| name | remote |
| description | Set up or operate a remote Porcelain host, including exposure, pairing, revocation, an always-on service, updates, and reachability. Use for Porcelain-specific remote administration, not general SSH or network work. |
| license | MIT |
Porcelain remote
This skill covers Porcelain's host launcher and service lifecycle. It does not replace the domain
MCP tools from $companion, and it does not make an SSH target implicit.
Use the current CLI help and the public
remote-access guide as
the operational authority. Do not reconstruct commands from remembered versions.
Working flow
- Confirm the exact host, user account, installed version, and whether the daemon is development,
foreground production, or an always-on service.
- Start with loopback. Add only the narrowest exposure the human needs: LAN on the same trusted
network, Tailscale for a private mesh, or Cloudflare for an explicitly public HTTPS endpoint.
- Pair each client with its own short-lived link. Never distribute the administrator token.
- When the daemon must survive logout or reboot, use the documented user-level systemd service and
linger setup. Keep credentials in a protected environment file rather than command arguments.
- Verify the daemon locally, inspect
share status, then connect a real intended client. A running
process alone does not prove that exposure, pairing, or fallback routing works.
- For updates, restart the documented
@latest service or follow the foreground command shown by
Porcelain. Wait for readiness and verify the reported version before declaring success.
Safety boundaries
- Treat the administrator token and every paired-client token like host login credentials. Do not
print, log, commit, or place them in plugin configuration.
- Porcelain's MCP channel is a profile-scoped local socket on the agent's machine. Do not proxy or
forward it through LAN, Tailscale, Cloudflare, or HTTP to control another Environment.
- Never replace a running daemon, port, service unit, or exposure mode until its owner and purpose
are identified. Stop only the process or service explicitly in scope.
- Network reachability and authorization are separate: a timeout is not a credential failure, and a
401 proves the network path already reached the daemon.
- Report which host, route, client, and version were actually observed, plus any untested fallback.
