English | 한국어 | 中文 | 日本語 | Español

vibe-sunsang (바선생)

An AI mentor agent that turns your Claude Code conversation history into a growth engine.

vibe-sunsang automatically collects, converts, and analyzes every conversation you have with Claude Code — so you can become a more effective AI collaborator over time.

Quick Start • Why vibe-sunsang? • How it works • Modes • Level system • Requirements

---

Quick Start

1. Install

/plugin marketplace add https://github.com/fivetaku/gptaku_plugins.git
/plugin install vibe-sunsang

Then restart Claude Code.

2. Run onboarding

In any project:

/vibe-sunsang 시작

Onboarding will:

1. Create ~/vibe-sunsang/ workspace (with CLAUDE.md, .gitignore, and git init)
2. Scan your Claude Code conversation history at ~/.claude/projects/
3. Let you assign readable names to each discovered project
4. Classify each project by workspace type (Builder / Explorer / Designer / Operator)
5. Convert all JSONL logs to Markdown for the first time
6. Ask whether to jump straight into mentoring or a growth report

3. Use it weekly

/vibe-sunsang 변환      ← convert this week's conversations
/vibe-sunsang 멘토링    ← get coaching on your AI usage patterns
/vibe-sunsang 성장      ← generate a growth report
/vibe-sunsang 지식      ← explore the level system and concepts

---

Why vibe-sunsang?

• Your conversations already hold the data — Claude Code stores every session as JSONL. vibe-sunsang reads those files so nothing new needs to be logged
• Not just "what did I build" — "how did I build it" — analysis focuses on your request quality, error recovery habits, and strategic thinking, not just output volume
• Workspace-type-aware — the level system adapts to what you are actually doing (coding, researching, planning, or automating)
• Six dimensions, not one score — DECOMP, VERIFY, ORCH, FAIL, CTX, META are measured independently so you know exactly which skill to develop next
• Longitudinal tracking — each growth report updates TIMELINE.md so you can see progress across months, not just a single session
• Non-developer friendly — plain-language coaching with analogies, no jargon wall

---

How it works

Claude Code conversation logs (~/.claude/projects/)
                    |
                    | /vibe-sunsang 변환
                    v
     Readable Markdown files (~/vibe-sunsang/conversations/)
                    |
                    | 6-axis analysis
                    | (DECOMP / VERIFY / ORCH / FAIL / CTX / META)
                    v
     Mentoring session  ←→  Growth report  ←→  Concept learning
                    |
                    | Fit Score  ×  workspace-type weights
                    v
     Level judgment: L1.0 – L7.0 (0.5-step increments)
                    |
                    v
     TIMELINE.md (longitudinal growth tracking)

Conversation data and plugin code live in separate directories, so plugin updates never touch your data.

---

Modes

Onboard (시작)

Sets up the ~/vibe-sunsang/ workspace, maps project names, classifies workspace types, and runs the first full conversion.

---

Convert (변환)

Converts Claude Code JSONL logs to readable Markdown files organized by project.

Command	Behavior
/vibe-sunsang 변환	Incremental — only new sessions
/vibe-sunsang 변환 force	Full re-convert
/vibe-sunsang 변환 <project name>	Single project only

Each converted file includes metadata frontmatter: date, token counts, model used, and tools called. An INDEX.md gives you the full overview at a glance.

---

Mentor (멘토링)

Four coaching modes, each focused on different skill dimensions:

Mode	Trigger	What it does	Core axes
A — Request quality	"요청 코칭해줘"	Grades your prompts A–D for clarity and context	DECOMP + CTX
B — Antipattern diagnosis	"뭘 잘못하고 있는지"	Detects bad habits with specific examples and fix strategies	FAIL + VERIFY
C — Concept learning	"이게 뭐야"	Explains concepts from your recent sessions in plain language	META
D — Full coaching (default)	"멘토링해줘"	Complete 6-axis checkup, level judgment, and action plan	all 6 axes

Mode D outputs a text radar chart and a three-step action plan: one thing to do right now, one this week, one within a month.

Coaching results can be saved to ~/vibe-sunsang/exports/mentor-YYYY-MM-DD.md.

---

Growth report (성장)

Delegates analysis to a dedicated growth-analyst subagent to keep the main context lean. The agent reads your session files, scores all six dimensions, and writes a full report.

