cmds-onboarding

name	cmds-onboarding
description	Interview-based vault context capture for cmds-vault. Personalizes BRAIN.md, fills 100 Themes stubs, and optionally fills the CMDS.md Vault Operator profile so AI assistants give context-aware answers from day one. Activate when the user says "온보딩해줘", "처음 시작할게", "cmds onboarding", "온보딩", "context onboarding", "내 컨텍스트로 채워줘", "fill my vault", "personalize my brain", "interview me", "어떻게 시작하지", or after a fresh clone before exploring slash commands. Complementary to gobi-onboarding (which focuses on Gobi Space publishing).
metadata	{"version":"1.0","author":"johnfkoo951","created":"2026-05-01T00:00:00.000Z","base":"https://github.com/johnfkoo951/cmds-system-files","cohort":"CMDS x GOBI Cohort 1기 (2026-05) — Session 1 4부 ⭐"}

CMDS Onboarding — Vault Context Capture

Interview-led skill that turns the empty cmds-vault into a vault that knows you. Total time: 15–20 minutes. Use after WELCOME's batch-replace ritual, or run end-to-end and have this skill drive both.

When to Use

Activate on any of:

• User says "온보딩해줘", "처음 시작할게", "cmds onboarding", "온보딩", "context onboarding", "내 컨텍스트로 채워줘", "fill my vault", "personalize", "interview me"
• User opens a freshly-cloned cmds-vault and says "어떻게 시작하지?" or "where do I start?"
• BRAIN.md still contains (your name) or (student / researcher / hobbyist) placeholders
• CMDS.md line 145 still reads ## The Vault Operator: (Your Name) with the TODO HTML comment intact
• 30. Permanent Notes/ is empty except for .gitkeep

This skill is text-or-voice agnostic. If the user is in voice mode (Gobi Desktop voice mode active), keep responses short and conversational, ask one question at a time. Otherwise, batch 2–3 questions per turn and use markdown.

Onboarding Philosophy

1. Context-first — fill the vault with who the user is before exploring tools. Without context, every /command gives a generic answer.
2. Answer-1-of-many — the agent asks 5–10 questions per turn but the user only has to answer 1–2. Context accumulates across turns.
3. Agent-does-the-typing — the user describes themselves in natural language; the agent writes the structured frontmatter, BRAIN.md sections, and stub notes.
4. Cohort-friendly — designed to run live during 1주차 4부 (15-min budget) and continue at home. Resume logic lets the user pick up where they left off.
5. Complementary to gobi-onboarding — this skill fills vault context; gobi-onboarding publishes that context to Gobi Space. Run cmds-onboarding first, gobi-onboarding later (or skip Gobi entirely).

Pre-boarding (verify before starting)

Check	Command	Expected	Fix if fails
Vault root reachable	ls CMDS.md WELCOME.md	both files exist	Run from inside cloned cmds-vault/ folder
Claude Code installed	claude --version	version string	npm install -g @anthropic-ai/claude-code
BRAIN.md placeholder check	grep -c '(your name)' BRAIN.md	1 (placeholder still there)	If 0: BRAIN already personalized — confirm user wants to overwrite or skip to Step 5
[[Me]] placeholder in user-scope	grep -rln '\[\[Me\]\]' --include='*.md' BRAIN.md "00. Inbox" "30. Permanent Notes" "20. Literature Notes" "60. Collections" "90. Settings/91. Templates" 2>/dev/null	non-zero or empty (both OK)	This is the user-scope only. System files retain [[구요한]] and are NOT touched.
30. Permanent Notes/ empty	ls "30. Permanent Notes/" | grep -v gitkeep	empty output	If has notes: confirm with user before adding stubs
Vault template version	cat VERSION (or grep template-version README.md)	semver string (e.g. 1.0.0)	If missing: pre-versioning starter — proceed but advise user to upgrade

If any check is unrecoverable, report state to user and ask whether to skip that step or fix the prerequisite first.

Flow

Step 0 — Greet + scope check (30 sec)

Agent says: "Hi — I'll help you turn this empty cmds-vault into one that knows your domain. About 15 minutes. I'll ask a handful of questions; you only need to answer the ones that feel relevant. The rest I'll leave blank and we'll come back later. Ready?"

