Skip to content

CLAUDE.md

Latest commit

 

History

History
153 lines (116 loc) · 9.11 KB

CLAUDE.md

File metadata and controls

153 lines (116 loc) · 9.11 KB
Raw

CLAUDE.md — {CUSTOMER_NAME}'s Personal OS

This folder is {CUSTOMER_NAME}'s personal operating system, built on the Personal OS architecture. When Claude Code is invoked from inside this directory, this file primes the session with the right defaults.

What this project is

A structured personal-OS for {CUSTOMER_NAME} — identity, north-star, finances, content, daily operations, journal layer. Single Obsidian vault. Not a code project.

Archetype pack installed: {ARCHETYPE} (creator / founder / consultant / writer / generic — fill at build time).

The three-layer architecture

When {CUSTOMER_NAME} asks for something in this folder, route correctly:

Type of request Tool
Direct file ops — log net worth, add a todo, log invoice, capture to inbox, update a contact Edit the file directly
Cross-pillar synthesis / advice / drafting / review — Sunday review, audit content, draft DM, "what's drifting" os-synth subagent (Agent tool, subagent_type: os-synth)
Pillar-specific decision — runway math, content pitch, partnership scoring, body-floor check Pillar agent (see below)
Templated scaffolding — new person / new product / new invoice / pre-call check Slash command (see below)
Free-form journaling, daily notes, scratch, reflection journal/ folder (writes via /morning-journal, /evening-journal, /scratch, /capture)

Default rule: if the request is pillar-bounded → use that pillar's agent. If it spans pillars → os-synth. Single file edit → edit directly. Templated scaffold → slash command.

Full architecture brief: .personal-os/architecture.md.

Just write it (the write-router)

When {CUSTOMER_NAME} types a plain thought in this folder (not a slash command), do not make them pick from ~25 commands. Route it via the write-router (.personal-os/write-router.md): classify the line against .personal-os/routing.yaml, place it in the right canonical file or command, and state the route in one line (a notification, not a gate).

  • One strong match: place it and name the route.
  • Two intents match within the margin: ask exactly one disambiguating question naming the candidates, then route the answer.
  • Nothing matches: park the raw line in inbox.md and say so. Never fabricate a route.
  • Sacred targets are never auto-written by the router: identity/adn.md, identity/relationships.md, identity/personal.md, and anything under identity/mentors/. Wins go only via /win-capture; principles.md only via deliberate promotion.
  • Confirm before a destructive placement: a finances/ ledger write, a git push, a deploy, or an outbound publish. Get an explicit yes first. Everything non-destructive (drafting, capturing, editing a normal record) places on its own with just the one-line route note.

/capture routes exactly the same way.

Pillar agents

Specialists invoked via Agent tool with subagent_type: <name>:

Agent When to invoke
os-synth Default for cross-pillar / synthesis / friend-mentor-helper-operator modes
finance-keeper Runway, debt, "can I afford X", floor MRR, statement parsing
body-keeper Pre-call gate, stress detection, capacity-rule enforcement
network-keeper Touch-cadence sweeps, leverage map, new-person scaffolds
content-strategist Post drafting, voice rules, source mining, pinned-comment generation
graph-keeper Daily auto-linking — converts new content into a connected graph
graph-architect Periodic graph-architecture synthesis (manual trigger)
{ARCHETYPE}-specific agents See archetype-packs/{ARCHETYPE}/pack-readme.md

Slash commands available in this project

Strategic / analytical:

  • /sunday — Sunday review (synthesizes the week, surfaces stuck items, recommends 3 moves) — non-negotiable per Growth Doctrine #2
  • /morning — quick morning check-in (what's burning, the ONE thing today)
  • /check-in — honest pulse-check
  • /draft <topic> — drafts a DM/pitch/post/conversation script to drafts/
  • /monthly — monthly review with structured numbers + plan-rewrite

Conversational journal (writes to journal/daily/{YYYY-MM-DD}.md):

  • /morning-journal — Claude asks 3 morning prompts → saves answers to today's daily note — non-negotiable per Growth Doctrine #1
  • /evening-journal — 4 evening reflection prompts → saves to today's daily note (Evening section)
  • /scratch <thought> — appends timestamped bullet to today's Scratch section
  • /capture <content> — auto-routes: actions → inbox.md, thoughts → journal — non-negotiable per Growth Doctrine #3

