video-compositing

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user-provided content including video footage, images, and text copy for rendering. While it provides instructions to filter transcription tokens, it lacks explicit boundary markers or sanitization requirements when interpolating this user data into renderer project files or command-line parameters.\n
  • Ingestion points: User assets, logo files, and exact text copy defined in references/production-guide.md.\n
  • Boundary markers: The skill does not define specific delimiters to isolate user-supplied text from rendering instructions.\n
  • Capability inventory: The skill creates project directories, executes local renderers (ffmpeg, Hyperframes), and makes network calls to media models (genmedia, fal MCP tools).\n
  • Sanitization: Instructions mention filtering non-word tokens for captions, but do not address escaping for shell commands or renderer project formats.\n- [EXTERNAL_DOWNLOADS]: The skill references the Hyperframes renderer from the heygen-com GitHub organization, which is a well-known entity in the AI video space. It also instructs the agent to install publicly documented dependencies, though it mandates that this only occurs within the user's authorized setup.\n- [COMMAND_EXECUTION]: The skill workflow involves executing local rendering tools like ffmpeg and third-party renderers. It includes security-positive instructions to check command-specific help documentation and perform draft renders for inspection before the final export.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 07:17 AM