skill-prd
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses imperative override instructions at the beginning of the
SKILL.mdfile (e.g., "STOP - SKILL ALREADY LOADED", "DO NOT load any more skills. Execute directly."). These instructions are designed to override the agent's standard orchestration flow and skill-loading protocols.
- [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-AI orchestration flow where user-supplied content is passed to external AI providers for review without adequate sanitization.
- Ingestion points: User responses provided during the 'PHASE 0: CLARIFICATION' step in
SKILL.md. - Boundary markers: Absent. The dispatch prompt in Phase 2.5 ("Challenge this PRD. [DRAFT]") does not use delimiters or instructions to ignore potential commands embedded within the draft PRD content.
- Capability inventory: The skill utilizes web searches in Phase 1, dispatches data to secondary AI providers in Phase 2.5, and performs file writing in Phase 4.
- Sanitization: None. User-influenced data is directly incorporated into the PRD draft and subsequently processed as instructions for adversarial review by another model.
Audit Metadata