The Agent Skills Directory

[PERSISTENCE]: The instructions include steps to modify the shell profile (~/.bashrc) to update the system PATH permanently. This ensures the manually installed Node.js binary remains available in future shell instances.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (e.g., Douyin video URLs) through the qhkit CLI tool, which may lead to processing untrusted instructions embedded in external content.
Ingestion points: Data is ingested via the resourceUrl and playVideo parameters in the qhkit command calls.
Boundary markers: The skill uses JSON string formatting for command parameters, providing some structural separation.
Capability inventory: The qhkit tool performs network operations and likely file system writes for generating and retrieving marketing assets.
Sanitization: There are no instructions to sanitize the output from external URLs before it is processed or presented to the user.
[EXTERNAL_DOWNLOADS]: The skill downloads software components from external repositories.
It fetches Node.js binaries from nodejs.org (a well-known official service) or Aliyun's npmmirror.com (a known mirror service).
It installs the @iqinghu/qhkit NPM package, which is a resource owned by the vendor (iqinghu.com).
Integrity verification is performed using sha256sum against official checksum files, which is a security best practice.
[COMMAND_EXECUTION]: The skill executes several shell commands (curl, tar, npm, qhkit) to manage its execution environment and perform image generation tasks.
[REMOTE_CODE_EXECUTION]: The skill installs and executes the @iqinghu/qhkit package and the Node.js runtime. While these originate from trusted vendor and official sources, they involve downloading and running code from remote servers.

qinghu-image-ad-assets