Knowledge Management Skill for Claude Code

ระบบจัดการความรู้ 4 layers สำหรับ Claude Code CLI - inspired by Claude-Mem แต่เก็บไว้ใน Git repository

Roadmap: ดู ROADMAP.md สำหรับแผนพัฒนา

Usage Guide: ดู USAGE.md สำหรับคู่มือการใช้งานฉบับสมบูรณ์

Features

🚀 4-Layer System: /mem → /distill → /td → /improve
📝 Before/After Context: จับบริบทก่อน-หลังเหมือน Claude-Mem
🔍 Searchable: ค้นหาด้วย grep, type filter
📁 Git-Tracked: Version control ทุก knowledge
🔧 Portable: ใช้ได้กับทุก AI tool ที่อ่าน markdown
🤖 Auto-Capture: บันทึก session อัตโนมัติ พร้อม AI analysis
📋 Session Management: /recap → /focus → /td workflow
🔗 GitHub Integration: สร้าง Issues + PRs อัตโนมัติ
🎫 Jira Integration: สร้าง/จัดการ Jira issues (Atlassian Cloud)
✅ Code Review: /review ก่อน push
🔔 Notification Hooks: แจ้งเตือนเมื่อ Claude ต้องการ input (รองรับ Multi-Tab Workflow)
🤝 Multi-Agent Compatible: รองรับการใช้งานร่วมกับ multi-agent-auto-skill

Quick Start

For detailed usage examples and workflow guides, see USAGE.md

Installation

# Install all skills globally
npx skills add goffity/dev-km --full-depth --all -g

Alternative: Manual clone

# 1. Clone the skill
git clone https://github.com/goffity/dev-km.git ~/.claude/skills/dev-km

# 2. Install symlinks (makes all skills globally available)
bash ~/.claude/skills/dev-km/scripts/install-symlinks.sh

# 3. Initialize in your project
cd /path/to/your/project
~/.claude/skills/dev-km/scripts/init.sh .

Uninstall

npx skills remove dev-km

Commands

Session Management

Command	Purpose	Usage
`/recap`	เริ่ม session - โหลด context เดิม	`/recap`
`/focus [task]`	ตั้ง focus + สร้าง issue (GitHub/Jira)	`/focus Implement feature X`
`/td [status]`	จบ session: สร้าง retrospective	`/td done`

Git & Code

Command	Purpose	Usage
`/commit`	Atomic commits (via TDG)	`/commit`
`/pr`	Test + Build + Review + Create PR + Auto-respond	`/pr`
`/review`	Manual code review	`/review`
`/pr-review`	ตอบ PR feedback	`/pr-review`
`/pr-poll`	จัดการ PR review polling daemon	`/pr-poll start` / `/pr-poll auto`

Knowledge Capture (4-Layer)

Command	Layer	Purpose	Output
`/mem [topic]`	1	Quick capture ระหว่างงาน	`docs/learnings/YYYY-MM/DD/HH.MM_slug.md`
`/distill [topic]`	2	Extract patterns	`docs/knowledge-base/[topic].md`
`/td`	3	Post-task retrospective	`docs/retrospective/YYYY-MM/retrospective_*.md`
`/improve`	4	Work on pending items	Implementation

Knowledge & Docs

Command	Purpose	Usage
`/cleanup [days]`	Retention policy management	`/cleanup 30`
`/consolidate`	Daily session file consolidation	`/consolidate --execute`
`/summary [period]`	Weekly/monthly summaries	`/summary weekly`
`/search [query]`	Search knowledge index	`/search "auth pattern"`
`/example [lang] [name]`	Save code examples	`/example go retry-backoff`
`/flow [name]`	Process flow diagrams	`/flow deployment`
`/pattern [name]`	Design pattern docs	`/pattern retry-with-backoff`
`/share [path]`	Cross-project knowledge sharing	`/share docs/knowledge-base/topic.md`

Integration & Config

Command	Purpose	Usage
`/jira [cmd]`	Jira issue management	`/jira list PROJ`
`/permission`	Manage Claude Code permissions	`/permission suggest`

See references/jira-integration.md for full Jira documentation.

