Free · ICM Method
A step-by-step kit to turn one job you already do into an agent: a single folder that holds how you do the work, the facts it needs, and every result it makes. Start with one real job. Grow it only when the work asks you to.
No vendor lock-in. The folder is yours, and it outlasts every model.
A chat forgets. You re-explain who you are, what the client wants, and how you like it done. An agent folder writes that down once, in plain text files. Then one short sentence starts real work, and the AI reads only the few files that job needs. You can open every file, correct it, and hand the folder to a colleague.
ICM means Interpretable Context Methodology. "Interpretable" means you can open the folder and see why the AI did what it did. Every agent folder has the same five parts.
| Part | What it does |
|---|---|
| Route | AGENTS.md says who the agent is and where to go for each job. A map on one screen. |
| Rules | CONTEXT.md holds the ALWAYS and NEVER rules, the steps, and the facts for a folder. |
| References | references/ is the stable reading material, read only when a rule cites it. |
| Runs | runs/ is one dated, sealed folder per execution: its inputs, outputs, and record. |
| Versions | versions/ logs every change. Every result says which version made it. |
Give a job as much context as it needs to execute, and no more. Too little and the model guesses. Too much and files start to disagree, so results drift. Quality follows an inverted U.
Free. One short form, then the full kit opens right away.
By requesting the kit, you agree that AttestiFi AI, LLC may email you the kit and occasional AI enablement resources. We do not sell your information, and you can opt out at any time. See our Privacy Policy.
They are the part people quote back to us. You unlock both when you get the kit.
Steps 1 to 7 give you a working agent. Steps 8 to 15 make it better when the work asks for more.
