hedra-character-video
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it processes external, untrusted assets (images, audio, and text scripts) which are then interpolated into prompts or API payloads for video generation.
- Ingestion points: The skill ingests user-provided character images, audio files for lip-syncing, and text prompts for performance direction (SKILL.md).
- Boundary markers: The skill references safety policies and includes a dedicated section on 'Safety, rights, and custody' to mitigate misuse.
- Capability inventory: The skill possesses network capabilities (calls to Hedra API), local file system access (storing generated assets), and potential execution of LiveKit agent scripts for real-time interaction.
- Sanitization: The instructions explicitly advise verifying likeness rights, obtaining consent for voice cloning, and excluding sensitive personal or children's data from inputs.
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of specific, vendor-related software packages to enable real-time and programmatic features.
- Python package: 'livekit-agents[hedra]~=1.0' for Live Avatar integration.
- Node.js package: '@hedra/create-hedra-avatar' for scaffolding real-time avatar applications.
- [COMMAND_EXECUTION]: The skill instructs the agent to automate video production through direct API interactions and local project management.
- API workflow: Directs the use of 'POST /assets', 'POST /generations', and status polling at '/generations/{id}/status' to coordinate asset lifecycle.
- Local custody: Recommends maintaining a local asset ledger for project inputs and outputs rather than relying exclusively on cloud state.
Audit Metadata