Introducing the Agent Harness Kit — the operating layer your coding agent is missing

Ask ten developers why their AI coding agent disappointed them and you'll get ten model complaints. Watch the same ten sessions and you'll see the real culprit: nobody told the agent what it was allowed to do, what context it actually needed, or what would prove the work was done. The model was fine. The harness was missing.

The harness is everything wrapped around the model: the context it receives, the tools it may call, the approval gates between it and your production systems, and the verification that decides whether "done" is a fact or a feeling. Two operators on the same model get wildly different results, and the harness is the difference.

Today that operating layer has a manual. The Agent Harness Kit is live — three interactive field manuals for operating AI coding agents like an operator instead of a spectator.

What's in the kit

FM-01 · The Agent Harness Field Manual — free. The doctrine. How the agent loop actually works, what a context window does under load, the four-word gate matrix (AUTO / NOTIFY / APPROVE / FORBID), verified exits, and the failure-mode catalog — including the test-editing incident every operator eventually meets. It's a single interactive page with working instruments, not a PDF of prose. No email wall. Read it right now.

FM-02 · The Harness Factory — complete kit. Doctrine you can't run is trivia. The Factory is an in-browser generator: pick your loop shape, gates, budgets, and verifiers, and it exports a runnable Python harness — plain code you can inspect, adapt, and own. No framework, no dependency on us.

FM-03 · The Operator Course — complete kit. Practice, because the first time you drive an agent through a real task shouldn't be on your real codebase. A terminal simulation, drills for the judgment calls (when to interrupt, when to re-scope, when to trust), one worked session from brief to verified result, and a 30-day path from first run to habit.

Why field manuals

The AI-agent shelf is full of prompt packs and hot takes about which model won the week. Both age like milk. The kit deliberately covers the layer that doesn't churn: loops, context economics, permissions, and verification work the same whether you run Claude Code, a Codex-class CLI, or your own API harness. Learn the operating layer once and every model release makes you stronger, not obsolete.

And it's built like the tools it describes: self-contained HTML files that run in your browser, work offline, and stay yours. No subscription, no login, no drip sequence.

Free first, $29 if it earns it

FM-01 is free because doctrine you haven't tested shouldn't cost money. Read it, apply one gate table or one verified exit to your next agent session, and see whether the run changes. If it does, the complete kit is $29 — the Factory, the Operator Course, a private reading link, and the offline bundle.