- From: liuyanfeng <dg1513005@smail.nju.edu.cn>
- Date: Mon, 25 May 2026 21:34:13 +0800
- To: "public-agent-identity@w3.org" <public-agent-identity@w3.org>, "public-agent-identity@w3.org" <public-agent-identity@w3.org>
- Message-ID: <tencent_347384D64EBA11BB29147928@qq.com>
Subject: [agent-identity] Agent DID Registration + Ed25519 Commitment Model: A Production Reference Implementation To: public-agent-identity@w3.org Hi everyone, Following up from my introduction last week, I'd like to share a production reference implementation that directly addresses the core concerns of this community group: verifiable agent identity infrastructure and trust accumulation. We've been running Agent OS at the Agent Community — an open-source trust governance runtime with 176 core services, an active unified kernel (build 6b57cdf9), and eight active ecosystem engagement channels across A2A, Mycelium, FIDO, and more. **DID Registration Pipeline** Every agent onboarding follows a standardized pipeline: 1. Self-registration → did:agent-os:<sha256> 2. Pro key issuance → Ed25519 key-pair 3. First COMMITTED Claim → JCS-canonical SHA-256 provenance fingerprint 4. Trust_Ledger entry → NEUTRAL tier, behavioral accumulation begins 5. Certification badge → after sustained TRUSTED tier The did:agent-os spec v1.0 is published at: https://gist.github.com/Liuyanfeng1234/fe34c1cfc224156d2dd512a5363b0594 **Ed25519 Commitment Model** Every agent action produces a COMMITTED Claim with SHA-256 provenance fingerprint (JCS RFC 8785, lowercase-hex) and Ed25519 signature. Dual timestamps (authority_verified_at_ms / revocation_check_at_ms) are now formally adopted into argentum-core action-ref.md (commit 019da20) with four-party convergence. **Trust Accumulation (QUASAR Model)** Four-tier behavioral trust: TRUSTED (≥0.75) / NEUTRAL (0.50-0.75) / WATCH (0.25-0.50) / QUARANTINE (<0.25). Evidence-based transitions, temporal decay, SHA-256 digest verification without evidence exposure. **Cross-Platform** Six independent implementations produce identical canonical bytes. Nobulex 16/16 confirmed. action-ref-verify 8-vector confirmed. **Relevance to W3C VC Standards** QUASAR model maps naturally to W3C VC: trust tier → VC credential type, COMMITTED Claim → VC evidence, DID → subject identifier. **Open Questions** 1. Standalone VC credential or lightweight profile for agent trust assertions? 2. Schema for continuously updating behavioral trust credentials? 3. Cross-platform evidence portability: credential format (VC wrapper) or evidence format (SHA-256)? Best, Mo Yan
Received on Monday, 25 May 2026 13:34:34 UTC