XNews

Open-Source Harness Engineering Course Offers 14 Lectures and 8 Projects for Coding Agents

Open-Source Harness Engineering Course Offers 14 Lectures and 8 Projects for Coding Agents

An open-source course on Harness Engineering provides 14 lectures, 8 projects, hands-on exercises and reusable templates for building infrastructure around coding agents. The course covers repository instructions, persistent state, verification, scope controls and session lifecycle, with later projects addressing maker-checker loops and graph-based workflows.

Event version 1.

Key insights

Harness surrounds the coding model

A capable coding model is only one part of an agent. Reliable operation inside a real repository across multiple tasks and sessions requires a system that controls actions, remembers progress, verifies work and supplies context for the next session.

Five fundamental harness components

The course breaks the harness into repository operation instructions, persistent state across sessions, verification through tests, linting, type checks and full-pipeline runs, scope controls for focused runs, and session lifecycle for starting, handing off and finishing work.

Projects build progressively

Projects begin by comparing a prompt-only coding agent with one that has repository rules, then add multi-session continuity, runtime feedback and verification before building a complete harness. Later projects cover automated maker-checker loops and graph-based workflows with shared state, parallel execution, rollback, routing and human approval.

Reusable repository templates

The course includes templates for AGENTS.md, init scripts, feature lists, progress tracking, loop state, maker-checker prompts and a harness-creator skill for scaffolding a new setup.

Graph engineering guidance

Later sections explain when a single agent loop stops being sufficient and when graph engineering with separate nodes, shared state, parallel branches and recovery paths makes sense.

Source posts