-
Notifications
You must be signed in to change notification settings - Fork 0
First week
The route that matches what is actually in this repository. (docs/ONBOARDING.md describes a different one — see the note at the bottom.)
You do not have to be a programmer. Nothing here needs you to write code; the assistant does the installing and you check the receipts.
- A Claude Pro or Max subscription, then Claude Code:
npm install -g @anthropic-ai/claude-code(no npm → macOSbrew install node, Windows nodejs.org). - Run
claude, sign in with the subscription. - Install Obsidian — you will want to look at your own second brain with your eyes, not only through the assistant.
- Clone this repository (or Code → Download ZIP).
- Open
SEED.md, put your own name and context in it, paste it as the first message of a Claude Code session.
From there the assistant follows BOOTSTRAP-CLAUDE.md itself. Its six steps are: build your ~/.claude/CLAUDE.md from the principles, create the vault, install and adapt the skills, set up the CRM if you have leads or clients, solve one real problem and save the result, then write the reflexes back into CLAUDE.md.
Ask for a receipt after every step. "Saved X → note Y, 14 files copied" is a receipt. "Done" is not.
Installing all 101 on day one gives you a menu you will not read. Adapt these five first, in this order, because they are the ones that keep the rest honest:
| Skill | Why it comes first |
|---|---|
/tt |
proves that what was just built actually runs — the antidote to "done" meaning two-thirds done |
/retro |
closes a session: what was built, what is worth keeping, where it goes |
/intake |
takes a new rule you just invented and files it where it will surface again |
/1 |
recovers the thread after a crash or a lost context |
/ask |
searches your own vault; without a retrieval engine, start with the grep version — it is enough for a long time |
Second wave, when the first five are boring: coach, cofounder, dedup, relink, obsidian-backup, obsidian-ingest, alfa-search-recall-deepresearch. The rest as the need shows up.
Check: type / in Claude Code — your new commands appear in the list.
Pick something that annoyed you this week and let the assistant solve it and save the result to the vault. The point is not the answer; it is watching the loop close — problem → work → a note that is linked and findable next month.
If you have clients, students or leads, this is also where the CRM earns its place: deploy crm-template/ into 04-Projects/<domain>/crm/, one markdown card per person, and fill the checklist against your real process rather than the sample one.
These are what turn a clever session into a system. Have them written into your CLAUDE.md:
- An action repeated two or three times → propose turning it into a skill.
- End of session → memory check: what did I learn that is worth keeping, and where did I put it? — with a receipt.
- Stepped on a rake → propose a rule (the rake, the conclusion, where the rule goes).
- Saw something broken or drifting → raise it and offer the fix, rather than working around it silently.
The skills were written inside a different system. A skill that points at an engine you do not have has three honest options: build a simple local replacement (grep or a stdlib script), degrade to manual mode, or park the skill and say plainly what is missing. A silently broken link is the one option that is not allowed. Details: Adapting it to your machine.
Why this page exists: docs/ONBOARDING.md routes the reader through 05-Protocols-Bible/, 01-Concepts/, crm-lite/, 06-Templates/ and 00-MOCs/, which belong to the vault repository this kit was extracted from and do not exist here. Filed as issue #1. When that lands, this page should shrink to a pointer rather than stay a second source of truth.