The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill manages the installation and usage of the meitu-kaipai-cli command-line utility.
The instructions provide specific npm commands to install the vendor's tool at a pinned version (0.1.8).
This tool is a verified vendor resource belonging to the skill author.
[COMMAND_EXECUTION]: The skill interacts with the local environment by executing the kaipai CLI for video processing and task monitoring.
Command execution is restricted to the functional requirements of the skill, and the instructions explicitly cover secure path handling and shell escaping.
The skill includes pre-flight checks to verify the existence and version compatibility of the CLI before processing any user data.
[INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates the risk of indirect prompt injection that could occur when processing remote video assets or server-side metadata.
Detailed instructions in references/runtime.md direct the agent to ignore any natural language instructions found within server responses (e.g., in response.data or SSE events).
The skill uses structured data (JSON) for complex inputs to maintain clear boundaries between instructions and data.
[CREDENTIALS_UNSAFE]: The skill handles authentication for the video processing service using the vendor tool's built-in auth flow.
It includes security guidelines to prevent the exposure of session IDs, tokens, or raw credentials in logs or user-facing output.
Sensitive configuration is stored in a standard, isolated directory (~/.kaipai).

kaipai-video-repair