video-toolkit
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes numerous local commands including python3, npm, and modal to manage video assets and rendering pipelines. These commands are executed from a specific toolkit path (~/.openclaw/workspace/claude-code-video-toolkit) and include verifying setups, generating assets, and rendering video files.- [EXTERNAL_DOWNLOADS]: Installs Python dependencies using pip3 from a local requirements.txt file and Node.js dependencies via npm install. It also facilitates the deployment of AI models to cloud infrastructure using the modal CLI, which involves downloading and uploading containerized applications.- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting untrusted data from user-provided text briefs, voiceover scripts, and configuration files (demo-config.ts) which are then processed by the LLM and passed to shell commands.
- Ingestion points: User-provided text strings passed to --text and --prompt arguments in python3 tool calls, and content defined in VOICEOVER-SCRIPT.md.
- Boundary markers: Absent; user input is interpolated into command-line arguments without explicit delimiters or ignore-instructions warnings.
- Capability inventory: Subprocess execution via python3, package management via npm, and cloud resource management via modal.
- Sanitization: Absent; the instructions do not specify any validation or escaping of user-provided strings before they are passed to the generation tools.
Audit Metadata