generating-composite-and-video

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run() in scripts/merge_clips.py to execute FFmpeg commands. This is used for its primary function: concatenating generated video clips into a single final animation. The command uses static arguments and a generated concat.txt file pointing to local file paths, which is standard for video processing tasks.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from shots.json, specifically the action and veo_prompt fields, which are sent as prompts to external AI APIs (flux-2/pro-image-to-image and sora-2-image-to-video). While these are points where untrusted data could enter the context, the skill includes explicit user approval gates after every generation step, minimizing the risk of automated instruction following.
  • Ingestion points: shots.json, characters.json, backgrounds.json (read in scripts/generate_videos.py).
  • Boundary markers: None explicitly in the code, but the SKILL.md defines a strict approval workflow.
  • Capability inventory: File system writes (composites/, clips/), network POST/GET (kie.ai API), shell execution (FFmpeg via subprocess.run).
  • Sanitization: Standard JSON parsing is used; inputs are interpolated directly into API payloads.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 02:15 AM
Security Audit — agent-trust-hub — generating-composite-and-video