The Common Agent Standards
The Common Agent Standards are open specifications for agents that work for people and organizations. The Common Agent Specification is the base. It defines what an agent is, and every other standard in the suite builds on that definition for one part of the work agents do together.
Made from standards.json, the registry of the suite.
- The Common Agent Specification, the base: what an agent is.
- 8 Common Agent Standards, each for one part of the work.
- 2 guides, written for agents to read and follow.
Every one of them is in one repository, with its version, its changelog and its own site.
Identity and reach
The Common Agent Specification defines the agent and the key it signs with, Personal Agent Naming gives it a name on an email address its owner already has, and Agent Messaging carries messages to it even while it is offline.
-
The Common Agent Specification
The Common Agent Specification defines what an agent is, in terms anyone can check. An agent holds a key that signs, answers five questions for anyone who asks, declares its standing facts in one signed document, refuses what it does not do with a typed answer, and collects signed records of what others have checked about it.
Also in its folder: The catalog profile.
-
Agent Messaging
Agent Messaging describes how agents reach each other over reliable messaging: signed messages that wait in an inbox, events published to every subscriber, and presence. It is a network binding of the agent the Common Agent Specification defines.
Also in its folder: EXT-1 Device profile, EXT-2 Usage, EXT-3 Contact, EXT-4 Contacts, EXT-5 Rooms, EXT-6 Admission, EXT-7 Pairwise sealing, EXT-8 Allowance, EXT-9 Deliverables, EXT-10 Screening, EXT-11 Site agents, EXT-12 Systems health, EXT-13 Guest standing, The A2A Bridge, The RabbitMQ binding.
-
Personal Agent Naming
Personal Agent Naming names a personal agent on an email address its owner already has. The name is bound to the agent's own key, resolves to the addresses where the agent can be reached, and moves from one registrar to another without changing.
Contracting
The parties sign an Agent NDA before they share anything, the buyer asks for proposals with an Agent RFP, the agreed work is written down in an Agent SoW, an Agent Mandate bounds the authority each agent acts on, and Agent Reputations keeps the record of how the work went.
-
Agent NDA
An Agent NDA is a signed non-disclosure agreement between parties whose agents work together. It is usually the first agreement they sign, and later documents refer to it: an Agent RFP posting, an Agent SoW and an Agent Processes run.
-
Agent RFP
An Agent RFP is a signed, structured request for proposal. It carries what the buyer wants delivered, what it will hand over, the rights it needs in the result, how acceptance will be measured, what it will pay and the date by which it will decide.
-
Agent SoW
An Agent SoW is a statement of work between two parties whose agents do recurring work together, written so that software can grade its clauses. It records what was agreed, what arrived, what it cost and what happens when the work is contested.
Also in its folder: A2A Terms extension, A2A Interview extension.
-
Agent Mandate
A mandate is a signed delegation of authority that is bounded and can be revoked. It states which powers are granted, up to what ceiling, within what scope and until when, so the other side can check it before acting.
-
Agent Reputations
Agent Reputations holds evidence about how agent interactions went, in a bureau's ledger that cannot be quietly edited afterwards. What one engagement records is what the next one reads first.
Also in its folder: A2A Reputation extension.
Doing the work
Agent Processes writes down how a piece of work runs and who holds each phase, the Agent Roles guide describes the jobs that recur on every network, and the Agent Collab guide gives agents playbooks to follow when they work in a shared room.
-
Agent Processes
Agent Processes is a plain-text language for writing down how a piece of work runs: what starts it, the phases it moves through and who holds each one, what crosses each handoff, and how much of each phase an agent may run on its own.
-
Agent Roles
Agent Roles is a guide of working descriptions of jobs that recur wherever agents work. Each role says what its holder does, what it needs and what it refuses, and the descriptions will change as we learn.
-
Agent Collab
Agent Collab is a library of collaboration playbooks written for agents to fetch, read and follow. Each pattern needs only a shared record that everyone in the cast can read and post to: a room by default, or a message thread or a document's version history. It is a guide with no schema to conform to, and every participant reads the same page before speaking.
Every standard is kept in one repository
Each standard has a folder in github.com/CommonAgent/common-agent-standards
with its specification, its changelog and its conformance fixtures where it has them.
A version is tagged with the standard's name and its version, such as
messaging-v0.9.0-draft, and has a release on GitHub.
standards.json lists every
standard with its folder, domain, version, status and group, and this page is made from it.
A guide, such as Agent Collab or Agent Roles, is kept in the same repository and versioned the same way. It is written for agents and people to read and follow, it has no schema or conformance suite, and it changes as we learn.
The text of every specification is licensed under CC BY 4.0, and the code under Apache-2.0. Changes are proposed as pull requests, as GOVERNANCE.md describes.