Baksunseon Onboarding
Step 0: Prepare User Data Directory
"$HOME/vibe-sunsang/" Check whether the directory exists.
If it already exists (re-onboarding):
"There is previously configured Baksunseon data. Would you like to keep the existing settings and add only a new project, or set up again from scratch?"
EXECUTE: Immediately call the AskUserQuestion tool with the JSON below:
{
"questions": [{
"question": "이전에 설정한 바선생 데이터가 있습니다. 어떻게 할까요?",
"header": "재설정",
"options": [
{"label": "새 프로젝트만 추가", "description": "기존 설정을 유지하면서 새 프로젝트만 추가해요"},
{"label": "처음부터 다시", "description": "기존 설정을 백업하고 새로 시작해요"}
],
"multiSelect": false
}]
}
Depending on the choice:
- "Add only a new project" → Read the existing config file, skip mapped projects, and proceed only with new projects
- "Start over from scratch" → Back up the existing config file (
*.bak) and then create it anew
If it does not exist:
mkdir -p "$HOME/vibe-sunsang/config" "$HOME/vibe-sunsang/conversations" "$HOME/vibe-sunsang/exports" "$HOME/vibe-sunsang/growth-log/weekly"
Step 0.5: Configure Workspace Environment
Migration detection: $HOME/vibe-sunsang/ If a numbered-prefix directory (10-, 20-, 30-, 40-, 90- etc.) exists inside, determine it to be a pre-v1.3.x structure and inform the user:
"A previous version's folder structure has been detected. The plugin has been updated and no longer requires local command/skill files in the workspace. Existing data is preserved as is, and it will automatically switch to the new structure."
If migration is needed:
40-conversations/→conversations/(move if the existing conversations/ is empty, otherwise merge)90-exports/→exports/(same)30-growth-log/→growth-log/(same).claude/,10-scripts/,20-knowledge-base/,00-system/→ Deletion guidance (after user confirmation)
Create CLAUDE.md:
$HOME/vibe-sunsang/CLAUDE.mdIf it does not exist:
${CLAUDE_PLUGIN_ROOT}/references/CLAUDE-MD-TEMPLATE.mdRead the contents of$HOME/vibe-sunsang/CLAUDE.mdSave to with Write
If it already exists:
"Keep the existing CLAUDE.md."
Create .gitignore:
$HOME/vibe-sunsang/.gitignoreIf it does not exist, create it:
# Large conversation files
conversations/**/*.md
!conversations/INDEX.md
If it already exists, skip.
git init:
$HOME/vibe-sunsang/.gitIf it does not exist:
cd "$HOME/vibe-sunsang" && git init
If it already exists, skip.
Gotchas
- Do not hardcode CLAUDE-MD-TEMPLATE.md inline. Always
${CLAUDE_PLUGIN_ROOT}/references/CLAUDE-MD-TEMPLATE.mdread from. - During migration, be careful not to overwrite directories containing existing data. If there is a conflict, confirm with the user.
- Run git init only in the user workspace. Do not run it in the plugin directory.
Step 1: Welcome & Explanation
Show the following message to the user:
Welcome to Baksunseon!
Baksunseon looks back on your conversations with Claude Code, and how to collaborate better with AIis an AI mentor agent that teaches you.
If you review this week's conversations here once a week:
- Which mistakes I keep repeating
- How to ask AI for things more effectively
- Which concepts I skipped over without understanding
you can find.
Now I will proceed with the initial setup.
Step 2: Check Claude Code sessions
"$HOME/.claude/projects/" Check the directory to retrieve the list of available projects.
ls "$HOME/.claude/projects/"
If there are no projects:
"There is no Claude Code conversation history yet. Please use Claude Code in another project first, then come back." → Exit here
If there are projects, proceed to the next step.
Step 3: Map project names
Show the list of discovered project directories and guide the user to assign readable names.
Guidance message:
I found the projects saved by Claude Code. Please give each project an easy-to-recognize name.
Don't worry if the directory names look complicated - they are the actual project folder paths.
For each project directory:
- Infer the project name from the directory name (extract the last meaningful part of the path)
- EXECUTE: Immediately call the AskUserQuestion tool with the JSON below (repeat for each project):
{
"questions": [{
"question": "이 프로젝트(`-Users-xxx-my-project`)의 이름을 뭐라고 할까요?",
"header": "프로젝트명",
"options": [
{"label": "{추측한 이름}", "description": "디렉토리 경로에서 추측한 이름이에요"},
{"label": "다른 이름", "description": "Other에 원하는 이름을 입력해주세요"},
{"label": "건너뛰기", "description": "이 프로젝트는 분석하지 않아요"}
],
"multiSelect": false
}]
}
The question and the label of the first option are dynamically generated to match each project.
Rules:
- Ask about only up to 5 at a time (too many is tiring)
- If there are more than 5 projects, split them into groups of 5 and repeat. After each group, confirm "Shall we continue?"
- Projects with fewer than 5 sessions are automatically skipped (notify the user)
- Projects for which "Skip" was selected are excluded from mapping
- On re-onboarding: Skip already mapped projects and ask only about new projects
Save the results to "$HOME/vibe-sunsang/config/project_names.json".
Step 4: Classify workspace type
Read each project's CLAUDE.md or README.md to infer the type.
