design-color
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill audits and processes a wide variety of project files to extract color information, creating a significant ingestion surface for untrusted data.
- Ingestion points: Reads manifest files (
package.json,pubspec.yaml,build.gradle), configuration files (tailwind.config.js), and source code files across multiple frameworks. - Boundary markers: The instructions lack boundary markers or warnings to ignore potential instructions embedded in the comments or metadata of analyzed files.
- Capability inventory: The skill has the capability to write to the local file system (modifying CSS, Dart, and configuration files) and execute project build commands.
- Sanitization: No explicit sanitization or validation of the content ingested from the codebase is performed before it is used to influence the agent's actions.
- [COMMAND_EXECUTION]: The skill includes a 'Self-Healing Validation' phase that instructs the agent to 'Run the build command to verify no compile errors from color changes', which involves executing project-defined scripts.
Audit Metadata