browser-preview

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill defines a workflow where the agent scans user-controlled workspace files (such as package.json, index.html, app.py) to discover projects. These files serve as ingestion points for untrusted data that could contain malicious instructions designed to influence the parameters of the preview_serve tool or the agent's subsequent responses. No explicit sanitization or boundary markers for this untrusted content are defined in the instructions.
  • [COMMAND_EXECUTION]: The agent is instructed to execute system commands including curl for server-side port diagnostics on localhost and find for workspace discovery. These commands are scoped to the internal environment and used for legitimate diagnostic purposes.
  • [DATA_EXFILTRATION]: The skill accesses internal state files located at /data/previews.json and /data/preview_history.json. These files contain metadata about running services and historical preview activity within the platform's infrastructure, rather than sensitive user credentials or external data.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 08:03 AM