cali-product-ux-critique
Warn
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for installing multiple third-party tools and extensions from GitHub and NPM repositories that are not on the provided trusted vendor list, including packages from Prinova, nicobailon, juicesharp, and tintinweb.
- [COMMAND_EXECUTION]: To enable its audit workflow, the skill contains shell commands for installing these external plugins via 'pi install' or 'npx skills add' and utilizes tools like 'find' and 'git diff' to scan local directories for UI-related files.
- [PROMPT_INJECTION]: The skill's primary function is to audit untrusted external data from live URLs and source code, creating an indirect prompt injection vulnerability. 1. Ingestion points: Live site URLs and local codebase directories processed by the agent. 2. Boundary markers: Absent; tasks are passed to subagents without specific delimiters or instructions to ignore embedded malicious content. 3. Capability inventory: Usage of agent_browser for web interaction, subagents for task delegation, and standard file operations. 4. Sanitization: No explicit content validation, escaping, or filtering of the audited material is described.
Audit Metadata