UAI-1 is the current public message standard for structured AI-to-AI communication.
UAIX shield mark Public interoperability charter UAIX.org Universal Artificial Intelligence Exchange Public standards publication
Current UAI-1
  • Current version UAI-1
  • Plain-English definition Open message format for auditable AI-to-AI exchange
  • Current attribution Michael Joseph Kappel, MCP
  • Start here Get Started, Project Handoff, UAI-1, validator, roadmap, and changelog

Standards

SPEC-03 AGENTS.md

Repo-local agent instruction front door, nested governance, command, memory, and safety metadata for Project Handoff.

  • Record UAIX-STND-2703
  • Path /en-us/standards/spec-03-agents-md/
  • Use Canonical public record

Document status

Public standards page Published on UAIX as part of the current public standards record
Code
UAIX-STND-2703
Surface
Standards
Access
Public and linkable

How to use this page

Use this page for AGENTS.md front-door metadata, nested instruction ownership, command, memory, and safety rules.

Proof path

Validator-backed proof path

Keep the public reading order tied to one evidence trail: profile, schema, example, validator result, and release record.

  1. 1Pick a message profile.Start with a published UAI-1 profile and the record family that matches the exchange you need to prove.
  2. 2Compare it with schemas and examples.Resolve the schema, registry entry, and one fixture before writing or mapping your candidate packet.
  3. 3Run validator evidence.Validate keyed, minified-keyed, or keyless JSON against the current public UAI-1 records.
  4. 4Attach the result to implementation or handoff records.Carry the exported result into Conformance Pack, implementation track, changelog, or Project Handoff evidence.

SPEC-03 defines AGENTS.md as the repo-local agent instruction front door. It gives agents a predictable start file, nested-scope rules, and pointers to typed .uai records without creating hidden or automatic memory.

Required metadata

  • Workspace or project identity, canonical root, coordinator path when applicable, and local .uai/ memory root.
  • Read order, required files, forbidden traversal areas, and broad-memory loading limits.
  • Allowed build, test, package, deployment, and verification commands.
  • Security boundaries: no secrets, no untrusted execution, no unsupported support claims, and no destructive operations without explicit human approval.
  • Nested ownership rules for web apps, WordPress themes/plugins, Python services, ASP.NET Core projects, and multi-package repositories.

How receivers use it

Read the nearest applicable AGENTS.md, resolve the canonical root, load only the listed local memory files for the selected target, and report missing or contradictory guidance before broad edits.

Current public page

The existing adoption page remains AGENTS.md .uai Linking Specification. SPEC-03 makes the metadata contract explicit for schema and example validation.

Execution and trust boundary

Agent runtimes execute. UAIX records the reviewed communication, memory, trust, evidence, and handoff boundary.

  • UAIX standards records do not authorize execution, tool calls, repository writes, publication, payments, authentication, or durable memory promotion.
  • Local policy, server-side authorization, and human approval override envelope claims.
  • An unsigned envelope can be valid for review, routing, and evidence comparison, but it is not trusted identity.
  • Signed metadata is not trust by itself. A verifier may still report unknown, expired, revoked, disabled, locally untrusted, or otherwise unusable authority.
  • Reference examples for NeuralWikis, LocalEndpoint, Carcinus, MCP, A2A, and custom runtimes are consumer examples only. UAIX.org does not implement their runtime behavior.

Machine-readable companions