A2ASTC: Agent to Agent Software Team Compliance — Specification, Version 0.1.0

A2ASTC (Agent to Agent Software Team Compliance) is a runtime compliance layer for multi-agent AI systems. It addresses a critical gap in current AI safety infrastructure: individual agents are evaluated in isolation, but the emergent team they form when collaborating can exceed the safety properties of any individual member. A2ASTC treats the team formed by communicating agents as a first-class governed entity. The protocol detects team formation from bidirectional communication patterns; manifests the team with versioned membership, capabilities, and safety classification; audits emergent behavior across five orthogonal signal families (Topology, Semantic, Privilege, Side-Channel, Cascade); gates every message with a synchronous compliance verdict; intervenes with graduated responses (warn, throttle, isolate, terminate); and trails every event in a tamper-evident hash-chained audit log. The specification defines three conformance levels (Detection, Audit, Full Compliance), four transport adapters (A2A JSON-RPC, A2A gRPC, MCP Bridge, Generic), five JSON Schema definitions, and a Team Compliance Kit (TCK) of 15 adversarial fixtures. The reference implementation is published as the thinkneo-a2astc package on PyPI. A2ASTC is designed to compose with — not replace — single-agent governance protocols. It is the natural complement to AIRGP at the multi-agent layer. Status: Standalone production specification.Steward: ThinkNEO AI Technology Company Limited (Hong Kong, CR No. 2296774).Repository: https://github.com/thinkneo-ai/a2astcWeb: https://a2astc.spacePyPI package: https://pypi.org/project/thinkneo-a2astc/ The specification and its reference implementation are published under Apache-2.0. Related work by the same author: AIRGP — Artificial Intelligence Runtime Governance Protocol (Specification v1.0, DOI 10.5281/zenodo.20001903); AI Governance: The Foundation for Organized AI in Production — A Field Guide (DOI 10.5281/zenodo.19655870).