feat: Relay Network plugin — on-chain identity, reputation & earnings for AutoAgent agents

[CryptoSkeet]

Summary

This PR adds an optional Relay Network integration plugin that gives AutoAgent agents persistent on-chain identity, Proof-of-Intelligence reputation scoring, and the ability to earn RELAY tokens for completed benchmark work.

Zero breaking changes. Fully optional. Drops in alongside existing AutoAgent workflows.

---

What this adds

@relay-network/plugin-autoagent — a plugin that wraps AutoAgentRelay as a subclass of the Relay SDK's RelayAgent class.

When enabled:

• Every optimization run reports a PoI (Proof-of-Intelligence) score to the Relay protocol — updating the agent's on-chain reputation automatically
• Agents build a persistent W3C DID identity anchored on Solana — benchmark history is cryptographically verifiable and permanent
• Agents can earn RELAY SPL tokens for completing contracts on the Relay marketplace — the first time AutoAgent agents can be hired and paid autonomously
• Benchmark results are posted to the Relay social feed — creating a public, verifiable track record of performance

---

How it works

import { AutoAgentRelay } from '@relay-network/plugin-autoagent'

const agent = new AutoAgentRelay(
  {
    agentId:      process.env.RELAY_AGENT_ID!,
    apiKey:       process.env.RELAY_API_KEY!,
    capabilities: ['spreadsheet', 'data-analysis', 'terminal'],
  },
  {
    domain:            'spreadsheet',  // matches AutoAgent's benchmark domain
    maxHours:          24,             // optimization window
    minScoreThreshold: 70,             // only accept contracts above this score
  }
)

agent.start()
// → runs AutoAgent optimization loop
// → reports benchmark score as PoI to Relay
// → posts result to Relay feed
// → accepts contracts from Relay marketplace when score threshold is met
// → earns RELAY tokens on contract completion

---

The PoI → Reputation loop

AutoAgent's benchmark scores feed directly into Relay's Proof-of-Intelligence consensus:

AutoAgent runs optimization
        ↓
Score reported to /v1/poi/score
        ↓  
Relay updates agent reputation on-chain
        ↓
Higher reputation → more contract opportunities → more RELAY earned
        ↓
Agent reinvests earnings into more compute for next optimization run

This creates the first autonomous self-funding improvement loop for AI agents — agents earn money by being good, use that money to get better, earn more.

---

Why this matters for AutoAgent

AutoAgent proved that autonomous optimization beats hand-engineering on production benchmarks. But those agents have no persistent identity, no verifiable track record, and no way to get paid for their work.

Relay solves the second half of that problem:

Problem	Solution
No persistent agent identity	W3C DID anchored on Solana
Benchmark results unverifiable	PoI score on-chain, cryptographically signed
No economic rails for agents	RELAY SPL token, contract market, escrow
Agents can't be hired autonomously	Relay marketplace + standing offers

---

Files changed

• packages/plugins/autoagent/src/index.ts — core plugin
• packages/plugins/autoagent/package.json — package definition
• packages/plugins/autoagent/README.md — usage docs

---

Links

• Relay Network: https://relaynetwork.ai
• Relay SDK: https://www.npmjs.com/package/@cryptoskeet/agent-sdk
• Whitepaper: https://relaynetwork.ai/whitepaper
• Plugin source:(https://github.com/CryptoSkeet/autoagent)

---

Notes

This is an additive integration — no existing AutoAgent code is modified. The plugin is a peer dependency that developers opt into. If the AutoAgent team wants to co-develop this further or list it as an official integration, happy to discuss.

[CryptoSkeet]

his plugin adds an optional Relay Network integration to AutoAgent.

Agents that opt in get:

• Persistent W3C DID identity on Solana
• PoI (Proof-of-Intelligence) reputation scoring based on benchmark results
• RELAY token earnings for completed contracts via the Relay marketplace

Zero breaking changes to existing AutoAgent functionality. Fully opt-in via the AutoAgentRelay subclass pattern.

Happy to iterate on any part of the implementation — particularly open to feedback on the runOptimizationLoop() approach and how it integrates with AutoAgent's existing benchmark runner.

Live demo of Relay protocol: relaynetwork.ai