The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: A shell command in SKILL.md downloads a checksum file and pipes it to sha256sum. While automated scanners flagged this as potential remote code execution, manual review confirms this is a critical security best practice for verifying the integrity of the Node.js binary before installation, not an execution of untrusted code.
[EXTERNAL_DOWNLOADS]: The skill downloads Node.js and the @iqinghu/qhkit package from official, well-known sources including nodejs.org, Alibaba's NPM mirror, and the public NPM registry. These sources are considered reputable and safe for dependency retrieval.
[PERSISTENCE]: The installation instructions describe adding the path of the locally installed Node.js binary to ~/.bashrc. This is a standard and benign persistence mechanism used to ensure that the required runtime environment is preserved across future shell sessions for the agent.
[INDIRECT_PROMPT_INJECTION]: The skill provides an interface for processing user-supplied URLs, creating a surface for indirect prompt injection.
Ingestion points: The urls parameter in the qhkit video-edit generate command accepts external data (SKILL.md).
Boundary markers: None explicitly defined for the input strings.
Capability inventory: The skill executes shell commands via the qhkit CLI (SKILL.md).
Sanitization: The skill mitigates risks by recommending that parameters be passed via a JSON file (@params.json) rather than direct string interpolation, which effectively prevents shell command injection.
[CREDENTIALS_UNSAFE]: The skill manages authentication securely. It instructs the user to retrieve an API token from the official vendor dashboard and provides methods to set it via environment variables or CLI configuration, avoiding any hardcoded secrets within the skill instructions.

linkpix-video-watermark-remove