Classification flow:
- From the project path,
CLAUDE.mdorREADME.mdfind and read - Infer the type based on the content
- Show the inference result to the user and get confirmation
Type classification criteria:
| Type | Keywords/Patterns | Description |
|---|---|---|
| Builder (Implementer) | build, test, deploy, component, API, coding, development, app | Coding/development project |
| Explorer (Explorer) | research, study, analyze, research, learning, study group, Q&A, questions | Research/Q&A/Learning |
| Designer (Planner) | plan, design, ideation, planning, ideas, content, writing | Planning/Ideation |
| Operator (Operator) | automate, workflow, schedule, automation, integration, script, MCP | Work automation |
If CLAUDE.md/README cannot be found:
- Briefly check the file structure of the project (
.py,.jsif there are many files, Builder, etc.) - If inference is difficult, ask the user directly
For each project, EXECUTE: immediately call the AskUserQuestion tool with the JSON below:
{
"questions": [{
"question": "[프로젝트명]의 CLAUDE.md를 분석해보니 [유형] 워크스페이스로 보입니다. 맞나요?",
"header": "유형 확인",
"options": [
{"label": "Builder (코딩)", "description": "코딩/개발 프로젝트"},
{"label": "Explorer (리서치/학습)", "description": "리서치/Q&A/스터디"},
{"label": "Designer (기획)", "description": "기획/아이디에이션"},
{"label": "Operator (자동화)", "description": "업무 자동화/데이터처리"}
],
"multiSelect": false
}]
}
The question is dynamically generated with each project's name and inferred type.
Rules:
- If a project has multiple purposes, select 1 main purpose
- If the same type repeats across multiple projects, group them and confirm at once
Save the results to "$HOME/vibe-sunsang/config/workspace_types.json":
{
"schema_version": 1,
"type_definitions": {
"builder": "코딩/개발",
"explorer": "리서치/Q&A/스터디",
"designer": "기획/아이디에이션",
"operator": "자동화/데이터처리"
},
"default_type": "builder",
"workspaces": {
"-Users-xxx-my-app": {
"type": "builder",
"name": "my-app",
"detected_from": "CLAUDE.md",
"confirmed": true
}
}
}
Step 5: Run the first conversion
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/convert_sessions.py --force --names-file "$HOME/vibe-sunsang/config/project_names.json" --output-dir "$HOME/vibe-sunsang/conversations" 2>/dev/null || python ${CLAUDE_PLUGIN_ROOT}/scripts/convert_sessions.py --force --names-file "$HOME/vibe-sunsang/config/project_names.json" --output-dir "$HOME/vibe-sunsang/conversations"
Show conversion progress, and when complete, summarize the results:
- Total number of projects
- Total number of sessions
- TOP 3 most active projects
- Distribution by type (Builder N개, Explorer N개, ...)
Step 6: Usage guide
Setup complete!
You can receive customized analysis for each project type:
| Type | Analysis content |
|---|---|
| Builder (implementer) | Coding request quality, error handling, code comprehension |
| Explorer (explorer) | Question depth, source verification, critical thinking |
| Designer (planner) | Planning specificity, structuring, feasibility |
| Operator (operator) | Automation quality, error handling, reusability |
v2 level system:
Barseonsaeng analyzes AI utilization ability across 6 skill dimensions:
| Skill dimension | Easy explanation |
|---|---|
| DECOMP (task decomposition) | The ability to break a large request into small steps |
| VERIFY (verification strategy) | The ability to check and verify AI results |
| ORCH (orchestration) | The ability to combine and utilize multiple tools |
| FAIL (failure response) | The ability to identify the cause of an error and respond to it |
| CTX (context management) | The ability to convey necessary information to AI well |
| META (metacognition) | The ability to reflect on how I use AI |
Levels range from L1.0 (beginner) to L7.0 (master), measured in fine 0.5 increments. The important axes differ by type, so you can receive customized analysis that fits you.
Available features:
| Command | Description |
|---|---|
/vibe-sunsang 변환 |
Convert new conversations (recommended weekly) |
/vibe-sunsang 멘토링 |
AI utilization ability coaching (customized 6 axes by type) |
/vibe-sunsang 성장 |
Growth analysis report (including 6-axis radar chart) |
Recommended routine:
- Every Friday,
/vibe-sunsang 변환convert this week's conversations with - Review this week with "Mentor me"
- Carry out the action plan
Step 7: Ask whether to start right away
EXECUTE: Immediately call the AskUserQuestion tool with the JSON below:
{
"questions": [{
"question": "바로 이번 주 리뷰를 시작해볼까요?",
"header": "다음 단계",
"options": [
{"label": "멘토링 시작", "description": "AI 활용 능력 코칭 세션을 바로 시작해요 (6축 분석)"},
{"label": "성장 리포트 생성", "description": "성장 분석 리포트를 자동 생성해요 (6축 레이더 차트 포함)"},
{"label": "나중에 할게요", "description": "여기서 마무리할게요"}
],
"multiSelect": false
}]
}
Depending on the choice:
- "Start mentoring" → run the vibe-sunsang-mentor skill
- "Generate growth report" → run the vibe-sunsang-growth skill
- "I'll do it later" → exit