The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads Node.js binaries and the @iqinghu/qhkit NPM package to establish its operating environment. These resources are fetched from well-known and trusted domains (nodejs.org and the NPM registry).
Integrity Verification: The installation instructions include a mandatory step to verify the SHA256 checksum of the downloaded Node.js archive before extraction, which prevents the execution of corrupted or tampered binaries.
Trusted Mirrors: It suggests the use of a reputable mirror (npmmirror.com) for users in restricted network environments, which is a standard and safe development practice.
[COMMAND_EXECUTION]: The skill utilizes the qhkit command-line interface. It includes specific guidance to ensure commands are executed safely across different operating systems.
Injection Prevention: The instructions explicitly direct the agent to write complex parameters into a JSON file (@params.json) rather than passing them as raw shell arguments. This practice effectively mitigates command injection risks related to shell expansion and character escaping.
Automated Scan Resolution: The automated security alert regarding 'Remote Code Execution' is a false positive. The identified command sequence (curl | grep | sha256sum) is an integrity check for the Node.js installer, not a 'pipe-to-shell' execution of remote scripts.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (video titles and metadata) from social media platforms, which constitutes an attack surface for indirect prompt injection.
Ingestion points: External data enters the context via the fetch_user_videos tool output.
Boundary markers: The tool uses structured JSON for data exchange and directs the agent to store large results in local files, helping to maintain clear boundaries between instructions and data.
Capability inventory: The skill can execute the qhkit CLI and perform local file read/write operations for configuration and data management.
Sanitization: The use of structured JSON files for parameters and the requirement for explicit user authorization for API costs provide significant layers of protection against the accidental execution of instructions embedded in retrieved data.

qinghu-shipinhao-videos