render-stopmotion-hand-swatch-cycle
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONNO_CODE
Full Analysis
- [NO_CODE]: The skill package consists of documentation, configuration examples, and metadata but does not include the executable Python scripts (e.g., build_video.py, render_endcard.py) referenced in the pipeline files.
- [COMMAND_EXECUTION]: The instructions describe the assembly of media using shell commands for FFmpeg and browser automation with Playwright, which involves the agent constructing commands based on parameters derived from configuration files.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes external configuration data used in automated media generation and secondary AI prompts.
- Ingestion points: Data fields such as
tagline,subtitle, andplates[].promptwithin the JSON configuration. - Boundary markers: No delimiters or instructions are provided to the agent to isolate untrusted configuration data from system instructions.
- Capability inventory: The workflow utilizes local tools (FFmpeg) and browser automation (Playwright) to process the ingested data.
- Sanitization: The documentation does not specify any validation or sanitization procedures for inputs provided in the configuration files.
Audit Metadata