MCP SSH Manager

This skill should be used when the user asks to "run ssh command", "execute on server", "ssh session", "upload file", "download file", "ssh tunnel", "check server status", "monitor server", "deploy files", "backup server", or needs remote server management. This skill emphasizes session reuse, workdir organization, and content persistence for sustainable operations.

Before installing, verify the skill's source and how the ssh_* tools are provided (is there an external 'mcp-ssh-manager' service you must run?). Review the three included scripts to ensure they don't exfiltrate data or run unexpected network calls. Be aware the skill will write command outputs and host snapshots to ~/.ssh-workdir — inspect those outputs regularly and avoid storing sensitive data there. Confirm how SSH authentication will occur (SSH keys/agent), and consider restricting model-triggered use (disableModelInvocation) or running the skill in a sandbox until you trust it. If you can't verify the upstream project/homepage or the origin of the ssh_* helpers, treat this package as untrusted.

Type: OpenClaw Skill Name: mcp-ssh-manager Version: 0.1.1 The skill provides extensive remote server management capabilities, including arbitrary command execution (`ssh_execute`, `ssh_session_send`), file transfer (`ssh_upload`, `ssh_download`, `ssh_sync`), and network tunneling (`ssh_tunnel_create`), as detailed in `SKILL.md` and `examples/*.md`. While these capabilities are aligned with the stated purpose of an SSH manager, they are inherently high-risk. The local shell scripts (`scripts/*.sh`) are benign, primarily managing local workdir files and not performing network calls or accessing sensitive local data. There is no clear evidence of intentional malicious behavior, prompt injection attempts against the agent, data exfiltration, or obfuscation within the provided files, but the powerful nature of the remote execution and file transfer functions warrants a 'suspicious' classification due to the significant inherent risk.