The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is configured to access and extract content from local directories including 'SOPs/', 'internal/', and 'claudes-docs/'. While these paths are designated for documentation, they represent a broad access scope to internal organizational data and procedures.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites (via WebFetch) and GitHub repositories. It lacks mandatory evidence of boundary markers or instructions to sanitize this content.
Ingestion points: WebFetch tool for arbitrary external URLs and Read tool for local documentation and GitHub files.
Boundary markers: Absent. The instructions do not provide delimiters or warnings to ignore instructions embedded in the fetched content.
Capability inventory: Read, Glob, Grep, WebFetch, and Bash (including cat, grep, find).
Sanitization: Absent. There is no logic for escaping or validating external content before processing.
[COMMAND_EXECUTION]: The skill instructions mandate the use of the Bash tool to perform file system searches and data extraction using commands like cat, grep, and find. The availability of shell execution capabilities combined with the ingestion of untrusted external data increases the attack surface for potential exploitation.

000-jeremy-content-consistency-validator