Then check if the user has already done the WELCOME author batch-replace in user scope:

# Check user-scope only (system files always retain [[구요한]] and are excluded)
grep -rln '\[\[Me\]\]' --include='*.md' \
  BRAIN.md BRAIN_PROMPT.md \
  "00. Inbox" "30. Permanent Notes" "20. Literature Notes" "60. Collections" \
  "90. Settings/91. Templates" 2>/dev/null | head -3

• If output is empty → user has either already personalized or has no user-scope notes yet → skip Step 1·2, go to Step 3 (still confirm operator wikilink choice for naming new stubs in Step 5)
• If output has files → continue to Step 1

Step 1 — Pick operator wikilink (1 min)

Ask: "What name should your author field use? Real name, handle, Korean — your call. You can change later."

Examples:

• Real name: [[John Doe]]
• Handle: [[johndoe]]
• Korean: [[홍길동]]

Capture the choice. Confirm before proceeding.

Step 2 — User-scope author batch-replace ritual (1 min)

⚠️ Scope policy (Jin's feedback, 2026-05-02):

• System files retain author: [[구요한]] — upstream attribution. Never replace.
• User-scope only gets [[Me]] → [[<NAME>]]: BRAIN.md, BRAIN_PROMPT.md, 00. Inbox/, 30. Permanent Notes/, 20. Literature Notes/, 60. Collections/, 90. Settings/91. Templates/.

Reuse [[WELCOME]] § "1B. Run the user-scope author batch-replace ritual". Either:

Option A (Claude Code in-context):

Batch-replace [[Me]] with [[<NAME>]] in user-scope only — exclude system files.
INCLUDE: BRAIN.md, BRAIN_PROMPT.md, "00. Inbox/", "30. Permanent Notes/", "20. Literature Notes/", "60. Collections/", "90. Settings/91. Templates/".
EXCLUDE (system file authorship — leave [[구요한]]): vault root system .md files (CLAUDE/AGENTS/CMDS/Guide/HQ/WELCOME/README), .claude/, 90. Settings/94. Agent Settings/, 90. Settings/91. Skills/, 90. Settings/92. Prompts/.
Note: Body-text [[Me]] inside EXCLUDED files (frontmatter examples) is intentionally a placeholder for future user notes — leave unchanged.
Use exact-string replace (case-sensitive). Verify zero residual [[Me]] in INCLUDE scope only.

Option B (shell):

# User-scope replacement only — system files untouched
for target in BRAIN.md BRAIN_PROMPT.md "00. Inbox" "30. Permanent Notes" "20. Literature Notes" "60. Collections" "90. Settings/91. Templates"; do
  if [ -e "$target" ]; then
    find "$target" -name '*.md' -type f -exec sed -i '' 's/\[\[Me\]\]/[[<NAME>]]/g' {} + 2>/dev/null
  fi
done

# Verify zero residue in user scope (do NOT verify system files — they intentionally retain [[Me]] in body examples + [[구요한]] in frontmatter)
grep -rn '\[\[Me\]\]' --include='*.md' BRAIN.md BRAIN_PROMPT.md "00. Inbox" "30. Permanent Notes" "20. Literature Notes" "60. Collections" "90. Settings/91. Templates" 2>/dev/null && echo "REMAINING ABOVE" || echo "USER-SCOPE CLEAN"

Read the verification output to user. If "USER-SCOPE CLEAN", proceed. If BRAIN.md was empty/unchanged (system files exclude scope is correctly skipped), that's expected.

Step 3 — Persona interview (5–8 min) ⭐ Core

Ask in batches. User answers 1–2 per batch — that's enough.

Batch A — Current tools & environment

1. What tools do you currently use to manage knowledge? (Notion / Evernote / Apple Notes / paper / nothing yet)
2. Where do your files mostly live? (cloud / local / mixed)
3. How often do you use AI day-to-day? (daily / weekly / rarely)

Batch B — Domain & expertise 4. What's your professional / personal domain in 1–2 sentences? 5. Where in your domain do you feel AI is weak — what does AI not understand? 6. What 2–3 keywords are most central to your domain?

Batch C — What you want to remember 7. Imagine 6 months from now: what one piece of information do you really not want to lose track of? 8. What types of input do you want to capture most? (articles / meetings / books / conversations / your own ideas / data)

Batch D — 12-week intent (cohort-specific) 9. Over the next 12 weeks, what would you like this vault to help you produce? (book / newsletter / better notes / research / nothing in particular — just to think better) 10. If you had to write a one-line "PKM Mission" — what core question are you trying to answer?

Batch E — Voice & sharing 11. Who else, if anyone, will read these notes? (just me / partner / team / public) 12. What tone should this vault speak in when AI represents you? (formal / casual / Korean / English / mixed)

After each batch, summarize what you learned in 2–3 sentences and ask if the user wants to add or correct anything.

