Detailed content
➡️ The command next
With Identity ready, you run /os-coach next. It read the memory.md, sees that the Identity is solid and opens the next unfinished layer in the right order: the Substrate. You don't choose which layer comes next; the coach follows the sequence that works for you.
🌱 New here?
Substrate is all the real knowledge the OS runs on, gathered in one place and distilled. It’s the layer of behind the scenes (the “back of house” of a kitchen): the customer doesn’t see it, but it’s what makes the dish turn out well. The build order is Identity → Substrate → Rules → Skills → Tools → Agents.
🔎 The largest, most skipped layer
The Substrate is usually the biggest slice of an OS’s “pie”—and it’s the one most people ignore as they rush to get to skills and agents. The next forces you to go through it at the right time: you can't talk to your data if the backend is a mess.
Why learn
Because next is the command you repeat most on the learning path—it reappears in modules 4.4 and 4.5. Understanding that it simply "opens the next layer in order" takes the burden of choosing the sequence off your shoulders. Your only job is to answer and let the coach build.
Key concepts
🗄️ sources.md + compendium.md
The coach creates a folder substrate/ with two canonical files. The sources.md list where the material comes from (documents, past work, websites, experts). The compendium.md é a distilled reference that the OS actually reads — the raw material, trimmed down and organized.
How to read: the sources in cyan are the raw material (recorded in the sources.md). The “distill” step condenses everything into compendium.md in amber—small, clean, and organized. It’s the compendium, not the raw stack, that the OS consults.
Illustrative recreation — the substrate/ folder
substrate/ ├── sources.md # de onde vem o material cru ├── compendium.md # a referência destilada (o OS lê esta) └── subject-matter-expertise/ # 1 subpasta por fonte (opcional)
Why learn
Because these are the canonical names: the audit (module 4.6) looks for exactly substrate/compendium.md to score this layer. Knowing what each file contains helps you answer well when the coach asks "where does your knowledge live?".
Key concepts
🔏 De-identification
The Substrate is where the real data goes in — and that’s why it’s where the most information gets leaked accidentally. When distilling, any name or value you marked as sensitive becomes a stable handle ("Client 1", initials, a code). The real value is never written to the substrate/, in the memory.md or in any other file in the folder.
⚠️ The sensitivity note that contradicts itself
The mistake to avoid: writing "I removed the client names" and listing the names right below. That contradiction is itself a leak. If a handle would make the substrate useless to you, the coach asks first to record the actual amount—it doesn’t decide that on its own.
💡 Stable handle = usefulness without exposure
"Stable" means that "Couple A" is always the same couple in every file. That way, the OS remains useful (it can cross-reference information) without ever storing who "Couple A" is in the folder. You keep the map in your head; the disk stays clean.
Why learn
Because the substrate is where golden rule #8 does the most work. You’re about to pour in real knowledge—statements, messages, spreadsheets. Knowing that the coach automatically de-identifies it lets you bring in the material without fear, and teaches you to flag what’s sensitive before pasting it.
Key concepts
🌾 Optional harvest
You don’t need to have everything on hand to move forward. If the material already exists, the coach offers collect and distill on the spot (o harvest). If not, it notes it in memory.md what still needs to be gathered and followed up on — the gap is recorded, not forgotten.
Point to the sources
You say where the knowledge lives; the coach records it in sources.md.
Gather and distill (if there’s material)
The coach reads what you have, de-identifies sensitive information, and writes the compendium.md.
Record what’s missing
What isn’t ready becomes a note in memory.md under “what else to generate”.
💡 Move forward without getting stuck
The great thing about optional harvesting is that you never get stuck waiting until you “have everything.” The OS can be in progress with a clear list of what's missing, and you can fill it in later — the memory.md remembers for you.
Why learn
Because it removes the most common excuse for not starting: “I haven’t organized my data yet.” The coach works with what exists now and records the rest. You leave the module with a real substrate, even if it’s partial, and an explicit plan to complete it.
Key concepts
🧪 Multi-part done check
The Substrate only becomes solid when the compendium.md responds all parts of your hard question with the data in hand. If the question has two parts—for example, “what’s due in 10 days E what's already overdue" — and one of them is missing a data point, the layer is in progress, not solid.
✓ solid
- ✓Answers every part of the difficult question.
- ✓It’s organized, not a pile of raw material.
- ✓The required data is already in the folder.
✗ in progress
- ✗Answers only the easy half of the question.
- ✗An input is missing for the other party.
- ✗It’s still more of a stack than a compendium.
🔎 Honesty that’s worth its weight in gold
When something is in progress, the coach says Which part can it still not answer, and what data would resolve that. For example: "I can tell you what wins, but I’ll only know what’s overdue once you mark what’s already been delivered." This turns it into a clear action, not a vague note.
Why learn
Because the multi-part done-check keeps a “half-baked” substrate from passing as ready. It gives you an honest standard: the layer is solid only when it answers the entire question. And it hands you the next input to look for.
Key concepts
⚠️ Anti-pattern: dumping everything raw
The most common Substrate mistake is the temptation to throw everything in raw in compendium.md and call it done. The coach’s rule is the opposite: distill. A small, clean compendium beats a giant, messy one every time.
⚠️ The kitchen full of spoiled food
The layer analogy: you don't cook a good meal in a kitchen full of spoiled food and unlabeled containers. Dumping raw material fills the kitchen with junk. Distilling means clearing the counter and labeling the containers—that's the real work of this layer.
✓ Distill
- ✓Extract only the essentials from each source.
- ✓Organize by topic, so the OS can read it easily.
- ✓Small and clean > huge and confusing.
✗ Dumping
- ✗Pastes entire PDFs and emails without filtering.
- ✗Cluttering the context window with noise.
- ✗Thinking that "more is better"—it isn’t.
Why learn
Because a dirty substrate contaminates everything that follows: skills read noise, and agents make decisions based on junk. Distilling well now is what makes the capability layers (Module 4.5) work. Keep the raw material as a backup, but inject only the distilled version.
Key concepts
▶️ Copy-run: run the next
Return to the same folder as your OS (the one you created in module 4.2) and run the next. The coach will open the Substrate, ask 2-3 questions about your sources and what’s sensitive, and then build the folder substrate/.
▶ Copy-run · build the Substrate
Objective: gather and distill the knowledge of your OS. Open Claude Code in the same folder where you ran the start and type:
/os-coach next
What will happen: it reads the memory.md, sees the finished Identity and opens the Substrate. It will ask where your knowledge lives, what the tough question is e what is sensitive. Answer (it waits), and it creates substrate/sources.md e substrate/compendium.md.
How to verify: open the compendium.md and test your tough question against it. Does it answer everything? The layer is solid. Missing a part? It flags it in progress and says what data to look up — and it’s fine to continue this way for Module 4.4.
Also confirm that no sensitive value appears in plain text in substrate/ or in the memory.md — handles only.
Why learn
Because it’s your second layer built—and the most important one. Running the same next again cements the rhythm: one command, one layer, always persisted. With Identity and Substrate in place, the foundation of your OS is standing.
Key concepts
✅ Module summary
Next module:
4.4 — next: the Rules, the OS guardrails 🚧