The Agent Skills Directory

[SAFE]: The skill is designed for information retrieval from a local directory structure. It explicitly states it 'writes nothing' and contains no instructions for network operations, command execution, or sensitive data access. The logic is constrained to reading markdown files and formatting answers based on specific citation rules.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the knowledge/ and sources/ directories. Ingestion points: Loads text and metadata from files like knowledge/index.md and sources//.md. Boundary markers: The skill does not define specific delimiters for ingested text, though it mandates strict citation formats for output. Capability inventory: No high-risk capabilities such as file-writing, network communication, or shell command execution were found. Sanitization: The skill includes instructions to check for mangled content via a 'citing' field, though it does not implement specific sanitization for malicious prompt instructions within the ingested data. Given the lack of dangerous capabilities, the risk associated with this surface is negligible.

kb-expert