Session Lifecycle

┌─────────────────────────────────────────────────────────────────┐
│                        SESSION START                            │
└─────────────────────────────────────────────────────────────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │      /recap         │
                    │  โหลด context เดิม   │
                    └─────────────────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │      /focus         │
                    │  ตั้งงาน + สร้าง issue │
                    └─────────────────────┘
                               │
                               ▼
              ┌────────────────────────────────┐
              │         ทำงาน                   │
              │  /mem - บันทึก insight         │
              │  /commit - atomic commits      │
              │  /review - manual code review  │
              └────────────────────────────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │        /pr          │
                    │  test + build       │
                    │  code review        │
                    │  create PR          │
                    │  start auto-respond │
                    └─────────────────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │   [Auto-Respond]    │
                    │  poll for reviews   │
                    │  Claude handles     │
                    │  feedback auto      │
                    └─────────────────────┘
                               │
                               ▼
                    ┌─────────────────────┐
                    │        /td          │
                    │  retrospective      │
                    │  comment issue      │
                    └─────────────────────┘
                               │
                               ▼
┌─────────────────────────────────────────────────────────────────┐
│                        SESSION END                              │
└─────────────────────────────────────────────────────────────────┘

Quick Start Workflow

# 1. เริ่ม session
/recap                          # โหลด context + state เดิม

# 2. ตั้ง focus + สร้าง issue
/focus Implement feature X      # จะถามรายละเอียด (Overview, Technical Details, etc.)

# 3. ทำงาน...
/mem "JWT refresh pattern"      # บันทึก insight ระหว่างทาง
/commit                         # commit เมื่อเสร็จ chunk

# 4. สร้าง PR
/pr                             # test + build + review + สร้าง PR

# 5. จบ session
/td done                        # สร้าง retrospective + comment issue

Knowledge Layer Workflow

                    ระหว่างทำงาน
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  Layer 1: /mem                                                  │
│  Quick capture → docs/learnings/YYYY-MM/DD/HH.MM_slug.md       │
└─────────────────────────────────────────────────────────────────┘
                         │
                         │ (เมื่อมี 3+ learnings เรื่องเดียวกัน)
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  Layer 2: /distill                                              │
│  Extract patterns → docs/knowledge-base/topic.md               │
└─────────────────────────────────────────────────────────────────┘
                         │
                         │ (periodic review)
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  Layer 4: /improve                                              │
│  Work on pending items from knowledge-base + retrospectives    │
└─────────────────────────────────────────────────────────────────┘

Before/After Context

Feature เด่นจาก Claude-Mem ที่ช่วยให้เข้าใจบริบท:

## Context: Before

- **Problem**: MongoDB timeout under load
- **Existing Behavior**: Error "context deadline exceeded" after 30s
- **Metrics**: p99 = 2s, error rate = 5%

## Context: After

- **Solution**: Connection pool + retry with exponential backoff
- **New Behavior**: Connections stable under load
- **Metrics**: p99 = 200ms, error rate < 0.1%

Type Classification

ใน /td ระบุ type ใน frontmatter เพื่อ filter ได้ง่าย:

Type	Use When
`feature`	New functionality
`bugfix`	Bug fix
`refactor`	Code restructure
`decision`	Architecture decision
`discovery`	Research/learning
`config`	Configuration changes
`docs`	Documentation only

State Machine

          /focus
ready ─────────────▶ working ──/td done──▶ completed
  ▲                     │                      │
  │                     │ /td pending          │
  │ /focus new          ▼                      │
  └───────────────── pending ◀────────────────┘
                        │        /focus new task
                        │
                        │ /td blocked
                        ▼
                     blocked

