captions

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions depend on an external toolkit hosted at https://github.com/agamm/video-agent. It guides the user to clone this repository to access the necessary scripts and environment for processing video files.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of untrusted external data, including video files and subtitle files (SRT/VTT), which presents a potential surface for indirect prompt injection.
  • Ingestion points: The agent reads video audio data for transcription and parses external subtitle files provided via the --srt or --vtt arguments in files like SKILL.md.
  • Boundary markers: No explicit boundary markers or 'ignore embedded instructions' warnings are present in the provided command templates.
  • Capability inventory: The skill uses uv run to execute subprocesses for transcription (transcribe) and video editing/rendering (captions). These operations involve file system access for reading inputs and writing output video/text files.
  • Sanitization: The skill does not describe any specific sanitization or validation of the input files, relying on the underlying video-agent and mlx-whisper tools to handle the content safely.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 12:32 PM