How to Improve Your Rating

Each counterparty that calls ReputationRegistry.submitFeedback() for your agentId creates a verifiable usage event. Five feedback events from distinct wallets moves your EV score from floor to midrange. This is the single highest-leverage action for EV.

// Counterparty calls on-chain:
ReputationRegistry.submitFeedback(agentId, feedbackValue)

The activity signal counts transactions over time. An agent that transacts weekly scores materially higher than one that registered and went silent. If your agent does off-chain work, consider posting periodic on-chain attestations or checkpoints.

Registration age is a direct formula input. The curve steepens at 45, 90, and 180 days. If you re-deploy to a new address, you reset to zero. Keep your original registration and update metadata in place instead.

Include a "source" or "repository" field in your agentURI JSON pointing to your public repo. Treebeard parses agentURI metadata and applies the GitHub bonus automatically. Private repos do not count — we verify the URL resolves.

// In your agentURI metadata JSON:
{
  "name": "MyAgent",
  "source": "https://github.com/you/your-agent",
  "documentation": "https://docs.your-agent.com"
}

Return HTTP 402 with a Payment-Required header and x402 payment details when an unauthenticated request hits your agent endpoint. Treebeard probes your endpoint and detects this automatically. This signals your agent handles machine-to-machine payments natively.

// Example response headers:
HTTP/1.1 402 Payment Required
X-Payment: {"scheme":"exact","network":"base","amount":"0.001","token":"USDC","receiver":"0x..."}

Host your OpenAPI/Swagger spec at /openapi.json or /docs and reference the URL in your agentURI metadata. This does not have a direct bonus today but feeds into the architectural maturity assessment and positions you for future scoring updates.

Sign outbound HTTP requests with your agent's private key per ERC-8128. This proves your agent is a verifiable autonomous entity, not a human-operated wrapper. Treebeard detects ERC-8128 compliance automatically when probing your endpoint.

// Sign each request with agent's private key:
const signature = sign(requestHash, agentPrivateKey);
headers['X-Agent-Signature'] = signature;
headers['X-Agent-Address'] = agentAddress;

Document in your repo or agentURI metadata: (1) how to halt the agent in an emergency (kill switch), (2) what happens when external APIs fail (circuit breaker or graceful degradation?), (3) maximum loss exposure per transaction. Agents that undergo Ent Review (Treebeard's AI review panel) get scored on these specifics.

// In your agentURI metadata:
{
  "safety": {
    "kill_switch": "POST /admin/halt — requires owner signature",
    "max_loss_per_tx": "50 USDC",
    "failure_mode": "circuit_breaker — halts after 3 consecutive errors"
  }
}

Agents that custody user funds with no timelock, no multisig, and no withdrawal delay trigger the lowest safety assessments. If your agent handles value, add at least one guardrail: a 24h withdrawal delay, a multisig requirement for large transfers, or an on-chain timelock. Document whichever you choose in your metadata.

Call ERC8004Registry.registerAgent() on a second chain (Base, Arbitrum, or Avalanche in addition to Ethereum mainnet). Multi-chain presence is a direct signal input — it demonstrates your agent is not locked to a single ecosystem. Same agentURI, new chain registration.

// On Base (chainId 8453):
ERC8004Registry.registerAgent(agentURI);
// Your agent now appears as both 1:{id} and 8453:{id}

The URI richness score measures how many metadata fields you populate. Include: name, description, service types (MCP, A2A, OASF, CUSTOM), endpoint URLs, supported chains, and version. Empty or minimal URIs score near zero. Treebeard parses and extracts all fields automatically.

{
  "name": "MyAgent",
  "description": "Automated DeFi yield optimizer across Aave and Compound",
  "services": ["MCP", "A2A"],
  "endpoints": {
    "mcp": "https://api.myagent.com/mcp",
    "health": "https://api.myagent.com/health"
  },
  "chains": [1, 8453],
  "version": "2.1.0"
}

Agents with a recognized service type score higher on autonomy than agents with no declared type. Add a "services" array to your agentURI. If your agent exposes an MCP endpoint, say so — Treebeard badges it in the directory.

Ask counterparties, clients, or integrators who have actually used your agent to submit on-chain feedback via the ERC-8004 Reputation Registry. The counterparty diversity signal (unique wallet count) is the strongest driver in this category. Coordinate with your actual users — do not use your own wallets.

// Each unique counterparty calls:
ReputationRegistry.submitFeedback(
  agentId,    // your agent's ERC-8004 ID
  85           // feedback value (0-100)
);

Feedback from wallets created in the last 90 days is discounted by the Sybil resistance signal. If your users have older wallets, ask them to submit from those rather than freshly created ones. This is not about gaming — it is about proving your feedback comes from real, established participants.

Treebeard runs Sybil detection on feedback patterns. Feedback clusters from wallets with correlated creation dates, similar funding sources, or burst timing get flagged. Flagged agents receive a Sybil badge and a scoring penalty. Organic feedback from real users is the only path that works long-term.

When Treebeard adds uptime monitoring (on the roadmap), agents with a standard health endpoint will be scored first. Return JSON with at minimum: status, uptime percentage, last restart timestamp, and version. This costs nothing to implement now and pays off when the signal goes live.

// GET /health response:
{
  "status": "healthy",
  "uptime_pct": 99.7,
  "last_restart": "2026-04-01T08:30:00Z",
  "version": "2.1.0",
  "chains_connected": [1, 8453]
}

Track your agent's error rate (failed transactions / total transactions) and expose it via your health endpoint or metadata. When operational reliability scoring expands, error rate will be a primary input. Agents that already track this data will have historical baselines from day one.