Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

4.6 KB
Raw Blame

Agent Validation Report Format

Overview

An agent validation report is a JSON document that captures the results of testing an agent's GLB model and metadata against a suite of validation rules. Validators (auditors, testing services) generate these reports, answer the agent owner's on-chain request via validationResponse(), and end-users browse them to verify trust.

Schema

{
	"type": "agent-validation-report/0.1",
	"agentId": 1,
	"chainId": 84532,
	"validator": "0x1234567890123456789012345678901234567890",
	"ranAt": "2026-04-17T14:30:00Z",
	"suites": [
		{
			"name": "glb-schema",
			"status": "pass",
			"details": "GLB structure valid, all required chunks present"
		},
		{
			"name": "manifest-integrity",
			"status": "pass",
			"details": null
		},
		{
			"name": "skill-handlers-load",
			"status": "warn",
			"details": "Skill 'sing' references missing dependency @soundfont/npm"
		}
	],
	"verdict": "pass",
	"notes": "Agent passes baseline validation. Minor warnings on external deps."
}

Field Definitions

Field Type Required Notes
type string ✓ Always "agent-validation-report/0.1". Versioned for forward compatibility.
agentId number ✓ The agent token ID (from Identity Registry).
chainId number ✓ The blockchain chain ID (e.g., 84532 for Base Sepolia).
validator string ✓ The Ethereum address of the validator (0x-prefixed hex).
ranAt string (ISO 8601) ✓ UTC timestamp when tests ran. Format: YYYY-MM-DDTHH:mm:ssZ.
suites array ✓ Array of test suite results (see below).
verdict enum ✓ One of "pass", "warn", "fail". Typically "fail" if any suite fails.
notes string ✗ Human-readable summary or recommendations (up to 500 chars).

Suite Object

Each suite in the suites array represents a category of validation tests:

Field Type Required Notes
name string ✓ Unique identifier (e.g., glb-schema, manifest-integrity, skill-handlers-load).
status enum ✓ One of "pass", "warn", "fail".
details string | null ✗ Optional failure/warning message (up to 200 chars). Null if passing.

Validation Rules

  • GLB Schema (glb-schema): Confirms the GLB file structure is valid per glTF 2.0 spec.

    • Fails if: corrupted file, missing required chunks (glTF JSON, bin).
    • Warns if: unsupported extensions, non-standard materials.
  • Manifest Integrity (manifest-integrity): Checks the agent's on-chain metadata JSON.

    • Fails if: agentId, chainId, or registry address mismatch.
    • Warns if: fields are missing or malformed.
  • Skill Handlers Load (skill-handlers-load): Verifies that all declared skills can initialize.

    • Fails if: skill handlers.js has syntax errors or required imports are unavailable.
    • Warns if: optional dependencies are missing but fallbacks exist.

Example

See example-report.json for a complete, valid example.

Hashing & On-Chain Storage

  1. Client-side hash: The dashboard computes keccak256(JSON.stringify(report)) to create the proof hash.
  2. IPFS pinning: The full report JSON is uploaded to IPFS (via Pinata or /api/erc8004/pin).
  3. On-chain record: The contract stores (agentId, verdict, proofHash, proofURI, kind), where:
    • proofHash is the keccak256 hash.
    • proofURI is the IPFS URI (e.g., ipfs://QmXyz...).

Validators can verify the hash by fetching the report from IPFS and recomputing the hash locally.

Trust Model

  • Only Ethereum-validated wallets can submit validations.
  • The validator address is recorded on-chain.
  • End-users can view all validations for an agent and choose which validators to trust.
  • No endorsement or filtering by the three.ws team — raw records from all validators are visible.