After implementing ICM across several SMBs and solopreneur environments, one pattern became hard to ignore. The system itself was working, but I kept being asked the same questions at the start of each engagement. That repetition was useful because it exposed a design gap. The workspace gave the AI a clear operating structure, but it did not yet give the human user an equally clear way to enter it.
Problem
The recurring problem was not technical failure. It was repetitive onboarding friction. In company after company, I found myself answering the same questions: where should I start, is CONTEXT.md written for me or for the AI, where does source material go, what belongs in the Vault, what am I allowed to edit, and what should I leave alone. Those questions were reasonable. They reflected the fact that an interpretable system is not automatically a comfortable one.
Once the same questions appear across several implementations, they stop being a communication nuisance and start becoming a design issue. The problem was not that users needed too much support. The problem was that the workspace needed a more deliberate human entry point. Without that entry point, too much onboarding knowledge stayed in my head and too much time was spent repeating the same explanations from one client to the next.
Solution
The solution was to add a dedicated 00_HELP/ section at the top of every workspace and make its role explicit in 00_HELP/CONTEXT.md. The rule is simple: this section is for humans, and AI comes here only when specifically asked to do so. That boundary matters because it separates onboarding material from operating context and gives the client one place that is clearly theirs from the beginning.
The section now contains four core assets. The first is CONTEXT.md, which explains the boundary and the purpose of the folder in plain language. The second is the ICM paper in PDF format, so the methodology has a visible source and not just an implementation pattern. The third is a Reference_Material/ folder that contains lessons on every layer of the system, from the original vision to the operating model, including how the workspace is structured and how it should be maintained. The fourth is a Crash_course/ folder containing an interactive HTML course that can be hosted locally with minimal effort and produces a printable certificate at the end.
Results
The effect of 00_HELP/ was practical rather than cosmetic. People started asking questions to their own ICM environment instead of routing every uncertainty back through me. That meant they built working knowledge inside the system itself. It also meant I stopped wasting time answering the same questions repeatedly and could focus on the higher-value parts of the implementation.
There was a second-order benefit as well. Once 00_HELP/ became a stable part of the design, I could pollinate from one company to the next. The human onboarding layer stopped being reinvented each time and became a reusable capability inside the methodology. That made each new implementation easier to adopt without making the workspace looser or more generic.
What is ICM?
ICM, or Interpretable Context Methodology, is a workspace design method developed by Jake van Clief. It comes out of his broader work on practical AI operating systems for real businesses, and anyone who wants to explore the method further should start with his Skool community, cliefnotes.
At its core, ICM uses the filesystem as agent architecture. Numbered folders define stages or work areas. CONTEXT.md files define the operating contract inside each area. Reference material stays separate from working artifacts, and reviewable outputs remain on disk instead of disappearing into opaque application state. In practice, this can mean a consultant builds a workspace where an AI agent pulls source material from one folder, drafts a proposal in another, and leaves behind a structured output that is easy to inspect, review, and reuse.
In practice, however, visible structure is not the same thing as comfortable structure. A consultant, founder, or operator may appreciate that the workflow is explicit, but still feel unsure when opening the workspace for the first time. The architecture may be interpretable, yet the experience can still feel technical. That distinction is what 00_HELP/ addresses.
The Rule That Makes 00_HELP Work
The most important part of the folder is the rule in 00_HELP/CONTEXT.md: this folder is for humans. AI agents should not read it, summarize it, modify it, or rely on it unless the user explicitly asks them to do so.
That boundary matters for two reasons:
- Gives the client a space that is clearly theirs. They can read, learn, and orient themselves without wondering whether the AI is treating their onboarding material as operational context.
- Keeps the production workspace clean. Training material remains training material. Source material remains source material. Working folders remain working folders. In a methodology that depends on clarity of context, that separation is not cosmetic. It is structural.
What Goes Into 00_HELP
Over time, the folder settled into a stable structure because the same needs kept appearing across implementations. At the top level, I now include four elements.
00_HELP/
|- CONTEXT.md
|- ICM_Paper.pdf
|- Reference_Material/
`- Crash_course/
Each of those elements solves a different part of the onboarding problem.