State	Meaning	Next Action
`ready`	ว่าง รอ task	`/focus` เพื่อตั้งงาน
`working`	กำลังทำ	`/td` เมื่อจบ
`pending`	รอทำต่อ	`/recap` แล้ว `/focus`
`blocked`	ติดปัญหา	แก้ปัญหาก่อน
`completed`	เสร็จแล้ว	`/focus` งานใหม่

Project Directory Structure

After running init.sh, your project will have:

project/
└── docs/
    ├── current.md           # Current focus state
    ├── WIP.md               # Work in progress (paused tasks)
    ├── logs/
    │   └── activity.log     # Activity history
    ├── learnings/           # Layer 1: Quick capture (/mem)
    │   └── YYYY-MM/DD/
    ├── knowledge-base/      # Layer 2: Curated patterns (/distill)
    ├── retrospective/       # Layer 3: Full reviews (/td)
    │   └── YYYY-MM/
    ├── auto-captured/       # Auto-captured sessions
    │   └── YYYY-MM/DD/
    ├── examples/            # Code examples (/example)
    ├── summaries/           # Session summaries (/summary)
    ├── shared-knowledge/    # Cross-project knowledge (/share)
    ├── flows/               # Process flows (/flow)
    └── patterns/            # Design patterns (/pattern)

Search

# Find by type
grep -l "type: bugfix" docs/retrospective/**/*.md

# Search content
grep -r "mongodb" docs/

# Recent learnings (last 7 days)
find docs/learnings -name "*.md" -mtime -7

# List all decisions
grep -l "type: decision" docs/retrospective/**/*.md

Skill Structure

All 27 skills use the official Claude Code SKILL.md specification with YAML frontmatter.

dev-km/
├── SKILL.md                        # System overview (user-invocable: false)
├── skills/                         # 27 sub-skills
│   ├── mem/SKILL.md                # /mem - Quick knowledge capture
│   ├── focus/                      # /focus - Set task + create issue
│   │   ├── SKILL.md
│   │   └── jira-paths.md          # Jira integration detail
│   ├── td/                         # /td - Session retrospective
│   │   ├── SKILL.md
│   │   └── pr-commit-workflow.md
│   ├── recap/SKILL.md              # /recap - Load session context
│   ├── distill/SKILL.md            # /distill - Extract patterns
│   ├── improve/SKILL.md            # /improve - Work on pending items
│   ├── commit/SKILL.md             # /commit - Atomic commits
│   ├── pr/                         # /pr - Create PR
│   │   ├── SKILL.md
│   │   └── pr-post-create.md      # Auto-polling setup
│   ├── review/SKILL.md             # /review - Code review
│   ├── pr-review/                  # /pr-review - Handle PR feedback
│   │   ├── SKILL.md
│   │   ├── thread-resolution.md   # GraphQL thread resolve helpers
│   │   └── copilot-reviews.md     # Copilot review handling
│   ├── pr-poll/SKILL.md            # /pr-poll - PR review daemon
│   ├── cleanup/SKILL.md            # /cleanup - Retention policy
│   ├── consolidate/SKILL.md        # /consolidate - Session merger
│   ├── summary/SKILL.md            # /summary - Weekly/monthly summaries
│   ├── search/SKILL.md             # /search - Knowledge search
│   ├── jira/SKILL.md               # /jira - Jira integration
│   ├── example/SKILL.md            # /example - Code examples
│   ├── flow/SKILL.md               # /flow - Process flow docs
│   ├── pattern/SKILL.md            # /pattern - Design patterns
│   ├── share/SKILL.md              # /share - Cross-project sync
│   ├── permission/SKILL.md         # /permission - Permission management
│   ├── code-reviewer/SKILL.md      # Specialist (context: fork)
│   ├── session-analyzer/SKILL.md   # Specialist (context: fork)
│   ├── knowledge-curator/SKILL.md  # Specialist (context: fork)
│   ├── build-validator/SKILL.md    # Specialist (context: fork)
│   ├── code-simplifier/SKILL.md    # Specialist (context: fork)
│   └── security-auditor/SKILL.md   # Specialist (context: fork)
├── .claude/
│   └── settings.json               # Hook configurations
├── scripts/
│   ├── init.sh                     # Project setup script
│   ├── install-symlinks.sh         # Create 27 global symlinks
│   ├── uninstall-symlinks.sh       # Clean removal
│   ├── auto-capture.sh             # Auto session capture
│   ├── notify.sh                   # macOS notification script
│   └── ...
├── references/                     # Templates
├── assets/commands/                # Command source files (reference)
└── docs/                           # Knowledge data

