craft-critique
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The instruction file
references/adapters/html.mddirects the agent to execute a shell command involving a headless browser (e.g., Chrome) to generate screenshots for visual critique. While it uses a placeholder"<chrome>", it encourages direct shell execution for rendering tasks. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design data, including HTML/CSS source code and design briefs, creating an attack surface for embedded malicious instructions.
- Ingestion points: UI source code (HTML/CSS), project design briefs (
DESIGN.md), and canvas nodes from external platforms like Figma or Paper. - Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded text within the designs being reviewed.
- Capability inventory: The agent has capabilities to read/write local files (creating and deleting exploration folders) and execute shell commands for rendering, which could be abused if an injection is successful.
- Sanitization: No sanitization or validation of the design content's text or properties is performed before the agent analyzes and potentially modifies files during "refine" mode.
Audit Metadata