Detailed content
📊 What the audit does
O /os-coach audit is the command you run at any time to pressure-test your OS against reality. It rereads your goal in the memory.md, open the actual folder (not what memory says it contains) and scores the six layers against that specific goal. Then it delivers the three highest-leverage next moves and saves the report in OS-AUDIT.md.
🌱 New here?
Audit here it's just “give each part an honest score.” Scorecard is the report card: a table with each layer’s score and the reason behind it. Leverage (leverage) is "how much one move pushes you toward the goal"—the audit orders the next steps from highest to lowest return.
📊 What it is
An assessment tied to your goal. The question is never “is this a good OS in general?” but “does this layer move me toward MY goal?”
- •Scores the 6 layers across 4 levels.
- •Give the 3 highest-leverage moves, in order.
- •Write OS-AUDIT.md (storable) and feeds back into the loop.
Why learn
Because building is only half the work. Without auditing, you accumulate layers without knowing whether they serve the goal—and the OS quietly decays (Module 1.1: cultivate, don’t install). The audit turns “I made a bunch of files” into “I’m actually closer to never missing a deadline again.”
Key concepts
🏅 The 4-level scale
Each layer gets one of four levels. Be honest: most OSs are almost entirely at the beginning in Missing or Started, and that's normal. The scale shows direction; it doesn't grade you to make you feel bad.
How to read: the goal isn’t “everything Compounding already.” It’s to move up one step at a time in the layer that best advances your goal. A Solid layer that improves on its own (growing substrate, versioned skill, audits that feed back) reaches Compounding.
It doesn't exist, or it's an empty placeholder.
It exists, but it’s thin, generic, or answers only part of what the objective requires (e.g., shows “due soon” but not “already overdue”).
Real, specific, and passes the layer’s complete done-check (answers the whole hard question, not just the easy half).
Solid AND improves on its own over time (the substrate grows, skills are versioned, audits feed back into it).
Why learn
Because the scale gives you the vocabulary to see real progress. “Started” isn’t failure: it signals that a specific piece is missing—and the audit tells you which one. Knowing the difference between Started and Solid keeps you from confusing “there’s a file” with “it solves the problem.”
Key concepts
🎯 The non-generic test
The rule that keeps the audit useful: every line must fail if pasted into a stranger’s audit. If a sentence would work for any OS, it’s generic and banned—rewrite it using your actual files, names, and goal. A generic audit is a failed audit.
✗ Generic (banned)
"Add more documentation to your substrate."
Works for anyone. Moves no one.
✓ Specific (required)
"Your goal is to win back subscribers who canceled, but substrate/ only has sources.md and a compendium.md empty. The highest-leverage move is to pull the last 90 days of cancellation messages into substrate/churn/ and distill the three most common reasons.”
💡 The key insight
Each recommendation cites a actual file name, one real gap e o real objective. If you can swap your OS for someone else’s without changing the sentence, it doesn’t belong.
Why learn
Because it’s exactly what makes the OS Coach different from internet advice. You already have plenty of generic advice; what moves the needle is “your file X is missing Y to achieve Z.” This is also the test you use to tell whether an audit was done well—yours or one from any tool.
Key concepts
🥇 The 3 highest-leverage moves
The audit doesn’t give you an endless list. It gives you three moves, ordered by leverage — not by layer order. The priority follows a clear logic, where the rule above takes precedence when two compete:
How to read: the hard rule only jumps the queue if the damage is imminent or irreversible (wrong amount of money, data leak, tight deadline). Otherwise, the natural order is substrate → skills → agents. Never recommend an agent before the skills it would orchestrate.
1. Missing hard rule is #1 only if the worst case is imminent/irreversible. Otherwise, it stays below the substrate.
2. Otherwise, substrate gaps win. A weak foundation means everything else is built on sand.
3. Skills only matter when the substrate can feed them.
4. Agents last. Never before the skills that would orchestrate them.
✓ Order by leverage
- ✓Targets the objective’s bottleneck.
- ✓Weak substrate before skills.
- ✓A hard rule cuts the line only when it’s urgent.
✗ Sort by layer
- ✗Polish layer 1 “because it’s the first one.”
- ✗Beautiful agent on an empty substrate.
- ✗Spend an hour away from the real bottleneck.
Why learn
Because ordering by leverage is what helps you make progress with little effort. Without this logic, you spend energy polishing the wrong layer (a polished agent on top of an empty substrate). With it, every hour you invest tackles the real bottleneck in your goal.
Key concepts
📷 Guided example · start + Identity
Let’s look at a real, condensed example (from the OS Coach README). A solo wedding photographer want to stop missing deadlines. Clients show up as Couple A, Couple B because it marked the names as sensitive — so the coach keeps handles non-identifiers in place of real names.
