The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions to modify the user's shell environment to ensure the Node.js binary is available in future sessions.
Evidence: Instructions in SKILL.md to append export PATH="$HOME/.local/lib/node-v22.22.3-linux-x64/bin:$PATH" to ~/.bashrc.
[EXTERNAL_DOWNLOADS]: The skill downloads software and dependencies from external repositories and mirrors to set up the execution environment.
Evidence: Downloads Node.js binaries from the official Node.js distribution site.
Evidence: Installs the @iqinghu/qhkit and sharp-cli packages from npm registries.
Evidence: Directs users to fetch API keys from the iqinghu.com domain.
[COMMAND_EXECUTION]: The skill relies on executing external CLI tools to perform its primary functions, including task management and system setup.
Evidence: Execution of the qhkit binary subcommands (video, video-inspire, config) to interact with remote APIs.
Evidence: Use of system tools like tar and sha256sum for software installation and checksum verification.
[DYNAMIC_EXECUTION]: The skill generates and executes code snippets at runtime for image processing and dependency management.
Evidence: Execution of inline Python scripts using python -c to perform image resizing and compression.
Evidence: Use of npx to dynamically download and execute the sharp-cli tool for media processing.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external video platforms which could be manipulated to influence the agent's behavior.
Ingestion points: Untrusted video sharing URLs from platforms such as Douyin or TikTok are processed by the tool to generate scripts.
Boundary markers: Absent. The skill instructions do not specify structural delimiters or explicit instructions for the AI to ignore embedded commands in the processed data.
Capability inventory: The skill possesses capabilities for network operations (via qhkit), file system modifications (for image processing), and command execution.
Sanitization: The agent is instructed to "rewrite" generated scripts before final submission, providing a manual verification step but lacking programmatic sanitization.

linkpix-wanx-3-clone