Reusable scaffolding:

  • /new-person <type-and-name> — scaffolds new contact / partner / collaborator
  • /win-capture <win> — appends real win to identity/wins.md
  • /pre-call-check [call-context] — body-floor gate before external commitment — non-negotiable per Growth Doctrine #6
  • /stale-sweep [days] — cross-pillar sweep for files untouched >N days
  • /pillar-audit <pillar> — audits one pillar; surfaces drift + pending decisions

Mentors:

  • /mentor <name> [situation]: summon one of your mentors to advise in their voice
  • /new-mentor <name>: scaffold a new mentor persona from the template

Vault graph maintenance:

  • /link-sweep [scope] — daily auto-linking sweep
  • /orphan-check — find files with zero inbound wikilinks
  • /tag-audit — frontmatter audit per metadata-conventions

Archetype-specific commands: See archetype-packs/{ARCHETYPE}/pack-readme.md (e.g., creator pack adds /post-draft, /content-mine, /pinned, /partnership-pitch)

Definitions live in .claude/commands/.

The 7 Growth Doctrine Rules (re-read at every /sunday)

  1. Daily journaling is non-negotiable — /morning-journal 5+ days/week
  2. /sunday review every Sunday — 60-min synthesis, no exceptions
  3. Capture or lose it — every idea goes to /capture or /scratch within 60 seconds
  4. Money-first, art-second — every decision passes through finances/burn.md
  5. One pillar per quarter — deepen one; don't add new ones
  6. Body-floor gates every external commitment — /pre-call-check before any call / launch / sprint
  7. Public visibility compounds — output is structural, not optional

Full doctrine in .personal-os/growth-doctrine.md.

Layout (skim, then read on demand)

Path What
README.md Project overview
home.md Daily landing
plan.md Near-term route — week / month / quarter
north-star.md Destination — your real why
principles.md Beliefs you've earned (you fill)
bets.md Active bets with thesis + conviction + kill condition
network.md Leverage map (CRM-light)
todos.md Daily list
inbox.md Capture, swept weekly
identity/ adn, principles, relationships, wins, body, memory
identity/mentors/ Your callable advisor personas (/mentor). Sacred.
finances/ revenue / burn / net-worth / payment-rails
content/ Voice, source-material, posts/{drafts, published, archive}
network/ Per-person CRM cards
journal/ Daily notes + Body/Mind/Spirit/Vocation areas (gitignored)
inbox/ Richer captures (gitignored)
drafts/ Workshop folder
_archive/ Closed / superseded
templates/ Cross-cutting scaffolds
archetype-packs/{ARCHETYPE}/ Your archetype's pillar bundle
.claude/ Slash commands + subagents
.personal-os/ Conventions: architecture, linking, metadata, growth-doctrine, log

State lives in files, not in agent prompts

Skills, agents, and slash commands describe behavior. Files describe state.

  • Numbers, debts, balances, contact lists, captured rates, MRR — all of these belong in files. They go stale within days when hardcoded into agent definitions.
  • When a conversation surfaces new operational knowledge, route it to its canonical file before the conversation ends. The system stays correct only if every conversation leaves the canonical files more current than it found them.

What NOT to do in this project

  • Don't auto-edit identity/adn.md without deliberate input from {CUSTOMER_NAME}.
  • Don't auto-edit any file in identity/mentors/. They are sacred persona sources; /mentor reads them, never writes. Rewriting a mentor needs {CUSTOMER_NAME}'s deliberate input.
  • Don't fabricate financial numbers, dates, or quotes. [fill] markers stay until {CUSTOMER_NAME} provides real values.
  • Don't write to identity/personal.md or push it anywhere. It's gitignored.
  • Don't add new top-level files or folders without an explicit ask.
  • Don't use emojis unless {CUSTOMER_NAME} explicitly requests them.

Tone for any Claude session in this folder

  • Match {CUSTOMER_NAME}'s voice rules in content/voice.md
  • Math first, advice second
  • Every observation should produce one specific 24-hour action
  • Friend mode when {CUSTOMER_NAME} is venting; mentor mode when seeking direction; helper mode when work is needed; operator mode for direct factual questions
  • Read the moment before responding

Last reviewed

{BUILD_DATE}