video-editing
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on a suite of Python scripts and Bash wrappers to automate media processing tasks using FFmpeg and FFprobe. These operations are essential for functionalities such as audio extraction, video rendering, and crop suggestion. While the cleanup script contains a potential shell injection point if filenames are maliciously crafted by the user, the implementation aligns with its stated purpose of managing project assets.\n- [PROMPT_INJECTION]: The skill involves indirect prompt injection surfaces because it processes untrusted transcript data (SRT/TXT files) derived from user-provided media. The agent is instructed to read these transcripts to validate edits and scan for errors.\n
- Ingestion points: Captions are read and parsed in
scripts/parse_srt.py,scripts/cut_srt.py,scripts/scan_flubs.py, and reviewed inwork/edit/clean-preview.txt.\n - Boundary markers: No specific delimiters or safety instructions are used to isolate transcript text from the agent's instruction processing logic.\n
- Capability inventory: The agent can execute shell commands via the provided scripts and perform filesystem operations.\n
- Sanitization: Content is parsed for structure but the text within the captions is not sanitized against embedded instructions.
Audit Metadata