Time range options:

Range	Coverage
This week	Last 7 days
Last 2 weeks (recommended)	Last 14 days
This month	Last 30 days
Specific project	Full project history

Report includes:

1. Basic stats — session count, message count, token usage
2. 6-axis technical dimension analysis with scores and weights
3. Text radar chart
4. Gate condition status table (requirements to reach the next level)
5. Antipattern detection (calibrated to your workspace type)
6. Longitudinal comparison against all previous reports
7. Level card with progression bar
8. Level-up message if you advanced (L4→L5 gets a special "80% wall" event)
9. TIMELINE.md auto-update

---

Knowledge (지식)

Answers questions about the vibe-sunsang framework itself:

Topic	Description
Level system	7-level progression (0.5 increments) and what each level means
Six axes	Definitions and examples for DECOMP, VERIFY, ORCH, FAIL, CTX, META
Workspace types	What Builder / Explorer / Designer / Operator means for your analysis
Antipatterns	Bad habits to watch for, calibrated to your workspace type
Request quality	How to write better prompts
Growth metrics	Fit Score formula, gate conditions, type-specific weights

---

Level system

6 technical dimensions

Code	Dimension	One-line definition
DECOMP	Task decomposition	Breaking complex requests into units AI can handle
VERIFY	Verification strategy	Critically reviewing AI output rather than accepting it
ORCH	Orchestration	Combining tools, agents, and workflows
FAIL	Failure recovery	Diagnosing errors and finding alternatives
CTX	Context management	Supplying background, constraints, and goals to AI
META	Metacognition	Recognizing and strategically adjusting your AI usage patterns

Workspace types and weights

Analysis weights differ by what you are actually doing:

Dimension	Builder	Explorer	Designer	Operator
DECOMP	25%	15%	20%	15%
VERIFY	25%	15%	15%	20%
ORCH	15%	10%	10%	25%
FAIL	15%	20%	10%	20%
CTX	10%	20%	25%	10%
META	10%	20%	20%	10%

7-level progression (per workspace type)

Level	Builder	Explorer	Designer	Operator
L1	Observer	Asker	Dreamer	User
L2	Tinkerer	Curious	Sketcher	Repeater
L3	Collaborator	Digger	Shaper	Optimizer
L4	Pilot	Investigator	Planner	Builder
L5	Architect	Analyst	Strategist	Engineer
L6	Conductor	Synthesizer	Director	Orchestrator
L7	Forgemaster	Scholar	Visionary	Automator

Levels are scored to two decimal places internally, then rounded to 0.5 increments for display. The criterion is not "can you do this" but "do you do this consistently."

Gate conditions

Gate	Requirement
L3	Context specificity ratio > 0.5
L4	Verification ratio > 0.15 AND correction ratio > 0.05
L5	Tool diversity > 8 (or orchestration tool present) AND strategic ratio > 0.05
L6	Multi-agent experience present
L7	L6 gate + evidence of external contribution

---

Suggested routine

Every Friday (15 min)

1. /vibe-sunsang 변환      ← convert this week's sessions
2. 멘토링해줘               ← weekly review
3. Pick one thing to try next week

Monthly (30 min)

1. /vibe-sunsang 변환 force  ← full re-convert
2. 성장 리포트 만들어줘        ← monthly growth report
3. Compare with previous reports — trends are shown automatically

---

File structure

All user data lives in ~/vibe-sunsang/. Plugin code lives separately.

~/vibe-sunsang/
├── CLAUDE.md                 ← workspace description (created by onboarding)
├── .gitignore
├── config/                   ← project name map, workspace type classifications
├── conversations/            ← converted Markdown sessions
│   └── INDEX.md              ← full project index
├── exports/                  ← growth reports, mentor session saves
└── growth-log/
    ├── TIMELINE.md           ← level and 6-axis history across all reports
    └── weekly/               ← weekly retrospective notes

---

Requirements

• Claude Code CLI
• Python 3.8+

Conversation source path

By default vibe-sunsang reads sessions from ~/.claude/projects/. The resolution order:

1. --projects-dir <path> flag (passed to convert_sessions.py)
2. $CLAUDE_CONFIG_DIR/projects if CLAUDE_CONFIG_DIR env var is set
3. ~/.claude/projects/ (default)
4. WSL fallback: /mnt/c/Users/$USER/.claude/projects (only if 1–3 don't resolve)

Use --projects-dir when running across environments (e.g., WSL ↔ Windows) or when your Claude config lives outside ~/.claude/:

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/convert_sessions.py \
  --projects-dir /custom/path/.claude/projects \
  --output-dir "$HOME/vibe-sunsang/conversations"

---

License

MIT

---