Issue Tracker Integration

/focus รองรับทั้ง GitHub Issues และ Jira:

Supported Trackers

Tracker	Setup	Usage
GitHub Issues	ไม่ต้องตั้งค่า (ใช้ gh CLI)	Default option
Jira	`/jira init` ครั้งแรก	เลือกเมื่อรัน `/focus`

`/focus` Workflow

/focus [task description]
     │
     ├── ถ้ามี Jira config ──→ เลือก: GitHub / Jira (new) / Jira (existing)
     │
     └── ถ้าไม่มี Jira ──→ ใช้ GitHub Issues (default)

Jira Setup

# 1. สร้าง API Token ที่ https://id.atlassian.com/manage-profile/security/api-tokens

# 2. ตั้งค่า credentials
/jira init

# 3. ทดสอบ connection
/jira test

See references/jira-integration.md for full Jira documentation.

GitHub Integration

/focus และ /td integrate กับ GitHub โดยอัตโนมัติ:

Issue Format (/focus)

## Overview
Brief description of the feature/bug.

## Current State
What exists now.

## Proposed Solution
What should be implemented.

## Technical Details
- Components affected
- Implementation approach

## Acceptance Criteria
- [ ] Specific testable criteria
- [ ] Performance requirements

PR Format (/td)

## Summary
Brief summary of the changes made.

## Changes Made
- List of key changes.

## Testing
- Description of tests performed.

## Related Issues
- #issue_number

Fixes #issue_number

Pre-Push Checklist (via /pr)

Check	Command	Required
Tests	`make test`	Must pass
Build	`make build`	Must pass
Code Review	Subagent	No critical issues

Note: /pr จะรันทุกขั้นตอนนี้ให้อัตโนมัติพร้อม update issue ทุก step

PR Review Auto-Respond

หลังจาก /pr สร้าง PR เสร็จ จะ auto-start polling daemon ที่จัดการ reviews อัตโนมัติ:

/pr สร้าง PR เสร็จ
        ↓
Start PR Poll Daemon (--auto-respond)
        ↓
[รอ reviewer...]
        ↓
เมื่อมี review → Daemon ตรวจจับ
        ↓
Spawn Claude CLI → run /pr-review อัตโนมัติ
        ↓
แก้ code, reply comments, push changes

Event	Action
PR Approved	Notification (Glass sound)
Changes Requested	Claude auto-responds
Reviewer Comments	Claude auto-responds

Commands:

# ดูสถานะ daemon
/pr-poll status

# หยุด auto-respond
/pr-poll stop

# ดู Claude logs
tail -f ~/.pr-review-claude-*.log

Forbidden Actions

Action	Reason
Merge PR	ต้องรอ reviewer approve
Close Issue	จะปิดอัตโนมัติเมื่อ PR merge
Force Push	อันตราย ห้ามใช้
Skip Tests	ต้อง pass ก่อน push

Why Not Claude-Mem?

