refocus
Warn
Audited by Gen Agent Trust Hub on Jun 28, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to run 'npx clean-room-skill@latest' if the local binary is unavailable, which fetches a package from the public NPM registry at runtime.
- [REMOTE_CODE_EXECUTION]: The use of 'npx' with an external package enables the execution of remote code downloaded from a third-party registry.
- [COMMAND_EXECUTION]: The skill relies on executing shell commands (specifically the 'clean-room-skill' tool) to perform its core audit and verification functions.
- [PROMPT_INJECTION]: The skill acts as an audit layer over potentially untrusted external data, which creates a surface for indirect prompt injection.
- Ingestion points: Reads project files including 'task-manifest.json', 'preflight-goal.json', 'clean-run-context.json', 'implementation-plan.json', 'implementation-report.json', 'qc-report.json', and 'polish-report.json'.
- Boundary markers: The skill does not use delimiters or explicit boundary markers to separate the content of these files from the agent's instructions.
- Capability inventory: Authorized to execute the 'clean-room-skill' command and provide process corrections for the workflow.
- Sanitization: No input validation or content sanitization is described before the files are analyzed by the agent.
Audit Metadata