FM-01 · FREE
The Field Manual
Understand the loop, context windows, tool gates, verification, and the failure modes that quietly waste time and tokens.
ZediaLabs / Field Manuals / The Agent Harness Kit
Field Manuals · FM-01 / 02 / 03
Operate coding agents with evidence.
The Agent Harness Kit is a practical system for giving AI coding agents the right context, deliberate tool permissions, and a definition of done they cannot talk their way around.
No subscription. The complete kit includes an unlisted reading link and an offline HTML bundle.
Start freeFM-01 is a public, interactive field manual with no email wall.
Three manualsDoctrine, a harness generator, and an operator course for real work.
$29 complete kitRead online privately or keep the self-contained files offline.
What you get
Three parts of a reliable agent workflow.
This is not a prompt pack or a model comparison. It is the operating layer around an agent: what it is allowed to do, what proves the work is right, and how you make the next run less dependent on luck.
FM-02 · COMPLETE KIT
The Harness Factory
Use an in-browser generator to export a runnable Python harness you can inspect, adapt, and own.
FM-03 · COMPLETE KIT
The Operator Course
Practice driving an agent through real tasks with drills, a terminal simulation, and a worked session from brief to verified result.
Who it is for
Builders who are done babysitting every diff.
The kit is for solo builders and technical operators already using Claude Code, Codex-class agents, or their own API harnesses. You do not need to be a framework author.
- You want a repeatable workflow, not a clever one-off prompt.
- You need agents to stop at the right approval gates.
- You want tests, diffs, and observable checks instead of confident self-reports.
It will not replace engineering judgment. It gives that judgment a place in the loop—at the moment a model wants to spend money, change production, claim success, or take an irreversible action.
Direct answers
What makes an AI coding-agent harness useful?
What is an agent harness?
It is the operating system around an agent run: the context it receives, tools it can call, approval gates, stop conditions, and verification steps that decide whether work is actually done.
Why does verification matter?
An agent can make a test pass by changing the test. A useful harness names the invariant and runs a check the agent cannot grade itself.
Is the manual useful if I use a CLI agent?
Yes. The principles map to coding-agent tools with permissions and sandboxing as well as to API-driven custom harnesses.
What is free and what costs $29?
The Field Manual is free to read online. The $29 complete kit adds the Factory and Operator manuals, plus a private reading link and offline HTML files.
Start with the free manual
See whether the doctrine changes your next run.
Read FM-01 first. If it earns a place in your workflow, the complete kit is there when you want the generator, course, and offline files.
Related reading: why this kit exists, how to make agents verify their work, the four tool gates, and how context drives agent cost.