design-workflow-design
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill (Step 4.5) directs the agent to initiate a local HTTP service for 'interactive visual comparison.' This functionality involves the execution of shell commands to spawn a web server on the local development environment.
- [INDIRECT_PROMPT_INJECTION]: The workflow processes potentially untrusted external data, creating an injection surface.
- Ingestion points: The agent reads the user-controlled
01-spec.mdand incorporates results from 'Web search' queries into its decision-making process. - Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat external specification content as untrusted data.
- Capability inventory: The skill possesses the ability to execute local shell commands (HTTP server), perform file system write operations (updating
DESIGN.mdand feature-specific design files), and access the network via web searches. - Sanitization: The skill lacks instructions for sanitizing or escaping data from external specs or search results before interpolating them into prompts for design generation or other agent personas.
- [EXTERNAL_DOWNLOADS]: The skill utilizes 'Web search' as a fallback mechanism in Step 3 to retrieve design inspiration and patterns from external sources when local references are unavailable.
Audit Metadata