audience-site-brief
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill captures raw user input in 'Phase 1: Broad Intake' (e.g., descriptions of brand goals and product offerings) which is then used as context for research, persona generation, and architecture planning.\n
- Ingestion points: User input captured in the 'Intake' block in
SKILL.md.\n - Boundary markers: The input is stored in a markdown list but lacks explicit 'ignore embedded instructions' guards for the agent.\n
- Capability inventory: Orchestrates child skills such as
audience-researchandvisual-identity, and ultimately generates a 'build prompt' for site generation. The 'audience-research' phase additionally involves searching external user sources, which introduces a surface for tool output poisoning if malicious content is encountered.\n - Sanitization: No sanitization or validation of the user's text description is performed before it is processed by downstream skills.\n- [DYNAMIC_EXECUTION]: The skill generates a complex 'build prompt' deliverable by assembling user-specific content with pre-defined module blocks and presets. This generated prompt acts as a functional specification or instruction set for the
motion-web-designskill to execute a website build.
Audit Metadata