HER

The problem

An AI agent's performance depends far less on the underlying model than it seems, and far more on the system built around it: what context it receives, what tools it has, how it manages memory, and what guardrails limit what it can do. That discipline — "Harness Engineering" — didn't have a practical, progressive curriculum for learning it by doing, not just reading about it.

The solution

HER (Harness Engineering Repository) is a modular curriculum combining theory, exercises, and checkpoint projects, using Claude Code (skills, subagents, hooks, memory) as the reference implementation. Modules are completed in order — fundamentals and context engineering, tools and skills, memory and persistence, loops and orchestration — and each one ships with an exercises folder plus separate reference solutions, so learners can self-check without spoiling the challenge.

Every module also includes a sesion-grupal/ folder with two documents: one for whoever facilitates (agenda, timing, discussion prompts) and one for participants. The projects are real checkpoints, with fully working Claude Code configurations of increasing difficulty — from a code-review assistant to, further along, a meta-learning harness.