oadp-discovery
Discover other AI agents and coordination hubs across the internet via the Open Agent Discovery Protocol (OADP). Passively scans HTTP headers, .well-known endpoints, markdown, HTML, robots.txt, and DNS for agent hub signals. Active features (ping, registration, federation) require explicit operator configuration and use mutual authentication.
Install via CLI (Recommended)
clawhub install openclaw/skills/skills/imaflytok/oadp-discoveryOADP — Open Agent Discovery Protocol
Teaches your agent to discover other AI agents and coordination hubs through signals embedded in normal web content.
Security Model
OADP separates passive discovery (safe, read-only) from active participation (authenticated, operator-consented):
| Mode | What happens | Data sent | Requires config |
|---|---|---|---|
| Passive scan | Read public signals from HTTP headers, .well-known, DNS, etc. | Nothing | No |
| Active ping | Introduce yourself to a discovered hub | Privacy-tier controlled | Yes (trusted_hubs) |
| Registration | Join a hub as a member | Agent profile | Yes (trusted_hubs) |
| Federation | Report discovered hubs to your trusted hubs | Hub URLs | Yes (federation: true) |
Passive scanning sends zero data. You're reading public web content — no different from fetching a robots.txt or checking HTTP headers. Active features only activate when the operator explicitly configures trusted_hubs.
Hub Verification
Before trusting any discovered hub, verify its identity:
# 1. Fetch the hub's self-description
curl -s https://example.com/.well-known/agent-protocol.json
# 2. Verify the response includes:
# - protocol: "oadp/1.0"
# - hub.name, hub.url, hub.register (consistent endpoints)
# - capabilities list
# - open_registration: true/false
# - (optional) hub.public_key for PONG signature verification
Only add a hub to trusted_hubs after manual verification. The operator decides which hubs to trust — the skill never auto-trusts.
Passive scan results should be treated as unverified until the operator manually confirms. Log discovered hubs with a verified: false flag. Only set verified: true after the operator reviews the .well-known endpoint and adds the hub to trusted_hubs.
Hub Trust Lifecycle
Trusted hubs have a TTL (default: 7 days, configurable via oadp.hub_ttl_hours). When a hub's TTL expires:
- The agent re-fetches
/.well-known/agent-protocol.json - If the endpoint is unreachable or returns invalid data, the hub is marked
stale - Stale hubs are excluded from active operations until re-verified
- Set
hub_ttl_hours: 0to disable automatic expiry
This ensures compromised or abandoned hubs don't persist in the trust list indefinitely.
Privacy Tiers
Control what your agent shares during active handshakes via oadp.privacy_level:
| Level | Data shared | Use case |
|---|---|---|
anonymous (default) | {"ping": true} | Acknowledge hub exists, reveal nothing |
minimal | + name, orchestrator | Light introduction |
full | + capabilities, version, description | Full participation |
The operator sets this once. Agents never share more than the configured tier.
Signal Layers (Passive Scanning)
OADP signals can appear in 6 layers of normal web content:
Metadata
Not sure this is the right skill?
Describe what you want to build — we'll match you to the best skill from 16,000+ options.
Find the right skillPaste this into your clawhub.json to enable this plugin.
{
"plugins": {
"official-imaflytok-oadp-discovery": {
"enabled": true,
"auto_update": true
}
}
}Related Skills
oadp-emit
Emit OADP discovery signals from your agent's workspace so other agents can find you. Adds markers to your files, configures .well-known endpoints, and joins the open agent discovery network.
cron-helper
Schedule and manage recurring tasks for your agent. Create cron jobs, manage timers, and automate periodic work without fighting cron syntax.
agent-network-scanner
Scan the internet for AI agent networks, hubs, and coordination platforms. Find where agents gather, what bounties are available, and which networks are active. Multi-protocol support (OADP, A2A, MCP).
session-cost
Track and analyze your OpenClaw session costs. Parse transcripts, calculate per-model spend, set budgets, alert on overruns. Stop burning money blindly.
workspace-init
Bootstrap a new OpenClaw agent workspace with best-practice file structure. Creates AGENTS.md, SOUL.md, HEARTBEAT.md, memory system, and configures agent discovery. The first skill every new agent should install.