stitch-design-taste
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill generates a DESIGN.md file that serves as instructions for the Google Stitch AI agent. It uses user-provided "vibe descriptions" to dynamically adjust design parameters (density, variance, motion). This creates a surface where a malicious user could provide input designed to influence or override the instructions intended for the downstream agent.\n
- Ingestion points: User-provided atmosphere/vibe descriptions ingested at runtime in SKILL.md.\n
- Boundary markers: None present in the generated DESIGN.md template to delimit user-provided content from system instructions.\n
- Capability inventory: The skill specifies file-write operations to the llmwiki/ directory via its Output Report logic.\n
- Sanitization: No explicit sanitization or validation of user-provided descriptions is performed before they are synthesized into the output instructions.\n- [COMMAND_EXECUTION]: The skill contains a structured Output Report section that instructs the agent to perform file system operations, including creating new markdown files in the llmwiki/wiki/draft/uiux/ directory and updating index and log files. These operations are localized to the project's internal wiki structure.
Audit Metadata