The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing @google/gemini-cli via bun or npm. This is an official package from Google, a well-known technology company.
[COMMAND_EXECUTION]: The skill provides scripts (scripts/install-gemini-coach.sh, scripts/setup-slash-command.sh) that install local wrapper scripts to ~/.local/bin/ and configure Claude Code slash commands. These are standard developer setup actions for CLI tools.
[INDIRECT_PROMPT_INJECTION]: The skill ingests codebase files to be processed by an external LLM (Google Gemini). This represents an attack surface for indirect prompt injection where malicious content in a file could attempt to influence the external model's output. The skill mitigates this by instructing the agent to synthesize results rather than execute them.
Ingestion points: Reads file content via cat or the --all-files CLI flag (SKILL.md).
Boundary markers: Uses specific AI-to-AI headers like [Claude Code consulting Gemini for peer review] to scope the external model's context (references/prompting-strategies.md).
Capability inventory: Uses subprocess command execution to call the gemini CLI.
Sanitization: The skill instructions require the agent to synthesize and review Gemini's findings rather than providing raw output directly to the user or system.

gemini-cli