The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from external sources that could contain malicious instructions. \n- Ingestion points: The mcp__asc-metadata__list_reviews tool is used to fetch user-generated review text and ratings (referenced as EV.reviews in item P4.2) into the agent's context. \n- Boundary markers: The instructions do not define explicit delimiters or boundary markers to isolate review content, nor do they include warnings for the agent to ignore embedded instructions within that content. \n- Capability inventory: The skill possesses the Write capability for local file generation (GROWTH.md), filesystem access (Read, Glob, Grep), and the ability to interact with users via AskUserQuestion. \n- Sanitization: There are no mentioned mechanisms for filtering, sanitizing, or validating the text content of reviews before it is processed by the agent.

store-growth-audit