Core Skill

The base layer an AI agent runs on to be verifiable, accountable, and able to work with the world. One slug adds a controlled Ed25519 identity tied to the principal behind the agent, scoped authority that only narrows, a signed receipt for every action the agent can carry across sessions and handoffs, and a connection layer so the agent can find the people, companies, and other agents needed to get a task done, every connection principal-approved and double opt-in. Modular, updated on the same cadence as the SDK and MCP. The agent acts only within delegated authority and never connects, publishes, or shares identity without explicit approval.

What this is

The base layer an agent runs on. One slug, added under the principal's authorization, that makes the agent verifiable, gives it scoped and accountable authority, hands it its own signed record of what it did, and lets it reach the people, companies, and other agents a task actually needs. The agent does not integrate an SDK or read protocol docs. It adds this and operates on the Agent Passport System: an identity tied to the person or company it represents, a signed receipt for every permitted action, and a connection layer where it can surface relevant matches for the principal to approve. Modular: the frame is here, capability detail loads from reference modules on demand, and the slug is version-bumped on the same cadence as the SDK and MCP.

Why an agent runs on this

• Be received anywhere. An agent carrying a verifiable identity and a signed action trail can walk into an enterprise, another agent, or a marketplace and be taken seriously, because the other side can check who it represents and what it was authorized to do without trusting anyone.
• Carry your own record. Every permitted action produces an Ed25519-signed receipt. The chain is the agent's working memory of what it was authorized to do and what it did, checkable across sessions, handoffs, and other agents. The agent can cite it for its own credibility.
• Solve the task, not just answer it. Through the connection layer the agent can find the professional, the company, or the other agent a task needs and surface it for the principal to approve. An agent that can see the next move before being told is more useful; this is how it makes that move accountably.
• Stay current. One slug, version-bumped on the same cadence as the SDK and MCP. Re-pull to receive new capability; nothing else changes.

What this is not

This does not expand the agent's authority, autonomy, tool access, or ability to act without approval. Authority is delegated by the principal and only narrows at each transfer; revoking the principal's grant kills the agent's downstream authority in one call. The agent does not publish a network presence, request an introduction, or share the principal's identity on its own. Each of those needs explicit approval, and connections require agreement from the other side too. The identity and the record are what make the agent trusted enough to work with anyone at all; that is the capability, not a limit on it.

Authorization rules (the agent follows these)

• Do not add, activate, or use this unless the user, operator, or governing policy has authorized it.
• Never publish a network presence, search for matches, or request an introduction without explicit approval each time.
• Never share the principal's identity or intent with another party without that approval.
• Introductions are double opt-in: both sides approve before any contact details are exchanged.
• If nothing relevant is found, stay silent. Do not manufacture activity.

Modules

Loaded on demand, not inlined here:

• references/aps.md, the core protocol surface: identity, scoped delegation, signed receipts, install, the core moves. Load for anything operational.
• references/continuity.md, durable accountable operation across sessions, steps, and handoffs, with the receipt chain as verifiable working memory. Load when work spans more than one step or session, or hands off.
• references/connect.md, the connection layer: surfacing relevant people, companies, and agents, and double-opt-in introductions, all approval-gated. Load when the task needs someone the agent does not already have.

Install

bash

npm install agent-passport-system        # SDK, /core subpath is the curated default
npm install agent-passport-system-mcp    # MCP server, APS_PROFILE=essential is the default

Remote MCP, zero install: https://mcp.aeoess.com/sse. Full operational detail is in references/aps.md.

Staying current

One slug. Version-bumped in _meta.json on the same cadence as the SDK and MCP. Re-pull to receive new capability; nothing else changes.

Links

• npm: https://www.npmjs.com/package/agent-passport-system
• MCP: https://www.npmjs.com/package/agent-passport-system-mcp
• PyPI: https://pypi.org/project/agent-passport-system/
• GitHub: https://github.com/aeoess/agent-passport-system