linkpix-seedance-25-clone
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download Node.js binaries directly from
nodejs.organd install the@iqinghu/qhkitpackage from the official NPM registry or a well-known Alibaba mirror (npmmirror.com). These are established and trusted services. - [COMMAND_EXECUTION]: The instructions involve executing shell commands to configure the environment, manage packages, and interact with the video processing API via the
qhkitCLI tool. - [PERSISTENCE]: The skill suggests appending the Node.js binary path to the user's
~/.bashrcfile. While modifying shell profiles is a persistence technique, in this context it is a standard and legitimate method for configuring the execution environment for developer tools. - [DYNAMIC_EXECUTION]: The skill uses
python -candnpxto execute inline scripts for image processing (compression and resizing) as a fallback mechanism. These operations are performed on local files using established libraries like Pillow and Sharp. - [INDIRECT_PROMPT_INJECTION]: The skill processes external video links and scripts which represent a potential injection surface. However, the skill includes a mandatory evidence chain for safety:
- Ingestion points: External short-video URLs (TikTok, Douyin, etc.) processed by
qhkit video-inspire. - Boundary markers: Present; instructions require the agent to present the "Original Script vs. Rewritten Script" for user verification before proceeding.
- Capability inventory: File system writes (temporary files), network access (API calls via qhkit), and subprocess execution (Python/Node for image processing).
- Sanitization: Relies on LLM rewriting and explicit human-in-the-loop confirmation before task submission.
Audit Metadata