The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional text for the LLM. It does not invoke any external tools, execute scripts, or access the underlying operating system. No malicious patterns were detected across any of the analyzed threat categories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data through the {{input}} variable. While it lacks explicit boundary delimiters to isolate this input from the system instructions, the skill possesses no capabilities (such as network access, file writing, or command execution) that could be abused via prompt injection. The only possible outcome of a successful injection would be a manipulation of the generated text response.

mental-models