Feature	Claude-Mem	This Skill
Auto-capture	✅ Automatic	✅ Hooks/Wrapper/AI
Git tracked	❌	✅
Portable	❌ Claude Code only	✅ Any tool
Editable	Limited	✅ Full control
Structure	Fixed	✅ Customizable
Dependency	Plugin required	✅ Just markdown
GitHub Integration	❌	✅ Issues + PRs

Notification Hooks (Multi-Tab Workflow)

รองรับการทำงานแบบ Multi-Tab ตามแนวทางของ Boris Cherny (ผู้สร้าง Claude Code) - เปิดหลาย terminal tabs พร้อมกัน และรับ notification เมื่อ tab ใดต้องการ input

Notification Types

Type	เมื่อไหร่	Sound
`idle_prompt`	Claude รอ input นานเกิน 60 วินาที	Ping
`permission_prompt`	ต้องการ permission	Basso
`elicitation_dialog`	MCP tool ต้องการข้อมูลเพิ่ม	Purr

Setup

เพิ่มใน ~/.claude/settings.json:

{
  "hooks": {
    "Notification": [
      {
        "matcher": "idle_prompt",
        "hooks": [{
          "type": "command",
          "command": "~/.claude/skills/dev-km/scripts/notify.sh"
        }]
      },
      {
        "matcher": "permission_prompt",
        "hooks": [{
          "type": "command",
          "command": "~/.claude/skills/dev-km/scripts/notify.sh"
        }]
      },
      {
        "matcher": "elicitation_dialog",
        "hooks": [{
          "type": "command",
          "command": "~/.claude/skills/dev-km/scripts/notify.sh"
        }]
      }
    ]
  }
}

Multi-Tab Workflow

┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐
│  Tab 1  │  │  Tab 2  │  │  Tab 3  │  │  Tab 4  │  │  Tab 5  │
│ Feature │  │ Bug Fix │  │Refactor │  │  Tests  │  │  Docs   │
└────┬────┘  └────┬────┘  └────┬────┘  └────┬────┘  └────┬────┘
     │            │            │            │            │
     └────────────┴─────┬──────┴────────────┴────────────┘
                        │
                        ▼
              🔔 macOS Notification
              "Tab X needs your input"
                        │
                        ▼
                 สลับไป Tab นั้น

ใช้ร่วมกับ Git Worktrees

# สร้าง worktree สำหรับแต่ละ tab
git worktree add ../project-feature-a -b feature-a
git worktree add ../project-bugfix -b bugfix-123

# Tab 1: cd ../project-feature-a && claude
# Tab 2: cd ../project-bugfix && claude

Test Notification

echo '{"notification_type": "idle_prompt", "cwd": "/test"}' | \
  ~/.claude/skills/dev-km/scripts/notify.sh

Optional: terminal-notifier

สำหรับ notification ที่ดีกว่า (clickable, groupable):

brew install terminal-notifier

Auto-Capture

บันทึก session อัตโนมัติเมื่อจบงาน - 3 options:

Option 1: Hooks (Recommended)

# Add to ~/.claude/settings.json
{
  "hooks": {
    "Stop": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "[ ! -f docs/retrospective/$(date +%Y-%m)/retrospective_$(date +%Y-%m-%d)_*.md ] && ~/.claude/skills/dev-km/scripts/auto-capture.sh . 2>/dev/null || true"
      }]
    }]
  }
}

Note: Duplicate check prevents multiple captures on the same day.

Option 2: Wrapper

# Add alias
alias claude='~/.claude/skills/dev-km/scripts/claude-wrap.sh'

# Usage - shows summary and asks to capture
claude

Option 3: AI-Powered

export ANTHROPIC_API_KEY='your-key'
./scripts/ai-capture.sh .

Output: docs/auto-captured/YYYY-MM/DD/HH.MM_session-*.md

See AUTO-CAPTURE.md for full documentation.

