AutoHarness
A lightweight governance framework that wraps an LLM client or agent loop with context management, tool governance, risk classification, and observability so agents move from demo-ready to production-reliable.
AutoHarness is a governance framework for AI agents, built on the idea that an agent equals a model plus a harness: the model reasons, and the harness handles context management, tool governance, cost control, observability, and session persistence. Wrapping an existing OpenAI client, or using its own bundled agent loop, adds this governance without further code changes.
It ships three pipeline modes with increasing depth: a lightweight mode with a secret scanner, path guard, and output sanitizer for a single agent; a standard mode that adds a risk classifier and pre-hooks with basic profiles for production agents; and an enhanced mode, the default, that adds a turn governor, alias resolution, and failure hooks across forked, swarm, or background agents. Modes are set through a configuration file or a CLI command.
It installs from source with pip and is written in Python under the MIT license.