The Agent Skills Directory

[DATA_EXFILTRATION]: Instructions facilitate the transmission of local project data to external third-party services.\n
Specifically, it suggests piping sensitive files like source code (src/auth.ts), version control history (git diff), and error logs to external LLM providers via the llm command-line utility.\n- [COMMAND_EXECUTION]: The skill relies on shell command execution for installation and core functionality.\n
It uses system commands to install the necessary tools (pip install, brew install) and to execute the llm binary for querying models.\n- [EXTERNAL_DOWNLOADS]: Downloads software and plugins from public package repositories.\n
Fetches the llm package and its associated plugins (e.g., llm-gemini) from standard registries like PyPI and Homebrew.\n- [PROMPT_INJECTION]: Potential surface for indirect prompt injection via the processing of untrusted local data.\n
Ingestion points: Reads data directly from source files, logs, and git diffs (e.g., src/auth.ts, error.log).\n
Boundary markers: Prompts lack explicit delimiters to separate the data being analyzed from the instructions provided to the external model.\n
Capability inventory: The agent has the ability to execute shell commands and interact with external network services via the llm tool.\n
Sanitization: Content from the ingested files is passed directly to the external model without validation or escaping of potential embedded instructions.

llm-advisor