→ See references/interview-question-bank.md for extended question pool by domain (researcher / executive / creator / student variants).

Step 4 — Fill BRAIN.md (2–3 min)

Based on answers, rewrite BRAIN.md (don't append — overwrite the placeholder template):

---
title: "<NAME>'s Second Brain"
description: "<1–2 sentence Brain summary built from interview>"
thumbnail: "[[BRAIN.jpg]]"
prompt: "[[BRAIN_PROMPT.md]]"
tags:
  - profile
  - second-brain
created: <today YYYY-MM-DD>
---

## About Me

- **Name**: <NAME>
- **Role**: <from interview>
- **Active since**: <YYYY>
- **Areas of interest** (7 areas inferred from interview):
  - <area 1>
  - <area 2>
  - <area 3>
  - ...
  - <area 7>

## How I Use This Vault

<2–3 sentences from Batch C answers about capture patterns>

## What I'm Working Toward

<1–2 sentences from Batch D answers — PKM Mission + 12-week intent>

## Pinned

- [[🏛 CMDS Head Quarter]] — vault navigation hub
- [[🏛 CMDS Guide]] — operational standards
- [[WELCOME]] — first-read onboarding

Show the draft to user. Ask: "Anything to change before I write?" Wait for confirmation. Then write the file.

→ ⚠️ Don't change the frontmatter keys (title, description, thumbnail, prompt) — Gobi looks them up by name. Only change values.

Step 5 — Create 100 Themes stubs (3–5 min) ⭐

Based on Batch B (domain keywords) + Batch C (capture types) + Batch D (intent), suggest 5–10 stub notes for the user's themes. Show the list before writing.

For each, propose:

• File name: 30. Permanent Notes/<keyword-slug>.md
• type: note
• CMDS: value: choose from 📚 101 Interests / 102 Topics / 103 Variables / 104 Terminologies based on the keyword's nature
  • Interest = ongoing passion (e.g., "재즈", "리더십")
  • Topic = subject being explored right now (e.g., "조직 문화", "API 설계")
  • Variable = research variable to operationalize (e.g., "팀 응집력", "일일 활동량")
  • Terminology = term to define (e.g., "SaaS", "OKR")
• 1–2 sentence body — why this theme + what user wants to find here later

Template per stub:

---
type: note
aliases: []
description: "<English 1–2 sentence — what this theme is about and when to add notes here>"
author:
  - "[[<NAME>]]"
date created: <today>
date modified: <today>
tags:
  - theme
  - <kebab-case-keyword>
CMDS: "[[📚 10X <type>]]"
index: "[[🏷 Research Notes]]"
status: inProgress
---

# <Keyword>

## Why this theme matters to me

<1–2 sentences from interview>

## What I want to capture here

<bullet list of triggers — when to add to this note>

## Connections (to fill as you go)

- (links to other notes will accumulate here)

Write all stubs in one batch. Report file count + list.

Step 6 — (Optional) Fill CMDS.md operator profile (3 min)

Ask: "Do you want to fill in the operator profile in CMDS.md now, or leave it for later? Skipping is fine."

If yes, locate ## The Vault Operator: (Your Name) (around line 145), remove the <!-- TODO ... --> HTML comment, and replace with:

## The Vault Operator: <NAME>

### Professional Identity

- **Role**: <from interview Batch B>
- **Primary work**: <from interview>
- **Background**: <1-line summary from interview>
- **Areas of expertise**: <2–3 from interview>

### Current Focus Areas (1–4개)

1. <focus from Batch D>
2. <focus from Batch B>
3. (...)

### Primary Activities

- <activity 1 from interview>
- <activity 2>

### PKM Mission

> <one-line mission from Batch D>

Confirm with user before writing.

Step 7 — Wrap-up + cohort handoff (1 min)

Summarize what's done:

• ✅ [[Me]] → [[<NAME>]] replaced in user-scope only (system files retain [[구요한]] upstream attribution)
• ✅ BRAIN.md filled with 7 areas of interest + How-I-Use + Pinned
• ✅ 30. Permanent Notes/ has <N> theme stubs ready to grow
• ✅ (optional) CMDS.md operator profile filled

Suggest immediate next steps:

1. Try /status — see your vault snapshot
2. Try /connect "<one thought you have right now>" — your first capture
3. (Cohort 1주차 학생) Read [[Claude Code 온보딩 인터뷰 가이드]] in the cohort guide pack for the post-class checklist
4. (Optional, later) Run gobi-onboarding skill to publish your Brain to Gobi Space

Cohort 1주차 학생용 메시지: "Session 2 (5/9) 입력은 지금 채운 BRAIN.md + 100 Themes stubs. 이 상태 유지해주세요. 디테일 보강은 한 주 동안 천천히."

