diverse-plan
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided task descriptions and project files to generate implementation plans, which creates a potential surface for indirect prompt injection.
- Ingestion points: External task goals, scope, and constraints enter the agent context in the 'Workflow' section of SKILL.md (Steps 1 and 2) to form model briefs.
- Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing these task descriptions.
- Capability inventory: The skill does not perform direct shell commands or network operations, but it delegates its output to the 'to-prd' skill, creating a chain where injected instructions could potentially influence downstream tools.
- Sanitization: No mechanisms for escaping, filtering, or validating the external task content are defined before the data is used to prompt the model branches.
Audit Metadata