claude-video-caption

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute system utilities including FFmpeg, FFprobe, and the whisper-ctranslate2 transcription engine. It also executes local shell scripts (scripts/preflight.sh and scripts/caption_pipeline.sh) to manage the video processing workflow.
  • [DYNAMIC_EXECUTION]: The skill instructs the agent to generate and execute an inline Python script (words_to_ass) to handle the conversion of transcription data into advanced subtitle formats. This involves writing a temporary file and executing it using the system's Python interpreter.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface related to the processing of untrusted data.
  • Ingestion points: Untrusted media files ($INPUT) are processed through the Whisper transcription engine in SKILL.md.
  • Boundary markers: There are no explicit instructions or delimiters provided to the agent to ignore potentially malicious content found within the generated transcriptions.
  • Capability inventory: The skill utilizes powerful capabilities including full Bash execution, file writing, and Python script execution.
  • Sanitization: No explicit sanitization or filtering of the transcription text is performed before it is handled by the subtitle generation script or burned into the video output.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 06:38 PM
Security Audit — agent-trust-hub — claude-video-caption