video-shotcraft
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Several automation scripts included with the skill execute shell commands to perform system tasks.
workbench/scripts/open.mjsuseschild_process.spawnto executenpm installandnpm run dev(Vite server) during the project setup phase.assets/scripts/smoke-render-demos.pyusessubprocess.runto call theremotionCLI for automated smoke testing and rendering.jianying-export/smoke_test.pyexecutespip install pyJianYingDraftto verify the environment.jianying-export/mac_draft.pyusesos.systemfor file operations andsubprocess.runto check the system configuration on macOS.jianying-export/windows_draft.pyusessubprocess.runto callreg.exefor querying the Windows registry to locate installation paths.- [EXTERNAL_DOWNLOADS]: The skill initiates downloads of third-party software and libraries during its normal operation.
- The JianYing export module automatically installs
pyJianYingDraftfrom PyPI if it is not present. - The Motion Workbench automatically installs its Node.js dependencies via
npm installwhen first opened. - The capture script
assets/scripts/capture-template.mjsutilizes Puppeteer, which may download a specific version of Chromium for headless browser operations. - [DYNAMIC_EXECUTION]: The skill contains logic that generates and then executes script files as part of its testing and indexing pipeline.
assets/scripts/smoke-render-demos.pydynamically generatestemplate/src/smoke-root.tsxby scanning thedemos/directory and then executes the resulting code via Remotion.workbench/scripts/gen-index.mjsgenerates metadata and index files (demo-index.ts,demoMeta.ts) at runtime to populate the Motion Workbench library.- [INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface due to its complex data processing pipeline.
- Ingestion points: The agent reads local shot recipe cards (
references/shots/), UI layout JSON files, and user-provided copy/branding information. - Boundary markers: Missing explicit delimiters to separate user-provided branding text from the agent's internal video generation instructions.
- Capability inventory: The skill allows the agent to write files, execute shell commands, and run local web servers.
- Sanitization: There is no visible sanitization of the text content before it is interpolated into the Remotion TSX templates, which could potentially lead to unexpected behavior if processed by downstream automated agents.
Audit Metadata