The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the @google/gemini-cli package and several extensions from GitHub repositories belonging to the gemini-cli-extensions organization if they are missing from the environment.\n- [REMOTE_CODE_EXECUTION]: The skill supports the installation and execution of external CLI extensions from GitHub. It also utilizes remote analysis through external AI services (Codex and Gemini), which may execute code within their own sandboxed or managed environments during the evaluation process.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform repository analysis with git commands (such as diff, ls-files, and log) and to interact with the gemini command-line interface.\n- [DATA_EXFILTRATION]: Local project data, including source code files, git diffs, and development guidelines like CLAUDE.md, are sent to external third-party services (OpenAI and Google) for processing, which is the primary intended function of the skill.\n- [PROMPT_INJECTION]: The skill accepts user-provided arguments that can override its internal workflow and processes external data (files and code) that might contain malicious instructions. It uses XML tags as a security measure to isolate and delimit this untrusted content when communicating with the external models.

second-opinions