN cache purging.\n
Sanitization: None; the skill relies on the agent to interpret filtered grep output.
[EXTERNAL_DOWNLOADS]: Performs a non-invasive header check using curl against the user's own production domain to verify favicon deployment and cache settings.

lovable-cleanup