The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides explicit instructions and workflows for piping local project data to external AI service providers (OpenAI, Google, and Anthropic) via the llm CLI tool. It specifically suggests analyzing "Security-sensitive code (auth, crypto, input validation)" and piping system logs (cat large-app.log). This poses a data exposure risk as secrets, PII, or proprietary logic contained within those files will be transmitted to the third-party LLM APIs.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by design.
Ingestion points: Text generated by external models via the llm command is ingested directly into the agent's context (SKILL.md).
Boundary markers: None. The instructions do not provide delimiters or warnings to the agent to treat external LLM output as untrusted data.
Capability inventory: The agent has shell access and file read capabilities to execute the advisor tasks, which could be exploited if an external model provides malicious instructions.
Sanitization: There is no automated sanitization. While the documentation advises users to "Verify suggestions," the agent itself has no built-in mechanism to filter malicious commands or advice returned by the external LLM.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the llm package and several plugins (llm-gemini, llm-anthropic) via pip or brew. While these are established tools, they represent external dependencies managed outside of the skill's primary environment.

llm-advisor