prd-write
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources to generate documentation, creating a surface for potential instruction injection.
- Ingestion points: Processes user-provided requirements, research, data, screenshots, and engineering constraints, along with existing project files, PRDs, and source code (defined in SKILL.md and README.md).
- Boundary markers: The skill enforces a structured output using a Markdown template (
template.md) and requires specific labeling for inferred content ([INFERRED]) and unknown items ([UNKNOWN]), which provides some structural separation but does not prevent malicious instruction obedience. - Capability inventory: The skill is designed to read project files and write new documentation to the
docs/prd/<feature-name>/directory. - Sanitization: There are no instructions for sanitizing or escaping the content derived from external/untrusted inputs before it is interpolated into the final document.
Audit Metadata