You type:
/os-coach start vivo perdendo prazos de cliente e quero nunca mais
perder uma data de entrega
The coach confirms the goal, prepares the folder, and asks three short questions: who it's for, a single question you wish you could ask it, e one thing it should always do + one it should never do. You answer: just for me; "what will win in the next 10 days, and what am I already behind on"; always remind me within 3 days of a contracted date; never message a client without my approval. It then writes the Identity:
CLAUDE.md (generated)
# Quem eu sou Sou o sistema operacional de uma fotógrafa de casamento solo. Existo para que nenhuma entrega de cliente atrase jamais. ... ## Recusas duras (nunca) - Nunca mando mensagem a um cliente sem você ler e aprovar antes. - Nunca deixo um prazo passar em silêncio. Se algo está em risco, eu trago à tona, alto e claro.
Why learn
Because it shows the Identity layer emerging from your own words: the goal becomes the mission, and “always/never” becomes concrete refusals. Notice that the handles (“Couple A”) appear right from the start—protecting sensitive data isn’t a separate step; it’s the coach’s default way of working.
Key concepts
🗂️ Guided example · next + Substrate
The photographer types /os-coach next. This opens the Substrate: the coach asks where the reservations are stored and what’s sensitive, then builds a tracker that calculates both deadlines from each wedding date (preview = wedding + 1 week; full gallery = wedding + 6 weeks) and answers the question right away:
Coach response (immediate)
Dentro de 3 dias, sua linha de "sempre avisar" disparou: - Casal B, galeria completa vence 2026-07-01 (faltam 3 dias) A vencer dentro de 10 dias: - Casal C, prévia vence 2026-07-02 (faltam 4 dias)
🔎 Notice the detail
The tracker answers “what’s due soon”—but not yet answers "what am I already behind on," because she didn't mark what had already been delivered. Remember this: that's exactly what will keep Substrato in Started in the next topic audit.
Why learn
Because it shows a Substrate that already creates real value (answers “due soon”) but is still partial. This nuance—answering half of a difficult question—is at the heart of the difference between Started and Solid. The OS is useful early on, and the audit pinpoints exactly what’s missing to make it complete.
Key concepts
🧮 Guided example · audit + scorecard
The photographer types /os-coach audit. The coach reads the actual files and produces the report. See how each level is tied to her case—without a single generic sentence:
Scorecard
| Layer | Note | Why |
|---|---|---|
| Identity | Solid | O CLAUDE.md states the mission and two refusals. |
| Substrate | Started | The tracker answers “due soon,” but not “already overdue” until you mark what was delivered. |
| Rules | Missing | Customer handles now live in the folder, with nothing preventing a public backup. |
The three moves that matter most
Rules: write rules/never.md (never make a public backup of this folder). A leak can’t be undone, and a gallery wins in 3 days.
Substrate: mark which galleries have already gone out, so that "expires soon" also answers "already overdue."
Skills: turn deadline checking into a repeatable command.
⚠️ Why the rule jumped the queue
Substrate usually comes before Rules. But here, the worst-case scenario is imminent and irreversible: the handles are in the folder, nothing prevents a public backup, and there’s a delivery due in 3 days. By Topic 4’s priority logic, this moves the missing hard rule to move #1.
After the audit, her folder looks like this:
The OS folder (handles, never real names)
photo-os/ ├── CLAUDE.md # Identidade, a alma ├── memory.md # onde você parou ├── substrate/ │ ├── sources.md # onde vivem os dados reais │ └── compendium.md # o tracker de prazos └── OS-AUDIT.md # o scorecard mais recente
Why learn
Because this is the entire audit in a single case: honest notes tied to her files, the hard rule cutting the line because it’s irreversible and imminent, and three concrete moves in order. Every file is real, grounded in her goal, and hers to keep—exactly the result you want from your own audit.
Key concepts
🔁 Feedback in the loop + copy-run
The audit isn’t a dead end: it feeds the loop. After scoring, the coach updates the memory.md — attach the dated headline to the audit_log, stamps the date again and rewrites the layer status block based on the notes. And saves the full report to OS-AUDIT.md for you to save. The three moves become your next next.
The closed loop
Build
Layers with start / next, one at a time.
Audit
Against the objective: scorecard + the three moves.
Edit
Your files (they're yours) and tackle movement #1.
Re-audit
Watch the score go up. The loop doesn’t end: the OS is cultivated, not installed.
🌱 New here?
You can and should manually edit what the coach creates — the files are yours. The recommended flow: edit, then run /os-coach audit again to see how your changes score against the goal. The write-back ensures the next session already knows the latest score.
Objective: run the audit in your own OS and get the scorecard + the three moves tied to your goal.
Paste into Claude Code (inside your OS folder):
/os-coach audit
The coach generates, in your folder:
OS-AUDIT.md # scorecard das 6 camadas + 3 movimentos + o que já funciona
How to verify: open OS-AUDIT.md and apply the non-generic test — each line should mention a your file e o your goal. If a sentence would work for any OS, ask for it to be rewritten. Then focus only on the move #1 and run /os-coach audit again to see that layer’s score go up.
Why learn
Because this is where Track 4 becomes a habit, not a one-time event. Building once is easy; auditing, tackling #1, and auditing again are what keep the OS alive and bring it ever closer to the goal—exactly the “cultivation” mindset from Track 1. With this, you’re ready for Track 5, where you apply everything to real domains.
Key concepts
✅ Module summary
Next track:
5.1 — Tax & Bookkeeping OS 🧾 (real domains & production)