walkthrough-record

Pass

Audited by Gen Agent Trust Hub on Jul 21, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The recorder script (record.mjs) executes ffmpeg and ffprobe to handle video conversion and metadata extraction. These commands are invoked via execFileSync, and parameters such as output filenames are sanitized using basename to prevent path traversal attacks.
  • [EXTERNAL_DOWNLOADS]: The setup.sh script automates the installation of the Playwright Chromium browser and the skill's Node.js dependencies (playwright, yaml) from the official npm registry. This is standard and expected behavior for a browser automation tool.
  • [DATA_EXFILTRATION]: While the skill can automate a browser to navigate to any URL (including local files or internal endpoints), it implements a strict schema validator that limits URL schemes to http, https, or file. The instructions explicitly warn users to verify targets before using authenticated sessions (storageState).
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted scenario YAML files. It mitigates risks by implementing a comprehensive schema validator in record.mjs that checks all keys, step kinds, and numeric ranges before launching the browser. Output file names are strictly sanitized to prevent writing to unexpected locations.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 21, 2026, 09:35 PM
Security Audit — agent-trust-hub — walkthrough-record