desktop-preview-loop
Pass
Audited by Gen Agent Trust Hub on Apr 1, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell commands to start development servers and verify their status. Examples include running 'npm run dev', 'vite', and 'curl' to poll localhost. These commands are restricted to the local environment and align with the skill's primary purpose.
- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by processing external application data.
- Ingestion points: The skill captures browser console logs and network request data during the 'Capture Results' phase (SKILL.md).
- Boundary markers: There are no explicit delimiters or instructions defined to isolate captured application data from the agent's operational instructions.
- Capability inventory: The agent is empowered to execute local shell commands and modify project source code (SKILL.md, Step 5).
- Sanitization: No mechanisms for sanitizing or validating the captured console or network data are mentioned before the agent processes it.
Audit Metadata