Resume Logic

If the user says "continue onboarding" / "온보딩 이어서" / "where was I", inspect state and resume:

# Step 1·2 — author wikilink decided & batch-replaced in user scope?
# IMPORTANT: scope is user-scope only — system files (CLAUDE/AGENTS/CMDS/Guide/HQ/WELCOME/README, .claude/, 90. Settings/{91. Skills,92. Prompts,94. Agent Settings}/) intentionally retain [[Me]] in body examples + [[구요한]] in frontmatter
grep -rln '\[\[Me\]\]' --include='*.md' \
  BRAIN.md BRAIN_PROMPT.md \
  "00. Inbox" "30. Permanent Notes" "20. Literature Notes" "60. Collections" \
  "90. Settings/91. Templates" 2>/dev/null > /tmp/me_residue
[ -s /tmp/me_residue ] && echo "Step 2 NOT done — [[Me]] still in user scope" || echo "Step 1·2 done (user scope clean)"

# Step 4 — BRAIN.md personalized?
grep -q '(your name)' BRAIN.md && echo "Step 4 NOT done — placeholder still in BRAIN.md" || echo "Step 4 done"

# Step 5 — 100 Themes stubs created?
THEME_COUNT=$(find "30. Permanent Notes" -name '*.md' -not -name '.gitkeep' | wc -l | tr -d ' ')
[ "$THEME_COUNT" -eq 0 ] && echo "Step 5 NOT done — no theme stubs" || echo "Step 5 done — $THEME_COUNT theme stubs"

# Step 6 — CMDS.md operator profile filled?
grep -q '(Your Name)' CMDS.md && echo "Step 6 NOT done — operator profile still placeholder" || echo "Step 6 done"

Resume from the first failing check. Report state to user before resuming.

Voice Mode Adjustments

When gobiDesktop_voice_mode == active:

• One question at a time (not batches)
• Skip the "Show the draft" step in Step 4·5·6 — just write and verbally confirm afterward ("BRAIN.md is now filled with your 7 areas. Want me to read it back?")
• Skip CMDS.md operator profile (Step 6) — it's text-heavy
• Wrap-up in 2 sentences max

Failure Modes

Symptom	Cause	Fix
"I have nothing to say to questions"	User feels imposter / blank	Switch to passive mode — ask user to read 3 random recent web articles or notes; agent infers themes from those. Then summarize back.
User answers everything in one paragraph	Voice user / quick typist	Parse the paragraph: tag each sentence to relevant batch (A–E). Confirm mapping.
Agent mis-classifies theme keyword (Interest vs Topic vs ...)	Inherent ambiguity	Default to 📚 102 Topics (most permissive). User can re-classify with /connect later.
Stub note overlaps existing one in 30. Permanent Notes/	User had pre-existing notes	Skip duplicate, log warning.
User wants to only personalize BRAIN.md, skip 100 Themes	Quick path	Only do Steps 0–4 + 7. Acceptable.
User wants to do whole thing for Gobi Space publishing	Confused with gobi-onboarding	Run cmds-onboarding to fill context, then immediately invoke gobi-onboarding for publish.

What This Skill Does NOT Do

• Run Gobi CLI commands (gobi init, gobi sync, etc.) → use gobi-onboarding
• Capture audio / transcribe → out of scope, use a separate STT pipeline
• Suggest specific note templates beyond stubs → use /connect after onboarding
• Build custom Brain homepage → use Create Brain Homepage (CBH) prompt

References

• [[WELCOME]] — first-read doc; Steps 1·2 here mirror its "author batch-replace ritual"
• [[BRAIN.md]] · [[BRAIN_PROMPT.md]] — output targets for Step 4
• [[CMDS]] — output target for Step 6 (Vault Operator section)
• [[🏛 CMDS Guide]] — frontmatter standard for stub notes
• 90. Settings/91. Skills/gobi-onboarding/SKILL.md — sibling skill for Gobi Space publishing
• 90. Settings/91. Skills/gobi-cli/SKILL.md — full Gobi CLI reference
• references/interview-question-bank.md — extended question pool by domain
• Cohort 1주차 학생용: 70. Outputs/73. Courses/CMDS-Gobi-Cohort-1/Guides/Claude Code 온보딩 인터뷰 가이드.md (cmds-vault 외부, mothership 위치)