motion-video-production
Warn
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill uses dynamic imports in its internal pipeline (e.g.,
probe.mjs,score.mjs, andvo.mjs) to load project-specific configuration and track files (track.ts,timing.ts) from computed paths based on command-line arguments. - [COMMAND_EXECUTION]: The skill makes extensive use of
subprocess.runin Python andexecFileSyncin Node.js to invoke system utilities such asffmpeg,ffprobe,curl, and thecomposioCLI. While command arguments are generally passed as lists (reducing standard shell injection risks), the large surface area for interacting with system binaries via user-provided or project-generated filenames warrants caution. - [EXTERNAL_DOWNLOADS]: In the voice-over assembly stage (
vo.mjs), the skill usescurlto download audio assets from remote URLs returned by Text-to-Speech service APIs (ElevenLabs and Fish Audio) via the Composio toolset. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources like
STORY.md,character.txt, and reference videos. This data is interpolated into generation prompts and used in processing gates without explicit sanitization or strict boundary markers, which could lead to the agent following malicious instructions embedded within the processed video project content.
Audit Metadata