CONTEXT.md defines the boundary. It tells the client what the folder is for, what it is not for, and why the AI should stay out unless explicitly directed. This is where the human-only rule becomes visible and unambiguous.
The ICM paper, included in PDF form, gives the methodology a source. Not every client will read it in full, and that is entirely fine. Its value is that it makes the workspace feel grounded in a coherent method rather than in a collection of ad hoc folder habits. For clients who want to understand the logic behind the structure, the source is present and available.
Reference_Material/ handles the recurring explanatory burden. This is where I place plain-language material on the five-layer method, the role of CONTEXT.md, the distinction between reference material and working artifacts, and the reason review gates matter. The purpose is to reduce repetitive support and let users answer common questions without waiting for me.
Crash_course/ is the most practical part of the folder. This is an interactive HTML course customized to the client’s workspace. I tell clients to open it in VS Code with the Live Server extension, go through the modules, complete the quiz, and receive the certificate at the end. The certificate creates a finish line, and who doesn’t love a finish line? Onboarding becomes a short, bounded learning path that you can share.
Why the Crash Course Matters More Than It Sounds
The crash course changed the adoption dynamic because it translated the workspace from a static structure into a guided experience. A folder map and a few markdown files may be enough for a technically confident user, but most SMB environments do not operate that way. People need a sequence. They need to understand not only what the folders are called, but how to move through them, what the AI is expected to do, and where human review begins.
The course gives them that sequence. It explains the mental model of the workspace, the function of CONTEXT.md, the role of the Vault, and the difference between onboarding material and operational material. It also gives the user a repeatable way to train the next person. That matters in small teams, where continuity often depends on whether knowledge can be transferred without another custom briefing.
The Smaller Friction That Still Matters
One lesson from these implementations is that minor interface frictions accumulate faster than people expect. Markdown is a good medium for ICM because it is portable, inspectable, and easy to maintain. For non-technical users, however, raw markdown can still feel more like code than like documentation. That perception matters because people avoid formats that make them feel unsure.
That is why I also recommend the Mark Down For Humans WYSIWYG extension for VS Code. The extension is not part of the methodology, but it improves the usability of the methodology. If the files feel readable, the workspace gets used. If the files feel forbidding, the human side of the system weakens, even if the underlying architecture is correct.
What Changed After 00_HELP Became Standard
Once 00_HELP/ became a standard part of the workspace, the onboarding pattern improved in predictable ways. Users had a clear first step. The number of repetitive setup questions decreased. The distinction between human orientation, source material, and working operations became easier to maintain. Most importantly, the workspace began to feel less like something the AI owned and more like something the client could navigate with confidence.
That shift matters because trust in an AI workspace is rarely determined by model quality alone. Trust is shaped by whether the user understands where to begin, what the system knows, and where their own judgment still applies. 00_HELP/ does not solve every adoption problem, but it removes a basic and recurring source of friction that should not exist in the first place.
The Broader Point
When people discuss AI adoption, they often focus on models, prompting, orchestration, or automation tooling. Those questions are real, but in small-company environments the first failure point is often simpler. The user does not know how to enter the system. If that entry point is weak, the rest of the architecture carries unnecessary resistance.
That is why I now treat 00_HELP/ as a required part of the workspace. If the filesystem is the agent architecture, then the human side of the system also needs its own architecture. In my implementations, 00_HELP/ has become that layer.
FAQ
What is 00_HELP/ in an ICM workspace?
00_HELP/ is the human onboarding layer in an ICM workspace. It provides orientation, reference material, methodology context, and a client-specific crash course before the user begins operational work.
Why should AI stay out of 00_HELP/?
The folder exists to give humans a protected onboarding area. If the AI treats that material as normal operating context, the boundary between training, production, and source logic becomes weaker than it should be.
What should the folder contain?
At minimum, 00_HELP/ should contain a CONTEXT.md, the ICM paper, a Reference_Material/ folder, and a client-specific Crash_course/ folder. Those four elements cover boundary-setting, methodology source, reusable explanation, and structured onboarding.
Why use an HTML crash course?
Because onboarding works better when it has sequence and closure. A short interactive course is easier to complete, reuse, and hand to the next person than a loose collection of notes.