highlight-cutter
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes external tools (ffmpeg and yt-dlp) via a Python script to perform video editing tasks.
- [EXTERNAL_DOWNLOADS]: The skill uses yt-dlp to fetch media content from URLs provided as arguments at runtime.
- [PROMPT_INJECTION]: The SKILL.md file contains an 'Iron Rule' (Token Budget Policy) that explicitly forbids the AI from reading the contents of the JSON plan files it processes. This instruction suppresses the AI's ability to inspect and validate the data it handles, which is a common technique to bypass the agent's internal oversight.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where data from an external, potentially untrusted JSON file is used to construct command-line arguments for system tools. Ingestion points: scripts/cut_highlight.py reads plan_json. Boundary markers: Absent. Capability inventory: Subprocess execution of ffmpeg and yt-dlp in scripts/cut_highlight.py. Sanitization: The script relies on Python's list-based subprocess.run to prevent shell injection, but does not validate the content of the JSON fields before they are used as parameters.
Audit Metadata