The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted data from intent files and user stories. Ingestion points: Reads intent files, functional/non-functional requirements, and user stories supplied in conversation or files (SKILL.md). Boundary markers: The skill emphasizes restating intent and requirements without changing meaning, and preserving identifiers, but does not use explicit delimiters like XML tags for untrusted input. Capability inventory: Uses file reading to inspect the repository and file writing to save specifications; utilizes network access for research purposes. Sanitization: No specific sanitization or escaping mechanisms for external inputs are defined before their inclusion in the design specifications.

design-specs