The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches and installs software and dependencies from external repositories during its environment setup phase.
It installs the @iqinghu/qhkit package from the NPM registry as a primary dependency.
It contains instructions to download Node.js binaries and verification checksums from the official nodejs.org website.
It suggests the use of the well-known Alibaba mirror (registry.npmmirror.com) as a fallback for connectivity issues.
The Node.js installation includes a security best practice: verifying the integrity of the downloaded archive using sha256sum before extraction.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data (text prompts and images) which are subsequently used as arguments for the qhkit CLI tool, creating a surface for indirect prompt injection.
Ingestion points: User-provided prompt strings and uploadedImages paths are ingested into the qhkit video generate and qhkit video-quick generate command templates.
Boundary markers: The instructions do not define boundary markers (like delimiters) or guidance to the agent to treat these inputs purely as data, which might allow maliciously crafted prompts to influence the agent's behavior or CLI execution.
Capability inventory: The CLI tool has the capability to read local files and perform network operations (uploading assets and submitting tasks to the LinkPix service).
Sanitization: There is no evidence of sanitization or escaping of the user input before it is interpolated into shell commands.
[DYNAMIC_EXECUTION]: The skill generates and executes small scripts at runtime to handle image pre-processing requirements.
It instructs the agent to use a Python one-liner (python -c) with the Pillow library to resize images if they exceed size limits.
Alternatively, it utilizes npx sharp-cli to perform image compression tasks dynamically.

linkpix-seedance-25-sales