Security

สคริปต์ทั้งหมดมีการป้องกันความปลอดภัย:

Path Validation

ป้องกัน path traversal attacks - ทุกสคริปต์ตรวจสอบ path ก่อนใช้งาน:

# ❌ จะถูก reject
./init.sh "../../../etc"
./auto-capture.sh "../../sensitive"

# ✅ ใช้งานได้
./init.sh .
./init.sh /path/to/project

Input Sanitization

notify.sh sanitize ทุก input ก่อนส่งให้ osascript:

Whitelist เฉพาะ alphanumeric, spaces, และ basic punctuation
Validate notification types (whitelist approach)
จำกัดความยาว input

Protected Scripts

Script	Protection
`init.sh`	Path validation, reject `..`
`auto-capture.sh`	Path validation
`ai-capture.sh`	Path validation
`notify.sh`	Input sanitization, type whitelist

Specialist Skills

Specialist skills run in forked context (context: fork) for automated workflows.

Available Specialists

Skill	Purpose	When to Use
`code-reviewer`	Review code for bugs, security, performance	Before commit/push
`code-simplifier`	Simplify code after writing	After coding
`security-auditor`	OWASP Top 10 security audit	Before push
`knowledge-curator`	Scan learnings, suggest distill topics	Weekly review
`session-analyzer`	Create retrospective drafts	End of session
`build-validator`	Validate build, tests, lint	Before push

Location

Specialists are defined as context: fork skills in skills/*/SKILL.md:

skills/
├── code-reviewer/SKILL.md
├── code-simplifier/SKILL.md
├── security-auditor/SKILL.md
├── knowledge-curator/SKILL.md
├── session-analyzer/SKILL.md
└── build-validator/SKILL.md

Contributing

Fork the repository
Create your feature branch (git checkout -b feature/amazing-feature)
Commit your changes (git commit -m 'feat: add amazing feature')
Push to the branch (git push origin feature/amazing-feature)
Open a Pull Request

See ROADMAP.md for planned features and current progress.

Related Projects

multi-agent-auto-skill - Multi-agent orchestration ที่ใช้ร่วมกับ skill นี้

Acknowledgments

Inspired by Claude-Mem
Inspired by weyermann-malt-productpage
Built for Claude Code
Multi-Tab workflow inspired by Boris Cherny

Name		Name	Last commit message	Last commit date
Latest commit History 163 Commits
.claude		.claude
.github/workflows		.github/workflows
assets/commands		assets/commands
docs		docs
references		references
scripts		scripts
skills		skills
.gitignore		.gitignore
AUTO-CAPTURE.md		AUTO-CAPTURE.md
LICENSE		LICENSE
README.md		README.md
ROADMAP.md		ROADMAP.md
SETUP.md		SETUP.md
SKILL.md		SKILL.md
USAGE.md		USAGE.md
hooks.json		hooks.json

License

goffity/dev-km

Folders and files

Latest commit

History

Repository files navigation

Knowledge Management Skill for Claude Code

Features

Quick Start

Installation

Uninstall

Commands

Session Management

Git & Code

Knowledge Capture (4-Layer)

Knowledge & Docs

Integration & Config

Session Lifecycle

Quick Start Workflow

Knowledge Layer Workflow

Before/After Context

Type Classification

State Machine

Project Directory Structure

Search

Skill Structure

Issue Tracker Integration

Supported Trackers

/focus Workflow

Jira Setup

GitHub Integration

Issue Format (/focus)

PR Format (/td)

Pre-Push Checklist (via /pr)

PR Review Auto-Respond

Forbidden Actions

Why Not Claude-Mem?

Notification Hooks (Multi-Tab Workflow)

Notification Types

Setup

Multi-Tab Workflow

ใช้ร่วมกับ Git Worktrees

Test Notification

Optional: terminal-notifier

Auto-Capture

Option 1: Hooks (Recommended)

Option 2: Wrapper

Option 3: AI-Powered

Security

Path Validation

Input Sanitization

Protected Scripts

Specialist Skills

Available Specialists

Location

Contributing

Related Projects

Acknowledgments

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 2

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`/focus` Workflow

Packages