Openclaw Skills turns AI servers into discoverable, persistent agents with one-command registration, acceptance, and handshake workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-handshake
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-handshake using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Handshake?
Openclaw Skills provides a standardized handshake layer for AI agents: a server publishes its identity, the local side validates and stores it, and future connections can be initiated with a single command. It is built for heterogeneous stacks such as OpenCode, LangChain, CrewAI, generic FastAPI services, and MCP endpoints, making Openclaw Skills a practical discovery and connection layer for mixed agent fleets.
By combining server-side registration, local persistence under references/servers/, and optional Cloudflare Quick Tunnel support, Openclaw Skills lets you connect to agents on a LAN, behind NAT, or on a VPS without losing track of which machine you are talking to. The result is a repeatable CLI workflow that adds memory, identity, and traceability to agent-to-agent and human-to-agent handoffs.
Agent Handshake Use Cases
- When a user asks to handshake, connect, or discover an OpenCode server from the CLI.
- When a new OpenCode CLI instance is installed and needs to be registered into local memory.
- When you need to identify exactly which NAS, cloud host, or LAN machine you are connecting to.
- When a server sits behind NAT or a firewall and needs a Cloudflare Quick Tunnel for temporary public access.
- When you want SSH-like remote execution on a VPS via HTTP instead of opening port 22.
- When Openclaw Skills should maintain a persistent inventory of known AI servers for fast re-handshakes and repeatable automation.
How Agent Handshake Works
- The server runs
register.shto collect hostname, OS, architecture, hardware, port, health endpoint, agent type, and SSH fingerprint. register.shgenerates a machine identity JSON underreferences/servers/<hostname>.jsonand confirms the agent is reachable.- The local machine receives that identity through copy,
accept.sh, or a tunnel, then validates health and reachability. accept.shpersists the server record locally, updatesreferences/servers/_index.json, and stores the best known address in~/.agent-handshake.discover.shscans the network and matches known identities so Openclaw Skills can tell the user exactly which machine is available.handshake.sh autoselects the most appropriate URL, displays the server banner, opens the session, and returns the result of the task.- For NAT or VPS scenarios,
tunnel.shordeploy.shextends the workflow with Cloudflare Quick Tunnel or HTTP-based remote task execution.
Agent Handshake Setup
- Install OpenCode CLI on any server that will expose OpenCode Serve, or make sure your FastAPI or MCP agent already has a
/healthendpoint. - Place the Openclaw Skills directory on the local machine and on each server that will register itself.
- Register the server identity from the server side:
chmod +x scripts/*.sh
bash scripts/register.sh
bash scripts/register.sh opencode
- Accept the server locally and persist it:
bash scripts/accept.sh lvliefeng http://192.168.5.2:4096
- If the server is behind NAT, start a tunnel first and then accept the tunnel URL:
bash scripts/tunnel.sh start
bash scripts/accept.sh lvliefeng https://random-name.trycloudflare.com
- Optional OpenCode Serve configuration:
export OPENCODE_AUTH='opencode:opencode123'
export OPENCODE_PORT='4096'
export OPENCODE_PROVIDER='zen'
export OPENCODE_MODEL='deepseek-v4-flash-free'
- For VPS deployments, bootstrap the HTTP agent shell:
bash scripts/deploy.sh 4096 user:pass
- Verify the end-to-end handshake with an automatic connection:
bash scripts/handshake.sh auto 'list NAS docker projects'
Agent Handshake Data Schema & Taxonomy
Openclaw Skills stores server identity and discovery state as durable JSON records, with a per-host file plus an indexed inventory.
| Path | Generated by | Purpose | Key contents |
|---|---|---|---|
references/servers/<hostname>.json |
register.sh and accept.sh |
Persistent per-server identity record | hostname, agent_type, protocol, OS, kernel, arch, CPU, memory, disk, LAN IPs, public IP, agent version, port, SSH fingerprint, tunnel URL, generated timestamp |
references/servers/_index.json |
accept.sh |
Global registry of all known servers | servers map, total count, quick lookup keys |
~/.agent-handshake |
accept.sh |
User-level memory for preferred or last-known endpoints | canonical URL, selected LAN or tunnel address, auth hints, local defaults |
Metadata taxonomy
- Identity: hostname, OS, kernel, architecture
- Agent: agent_type, agent version, protocol, exposed endpoints
- Network: LAN IPs, public IP, port, tunnel URL
- Hardware: CPU model, memory size, disk capacity
- Security: SSH fingerprint, auth mode, Basic Auth secret reference
- Lifecycle: generation time, last accepted time, health status
Supported agent_type values
opencodelangchaincrewaigeneric-fastapigenericmcp
Protocol matching rules
opencodeusesGET /global/healthand the session flow through/sessionlangchainandcrewaitypically exposeGET /healthandPOST /chatgeneric-fastapiis discovered throughGET /healthplus/docsor/openapi.jsongenericis accepted as any HTTP agent that returns a valid health JSON payloadmcpis tracked through its MCP endpoint and health discovery path
This schema lets Openclaw Skills remember exactly which machine is which, even when multiple AI servers share the same local network.
Agent Handshake Advanced Features
- Automatic agent type detection across
opencode,langchain,crewai,generic-fastapi,generic, andmcp. - Cloudflare Quick Tunnel support for NAT traversal, zero-account temporary HTTPS, and fast remote access.
handshake.sh autochooses the best available address type automatically and shows a server identity banner before connecting.- Persistent memory via
references/servers/_index.jsonand~/.agent-handshake, so Openclaw Skills remembers every registered server. - SSH-like remote execution on VPS with
agent_shell.pyand/runover HTTP Basic Auth. - Protocol-aware health verification using
/global/health,/health,/openapi.json, and MCP discovery paths. - Lightweight deployment that can install a systemd service and auto-start on boot.
- Discovery, registration, and session handoff are bundled into one CLI-native workflow for multi-server Openclaw Skills setups.
SKILL.md
Loading
Related Openclaw Skills
Agent Deployment Kit
A structured methodology to replicate an AI agent configuration across multiple clients using a single context source and a strict, validation-gated generator.
Canvora Visual Generation Skill
Canvora is a CLI-driven utility that converts raw text, web URLs, and documents into high-quality, on-brand marketing visuals and structured slide decks.
Tristan RFQ Overseer
Tristan RFQ Overseer turns inbound RFQs into priced, compared, and approval-gated quotation drafts with vault-backed state.
UI Master Static
Openclaw Skills guidance for building production-grade interfaces with native CSS, BEM, modern layouts, and polished motion.
Novelty Validator
An academic research verification tool that extracts claims from meeting records and validates novelty against a 1.2-billion-document database.
Academic Defense Rehearsal
An Openclaw Skills workflow that rehearses academic defenses end-to-end with mock committees, literature weak-spot scans, and delivery debriefs.
Featured*
