The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill requires installing google-generativeai via pip. This is a trusted source (Google/Google-Gemini), downgrading the download risk per safety protocols.
[COMMAND_EXECUTION] (LOW): The skill uses shell commands to interact with the gemini CLI. This is the intended operational mode for this skill.
[PROMPT_INJECTION] (MEDIUM): The skill exhibits a significant Indirect Prompt Injection surface (Category 8).
Ingestion points: Processes local files (via cat) and web search results.
Boundary markers: Absent. Content is interpolated directly into prompts (e.g., cat document.txt | gemini ... "Summarize this").
Capability inventory: Sends data to an external API and outputs results to the agent.
Sanitization: None detected. An attacker could embed malicious instructions in a file being analyzed to hijack the agent's behavior or exfiltrate data from the conversation context to the LLM output.
[DATA_EXFILTRATION] (INFO): By design, this skill transmits local file content and user prompts to Google's cloud infrastructure. Users should ensure they do not pipe sensitive or PII-containing files into the tool.

gemini