Harness Engineering (field guide)

Why pick it

You read the OpenAI essay and want the operational version — playbooks, a thesis index, and an agent context bundle you can point a coding agent at directly. Ryan Lopopolo (author of the original essay) maintains this as the companion repo: "Most people do not know that they can just point their agents at my writing, tweets, podcasts, and talks and improve the output of their agents by 100x."

Harness engineering holds the model and coding agent constant as a black box and improves the two external levers — context and tools — by curating the environment around them.

What's inside

• Thesis index — the core arguments, organized for direct reading or agent routing.
• Playbooks — applied procedures for specific harness-engineering scenarios.
• Agent context bundle — AGENTS.md routes a coding agent to the relevant arguments, cases, and proof. Point an agent at this repo alongside the system it should improve.
• Source library — curated influences and alternate framings, including the [un]prompted conference systems-level framing.
• Evals — evaluation harness for harness-engineering practices.

Why it over the others

vs Harness Engineering (OpenAI essay) — the essay is the story (what happened, what they learned). This field guide is the reference — the same ideas decomposed into playbooks and a thesis index an agent can navigate. Read the essay for the mental model; use this repo to put it to work.

vs Loop Engineering — Loop Engineering is a toolkit of CLIs and patterns for designing agent loops. This field guide is the theory and practice of shaping the environment around any agent — loops are one part of it, but harness engineering also covers context architecture, tool design, authority, and proof.

vs ECC — ECC is a full operating system you install. This is a knowledge artifact you read and route agents through. ECC is the runtime; this is the doctrine.

Who it's for

Engineers who want to shape the environment around their agent — not just prompt better, but build the context, tools, and feedback loops that make agent output reliable. Especially useful for teams adopting the OpenAI essay's approach.

Quick start

Point a coding agent at this repo alongside the system it should improve. AGENTS.md routes the task to the relevant arguments, cases, and proof. For direct reading, start with the thesis index; for an application, choose from the playbooks.

CC BY 4.0. The worker should be able to recover intent, operate the real system, respect authority, prove the outcome, and leave the next run better equipped.