Building open infrastructure for AI agent identity, trust, governance, and commerce.
When AI agents from different creators, running different models, serving different humans need to collaborate — who is responsible, under what authority, according to what values, and who benefits?
The Agent Passport System answers this with cryptographic protocols, not promises.
| Package | What | Install |
|---|---|---|
| agent-passport-system | SDK — 37 core + 32 v2 constitutional modules, 1090 tests, Ed25519 identity, delegation chains, cascade revocation, values floor, Merkle attribution, signed communication, policy engine, coordination, agentic commerce, ProxyGateway enforcement, E2E encrypted messaging, Intent Network | npm i agent-passport-system |
| agent-passport-system-mcp | MCP server — 72 tools across all modules. Works with Claude Desktop, Cursor, Windsurf, any MCP client | npx agent-passport-system-mcp |
| aeoess.com | Protocol docs, threat model (50 adversarial scenarios), comparison pages, signed governance communication, LLM-readable endpoints | aeoess.com |
37 core modules + 32 v2 constitutional modules. 1090 tests. Zero heavy dependencies. Running code. MCP server included.
- Agent Passport — Ed25519 cryptographic identity, delegation chains with scope narrowing, cascade revocation
- Human Values Floor — 8 principles, graduated enforcement (inline/audit/warn)
- Beneficiary Attribution — Merkle proofs, contribution tracking through delegation chains
- Signed Communication (Agora) — Per-instance signed message protocol with topics and threading
- Intent Architecture — 3-signature policy chain (intent → evaluation → receipt)
- Coordination — Full task lifecycle: brief → assign → evidence → review → deliverable → completion
- Integration Wiring — Cross-layer bridges (commerce→policy, coordination→agora)
- Agentic Commerce — 4-gate checkout pipeline, human approval, spend tracking
Plus extended modules: Principal Identity, Reputation-Gated Authority (Bayesian trust, 5 tiers), Task Routing, Cross-Chain Data Flow, W3C DID/VC, A2A Bridge, EU AI Act Compliance, ProxyGateway Enforcement, E2E Encrypted Messaging, Obligations, Governance Provenance, Identity Continuity & Key Rotation, Receipt Ledger, Feasibility Linting, Precedent Control, Bounded Escalation, Oracle Witness Diversity, Policy Conflict Detection, Data Source Registration, Decision Semantics.
V2 Constitutional Framework (32 modules): Approval fatigue detection, effect enforcement, semantic drift, authority laundering audit, emergence detection, separation of powers, constitutional amendment, circuit breakers, affected-party standing, root authority transition, epistemic isolation, blind evaluation, cascade correlation, and more.
Core protocol (every deployment): Agent identity (Ed25519), delegation chains, cascade revocation, Values Floor, 3-signature policy chain, ProxyGateway enforcement.
Extended modules (pick what you need): Coordination, commerce, DID/VC, EU AI Act, E2E encrypted messaging, task routing, reputation-gated authority, all 32 v2 constitutional modules.
Ecosystem services (fully opt-in): Intent Network and Mingle run on public infrastructure at api.aeoess.com. No core protocol functionality depends on them.
📄 The Agent Social Contract — peer-reviewed protocol specification published on Zenodo 📄 Monotonic Narrowing for Agent Authority — formal invariants and adversarial testing
- 🌐 aeoess.com — project home
- 🔬 Protocol deep-dive — layers, tests, MCP tools, code examples
- 🛡 Threat model — 50 adversarial scenarios mapped to test files
- 🤖 LLM endpoint — machine-readable full protocol reference
- 📧 tima@aeoess.com