The Agent Skills Directory

[SAFE]: The skill facilitates interaction with LLMs using an internal library (sdk.llm). It uses standard execution methods like run_python_snippet and shell_exec to perform its primary function.
[COMMAND_EXECUTION]: The skill documentation describes using shell_exec to run Python scripts. This is presented as a standard method for executing the programmatic LLM calls defined by the skill's SDK.
[SAFE]: The skill provides utility functions (file_to_url, image_to_base64) to process workspace files for use with vision models. This creates a data ingestion surface for indirect prompt injection (Evidence: Ingestion point in SKILL.md, Capability inventory includes shell_exec, Boundary markers are absent in examples, Sanitization is absent), which is considered a standard functional surface for LLM-related tools and generally low-risk.

using-llm