The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the lh connect command, which allows remote agents to execute tools on the user's local machine through a WebSocket gateway. This capability is explicitly described in references/device.md as a way for agents to "execute local tools" via commands sent from the LobeHub service.\n- [REMOTE_CODE_EXECUTION]: The lh skill install command allows the agent to install and execute code from external sources, including GitHub repositories and ZIP archives. This mechanism enables the runtime acquisition and execution of unverified third-party content on the host system.\n- [EXTERNAL_DOWNLOADS]: Several commands facilitate downloading content from remote sources. lh skill install fetches code from GitHub and external URLs, while lh file upload and lh doc upload allow the agent to retrieve and process files from arbitrary web addresses.\n- [DATA_EXFILTRATION]: The skill includes capabilities to read and transmit local data. lh kb upload and lh file upload allow sending local files or content from specific URLs to remote LobeHub storage, providing a potential path for exfiltrating sensitive local documents if the agent is compromised.\n- [CREDENTIALS_UNSAFE]: Multiple commands instruct the agent to handle sensitive credentials as plain-text CLI arguments. Examples include --api-key for provider configuration, --bot-token for chat platform integrations, and --token (JWT) for gateway authentication. Providing secrets via command-line flags can expose them in shell history, process listings, and logs.

lobehub-cli