awesome-harness-engineering

Why we included this project

Harness engineering is the practice of shaping the environment around an AI agent so it behaves predictably, and this curated list gathers the best material on that subject: articles, playbooks, benchmarks, and open-source projects. The organization is the real strength, with sections for context design, memory, guardrails, evaluation, observability, and sandboxed runtimes rather than a flat pile of links. Because it is a reference index rather than a single tool, its value lies in the editorial filter, which keeps generic agent tooling out unless it directly addresses reliability-critical primitives. Teams building agents that must run reliably through long coding and research tasks can use it to find the evaluation frameworks and runtime control layers worth studying before committing to a design.