The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute the gemini CLI. It constructs command strings based on user intent. This standard tool usage is secured by instructions to use non-interactive mode and mandatory human-in-the-loop confirmation before any suggested actions are implemented.
[PROMPT_INJECTION]: The skill is designed to analyze external files (code and documentation), which introduces a potential surface for indirect prompt injection. 1. Ingestion points: Local codebase files and documentation read via the Read tool. 2. Boundary markers: None explicitly defined within the generated prompt strings. 3. Capability inventory: Bash, Read, and Write tools used to run analysis and potentially apply fixes. 4. Sanitization: The risk is managed through strict process rules requiring the agent to summarize findings for user review rather than executing suggestions automatically.

gemini