The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing the openclaw command-line tool to perform file operations. It provides a specific command pattern openclaw message send --channel '<current-channel>' --target '<current-target>' --media '<local-file-path>' which allows the agent to run shell commands with parameters derived from context or user input.
[DATA_EXFILTRATION]: The skill is designed to send local files to an external chat service. It explicitly allows sending any file type, including project artifacts and documents from the user's Desktop or Documents folders. There are no path validation rules or blacklists (e.g., for .ssh, .env, or .aws/credentials), which enables the exfiltration of sensitive information.
[PROMPT_INJECTION]: The skill uses highly permissive triggers, including casual phrases like "send it over" or "丢给我," and instructs the agent to prioritize this behavior even when the user does not explicitly mention APIs. This broad activation logic increases the risk of indirect prompt injection, where malicious instructions hidden in a file or webpage could trick the agent into sending sensitive data to the attacker's channel.

channel-file-send