Agent Harness

Agent Harness

Agent Harness is a small Python adapter for running the same tool registry through OpenAI Agents SDK or Anthropic Claude Agent SDK provider implementations. The core package has no runtime SDK dependency; provider SDKs are imported lazily only when that provider is used.

[](https://github.com/evalops/agent-harness/actions/workflows/ci.yml) [](https://github.com/evalops/agent-harness/actions/workflows/bazel-rbe.yml) [](https://www.python.org/downloads/)

What It Provides

• A thread-safe global tool registry via @register_tool.
• JSON Schema generation from Python type hints, including Optional, T | None, list[T], dict, Literal, and multi-type unions.
• A common HarnessConfig and AgentResponse shape across providers.
• Lazy OpenAI and Claude provider adapters.
• Provider comparison helpers for running the same prompt against multiple adapters.
• A built-in optional You.com search tool adapter.

The registry and schema generation work without OpenAI or Anthropic packages installed. Actual model runs require the provider extras and API credentials.

Install

pip install -e .
pip install -e ".[openai]"
pip install -e ".[anthropic]"
pip install -e ".[all]"
pip install -e ".[dev]"

Provider credentials:

export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."

Quick Start

import asyncio

from agent_harness import AgentHarness, HarnessConfig, register_tool


@register_tool(description="Get weather for a city")
def get_weather(city: str) -> str:
    return f"Weather in {city}: sunny, 72F"


async def main() -> None:
    config = HarnessConfig(
        system_prompt="You are a helpful assistant.",
        tool_names=["get_weather"],
        ma