The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates by executing the gemini command-line utility. It provides detailed instructions for passing user prompts and project files as arguments to perform auxiliary processing.
[EXTERNAL_DOWNLOADS]: The skill utilizes the @google/gemini-cli package, which is the official tool from a well-known service provider. All installation and documentation links target official repositories and domains associated with Google and Vercel.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, which is a standard feature for its research and analysis purpose but represents a potential attack surface.
Ingestion points: External data enters the context through local file reading (read_file tool and @path syntax), web search results (google_web_search), and remote content fetching (web_fetch).
Boundary markers: The current prompt templates do not implement specific structural delimiters or "ignore instructions" markers when interpolating external content.
Capability inventory: The skill possesses the ability to execute shell commands (via the gemini tool), read local files, and perform network operations.
Sanitization: While the prompt templates do not sanitize incoming data, the skill documentation includes a robust 'Validation Pipeline' in patterns.md, recommending syntax checks, security scans (e.g., for XSS or eval usage), and functional testing for all generated output.

gemini-cli