---

name: knowledge-management description: 4-layer knowledge capture system for development sessions. Use when user says /mem (quick capture), /distill (extract patterns), /td (post-task retrospective with Before/After context), /improve (work on pending items), or /commit (atomic commits via TDG). Manages docs/learnings/, docs/knowledge-base/, and docs/retrospective/ directories.

Knowledge Management System

4-layer system for capturing and organizing development knowledge.

Commands

Command	Layer	Output	Trigger
/mem [topic]	1	docs/learnings/YYYY-MM/DD/HH.MM_slug.md	Quick insight capture
/distill [topic]	2	docs/knowledge-base/[topic].md	3+ learnings on same topic
/td	3	docs/retrospective/YYYY-MM/retrospective_*.md	Task completed
/improve	4	Implementation	Work on pending items
/commit	-	Git commits	Atomic commits via TDG
/pr-review	-	Learning doc + PR updates	Handle PR review feedback
/pr-poll	-	Notification daemon	Auto PR review notifications
/cleanup	-	Archive + cleanup	Retention policy management

Flow

ทำงาน → /mem "insight" → /distill topic → /td → /improve
         (Layer 1)        (Layer 2)      (Layer 3) (Layer 4)

Directory Structure

docs/
├── learnings/           # /mem output
│   └── YYYY-MM/DD/
├── knowledge-base/      # /distill output
└── retrospective/       # /td output
    └── YYYY-MM/

Setup

Run init script to create directory structure:

./scripts/init.sh $PROJECT_ROOT

Or manually:

mkdir -p docs/{learnings,knowledge-base,retrospective}

---

Command: /mem

Quick knowledge capture - ใช้ระหว่างทำงานเมื่อพบ insight

TZ='Asia/Bangkok' date '+%Y-%m/%d/%H.%M'  # Path format

Output: docs/learnings/YYYY-MM/DD/HH.MM_[slug].md

Template: See references/mem-template.md

Key sections: Key Insight, What We Learned, Gotchas, Tags

---

Command: /distill

Extract patterns - รวม learnings เป็น reusable patterns

When: มี 3+ learnings เรื่องเดียวกัน หรือ weekly review

Output: docs/knowledge-base/[topic-name].md

Template: See references/distill-template.md

Key sections: Key Insight, The Problem, The Solution (with code), Anti-Patterns, When to Apply

After: Mark source learnings as "Distilled"

---

Command: /td

Post-task retrospective with Before/After context

Output: docs/retrospective/YYYY-MM/retrospective_YYYY-MM-DD_hhmmss.md

Template: See references/td-template.md

Type Classification (frontmatter)

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

Required: Before/After Context

## Context: Before
- **Problem**: ปัญหาที่เจอ
- **Existing Behavior**: พฤติกรรมเดิม
- **Metrics**: ตัวเลขก่อนแก้

## Context: After
- **Solution**: วิธีแก้
- **New Behavior**: พฤติกรรมใหม่
- **Metrics**: ตัวเลขหลังแก้

Decisions Table

| Decision | Options Considered | Chosen | Rationale |
|----------|-------------------|--------|-----------|

---

Command: /commit

Atomic commits via TDG plugin

Delegates to: /tdg:atomic-commit from https://github.com/chanwit/tdg

What it does:

• Analyzes staged/unstaged changes
• Detects mixed concerns (multiple unrelated changes)
• Helps create clean, focused atomic commits
• Each commit is a complete unit of work

Usage: Simply run /commit and it will invoke TDG's atomic-commit skill.

---

Command: /improve

Work on pending items from all knowledge sources

Scan order (priority):

1. docs/knowledge-base/ - Patterns to apply
2. docs/retrospective/ - Future Improvements (- [ ])
3. docs/learnings/ - Gotchas to fix (skip if "Distilled")

Workflow:

1. Extract unchecked items
2. Present prioritized list
3. User selects items
4. Implement & commit
5. Update source file (- [ ] → - [x])

---

Search Commands

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

# Search content
grep -r "mongodb" docs/

# Recent learnings
find docs/learnings -name "*.md" -mtime -7

---

Command: /pr-poll

Automatic PR review notifications - Polling daemon ตรวจสอบ PR reviews

What it does:

• Poll GitHub for user's open PRs
• Detect new reviews, comments, and review decisions
• Send macOS notifications with review details
• Suggest running /pr-review to respond

Usage:

/pr-poll              # Show daemon status
/pr-poll start        # Start polling daemon
/pr-poll stop         # Stop daemon
/pr-poll check        # Check once without daemon

Notification sounds:

• APPROVED → Glass
• CHANGES_REQUESTED → Basso
• COMMENTED → Ping

Files:

• ~/.pr-review-poll.pid - Daemon PID
• ~/.pr-review-poll.log - Daemon logs
• ~/.pr-review-state.json - PR state tracking

---

Command: /cleanup

Retention policy management - จัดการไฟล์เก่าด้วย retention policy

What it does:

• Delete old auto-captured files (configurable retention period)
• Archive old files before deletion (optional)
• Dry-run mode to preview changes
• Clean draft learnings that haven't been distilled

Usage:

/cleanup                    # Preview (30 days default)
/cleanup 7                  # Preview with 7 days retention
/cleanup 14 --archive       # Archive & delete files older than 14 days
/cleanup --all              # Clean all targets

Protected directories (never auto-deleted):

• docs/retrospective/ - Permanent records
• docs/knowledge-base/ - Distilled knowledge

---

Auto-Capture

บันทึก session อัตโนมัติเมื่อจบงาน

Options

Option	Command	Trigger
Hooks	-	Auto on session stop
Wrapper	claude (alias)	Interactive prompt
AI-Powered	./scripts/ai-capture.sh	Manual with AI analysis

Setup (Hooks - Recommended)

# Add to ~/.claude/settings.json
{
  "hooks": {
    "Stop": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "~/.claude/skills/knowledge-management/scripts/auto-capture.sh . 2>/dev/null || true"
      }]
    }]
  }
}

Output

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

See AUTO-CAPTURE.md for full documentation.

---

References

• references/mem-template.md - Full /mem template
• references/distill-template.md - Full /distill template
• references/td-template.md - Full /td template
• references/improve-workflow.md - /improve detailed workflow