The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it incorporates the contents of existing documentation files into its workflow without sanitization.
Ingestion points: The skill reads docs/product/learnings.md and docs/product/vision.md as part of the context assessment and vision definition steps.
Boundary markers: No delimiters or safety instructions are used to distinguish documentation content from the skill's primary instructions.
Capability inventory: The skill possesses file-writing capabilities specifically for the docs/product/ directory.
Sanitization: There is no evidence of validation or filtering for data retrieved from the documentation files.
[SAFE]: The skill implements a mandatory human-in-the-loop checkpoint ([STOP — BLOCKING]) that prevents the agent from writing any files until the user has explicitly reviewed and confirmed the proposed changes.

recipe-vision