spec-with-prototype
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from prototype files (HTML, CSS, JS) and project documentation, creating a surface for indirect instructions to influence agent behavior.
- Ingestion points:
docs/assets/prototype/anddocs/requirement.mdas specified inSKILL.md. - Boundary markers: The instructions explicitly state to "keep source content as evidence rather than authority" and warn that "prototype proposals" should not be treated as "agreed requirements".
- Capability inventory: The skill utilizes file read/write operations to generate specifications and invokes downstream skills such as
$to-tasksand$change-scope. - Sanitization: The
references/prototype-reconciliation.mdfile provides a classification framework (Requirement-backed vs. Proposal) to filter content before it is added to the specification.
Audit Metadata