awesome-harness-engineering
Awesome list for AI agent harness engineering: tools, patterns, evals, memory, MCP, permissions, observability, and orchestration.
What is awesome-harness-engineering?
Awesome list for AI agent harness engineering: tools, patterns, evals, memory, MCP, permissions, observability, and orchestration.
Overview of awesome-harness-engineering
What it is
This is an awesome list for AI agent harness engineering. It curates resources, patterns, and templates for building reliable scaffolding around AI agents. The harness includes context delivery, tool interfaces, planning artifacts, verification loops, memory systems, and sandboxes. The list focuses on the harness, not the model, and is maintained at the repository https://github.com/ai-boost/awesome-harness-engineering with a companion site at https://agentcraft.dev.
Key features
- Curated collection of canonical essays and guides on harness engineering from organizations like OpenAI, Anthropic, Google, and LangChain.
- Covers design primitives such as agent loops, planning, context delivery, tool design, skills/MCP, permissions, memory, orchestration, verification, observability, debugging, and human-in-the-loop.
- Includes reference implementations, tutorials, demo harnesses, and generators.
- Addresses security, sandbox, permissions, evals, verification, templates, and production infrastructure.
- Multilingual README available in Deutsch, English, Español, Français, 日本語, 한국어, Português, Русский, and 中文.
Use cases
- Designing and improving the scaffolding that surrounds AI coding agents to increase reliability.
- Studying real-world harness patterns from production systems like Meta's REA and Microsoft's Azure SRE Agent.
- Learning how to build agent harnesses for long-horizon tasks, multi-agent orchestration, and scientific workflows.
- Understanding the discipline of harness engineering as distinct from model development.
Getting started
See the repository README at https://github.com/ai-boost/awesome-harness-engineering for the full list of resources, categories, and contribution guidelines.
License and pricing
The license is not stated (NOASSERTION). The project is listed as open source with no pricing information provided.
Based on the project's public README and repository metadata.