suede-component-lab
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to create fixtures and workbenches which involves executing local project commands. Safety is addressed by limiting commands to the existing project runtime and explicitly warning against running code from untrusted sources like source comments.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests repository content such as component source and PR comments, creating a surface for indirect prompt injection. The skill mitigates this risk by providing strict boundaries that define such content as task evidence rather than instructions for the agent.
- Ingestion points: local component source code, PR bodies, and repository comments.
- Boundary markers: explicit instructions in the 'Boundaries' section stating that external documents are evidence and not authority.
- Capability inventory: local file system access and shell command execution via the project's runtime.
- Sanitization: requirement for synthetic local fixtures and non-mutating test adapters.
Audit Metadata