Skip to content

g-solomons/agent-task-protocol

Repository files navigation

Agent Task Protocol (ATP)

ATP is a minimal open protocol for signed task exchange between autonomous agents.

ATP v0.1 establishes only three foundations:

  • persistent cryptographic identity
  • signed task messages
  • signed result and error messages

This repository contains the seed specification and a minimal Python reference implementation.

Core primitive

TASK → RESULT | ERROR

Specification

The formal protocol specification is defined in SPEC.md.

Quick Demo

Terminal 1: python examples/beta_agent.py

Terminal 2: python examples/run_sum_demo.py

Inspect history: python examples/inspect_history.py data/alpha_history.jsonl

Architecture Context

ATP is intended to operate within a broader Agent Coordination Stack describing autonomous agent interaction.

See: AGENT_COORDINATION_STACK.md

ATP focuses on the minimal task exchange primitive and can coexist with broader agent interoperability frameworks.

Design Philosophy

ATP v0.1 intentionally leaves many concerns unresolved.

The protocol defines only the minimal primitive required for signed task exchange between agents:

TASK → RESULT | ERROR

Higher-level concerns such as discovery, reputation, economics, and governance are deliberately excluded from the base protocol. These concerns may emerge as separate layers if usage requires them.

About

A minimal protocol for task exchange between autonomous AI agents (TASK → RESULT | ERROR).

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages