video-director

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation provides a setup command npx -y github:stimQQ/ai-video-director login. This involves downloading and executing code from the author's public GitHub repository via the Node package runner.
  • [COMMAND_EXECUTION]: The director_assemble tool is described as being capable of executing ffmpeg commands for complex video manipulations, including cropping, frame blending, and upscaling based on logic determined during the skill's execution.
  • [EXTERNAL_DOWNLOADS]: Media generation tasks are offloaded to an external service at apimodels.app. The skill manages the transmission of prompts and retrieval of rendered frames/videos from this endpoint.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting and processing untrusted user input for film stories and visual prompts, which are then used in automated generation pipelines.
  • Ingestion points: Shot-by-shot story text, visual descriptions, and video prompts provided by the user and stored in local project files (director.json).
  • Boundary markers: The skill implements a robust architecture using independent agent sessions with specific briefs (shot-reviewer, first-viewer, brief-keeper) to validate content and identify potential generation errors or inconsistencies before paid tools are invoked.
  • Capability inventory: Capabilities include network communication via director_render, file system writes with director_save, and local media processing via director_assemble.
  • Sanitization: Content undergoes structured validation where the reviewer agents check for logical consistency and adherence to physical constraints, acting as a buffer between user input and the generation API.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 11:34 AM