dreamina-video-production
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes local reference videos, creating a potential vector for instructions embedded within the media content to influence agent behavior.\n
- Ingestion points: Local reference videos are processed by dreamina_analyze_reference_video and annotated by dreamina-shot-annotator (SKILL.md).\n
- Boundary markers: The skill uses PreviewValidated and deterministic evidence gates, but does not explicitly instruct the agent to ignore prompt-like content within the video frames or audio.\n
- Capability inventory: The workflow includes state-changing operations such as dreamina_execute_video_batch and dreamina_compose_video (SKILL.md).\n
- Sanitization: The system validates resource integrity via machine fingerprints and digests but does not mention semantic filtering of media content.\n- [COMMAND_EXECUTION]: The skill facilitates the use of local command-line tools for media processing through a managed enrollment process.\n
- Evidence: SKILL.md describes the enroll_media_tools action which binds ffmpeg, ffprobe, and whisper by absolute path and cryptographic digest to ensure only approved binaries are used.
Audit Metadata