recordly
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.run()inscripts/check_x_export.pyto invokeffprobeandffmpeg. These calls are used for legitimate media verification and re-encoding tasks. The inputs are passed via standard command-line arguments. - [EXTERNAL_DOWNLOADS]: The skill includes several network operations targeting well-known and trusted services:
- Fetches
three.jsand related 3D libraries from the JSDelivr CDN. - Downloads environment maps (HDRI) from Poly Haven's public infrastructure.
- Installs the Recordly application via the official Homebrew repository or clones the source code from GitHub.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external media files (video/audio) and metadata (UI control names) which represent a potential ingestion surface for untrusted data.
- Ingestion points: Ingests file paths, video metadata (via
ffprobe), and Electron DOM content (via CDP). - Boundary markers: None explicitly defined in the scripts, though agent-browser usage follows standard platform patterns.
- Capability inventory: Includes file system writes (for video fixes), subprocess execution (
ffmpeg), and browser automation (agent-browser). - Sanitization: Uses
json.loadsforffprobeoutput and standard argument parsing; however, the agent interprets raw DOM strings from the Recordly HUD.
Audit Metadata