Meta Workshop

สร้าง AI ที่สร้าง Workshop

AI builds workshop → You learn how → You build your own

The Loop

nnn → gogogo → rrr

CodeActionResult
nnnวางแผนGitHub Issue
gogogoลงมือทำCode + Commits
rrrสรุปRetrospective

Plan → Execute → Reflect → Repeat

Short Codes in Action

วันนี้ใช้จริง:

nnn     → สร้าง issue #8 (plan slides)
gogogo  → สร้าง slides + PR
/trace  → หา context ข้าม repos

ผลลัพธ์: 6 issues, 3 PRs, 34+ slides

/trace - Parallel Agents

# /trace [project-name]

Launch 5 subagents simultaneously:

1. Current Repo Files  → grep/find
2. Git History         → git log --grep
3. GitHub Issues       → gh issue list
4. Other Repos         → find ~/Code
5. Retrospectives      → grep ψ/memory/

5 agents = 5x faster search

/trace - Header

---
description: ตามรอย Project ที่หายไป
allowed-tools:
  - Bash
  - Grep
  - Glob
  - Task
---

# /trace - ตามรอย Project ที่หายไป

## Step 0: Timestamp (REQUIRED)
date "+🕐 %H:%M (%A %d %B %Y)"

ARGUMENTS: $ARGUMENTS

allowed-tools = security boundary

/trace - Agent 1: Current Repo

### Agent 1: Current Repo Files

subagent_type: context-finder
model: haiku
prompt: |
  รัน bash แล้วสรุป:

  # Root files ที่ mention project
  grep -il "[PROJECT]" *.md 2>/dev/null

  # Files + Folders ที่ชื่อตรง
  find . -iname "*[PROJECT]*" -type f | head -20
  find . -iname "*[PROJECT]*" -type d | head -10

  # Content mentions ใน subfolders
  grep -ril "[PROJECT]" --include="*.md" \
    tools/ projects/ ψ/ .claude/ | head -20

  Return: list ของ paths ที่เจอ

/trace - Agent 2: Git History

### Agent 2: Git History

subagent_type: context-finder
model: haiku
prompt: |
  รัน bash แล้วสรุป timeline:

  # Commits ที่ mention project
  git log --all --oneline --grep="[PROJECT]" | head -15

  # Files created/deleted
  git log --all --oneline \
    --diff-filter=A -- "*[PROJECT]*" | head -10
  git log --all --oneline \
    --diff-filter=D -- "*[PROJECT]*" | head -10

  Return: timeline (Date | Action | Commit)

/trace - Agents 3, 4, 5

### Agent 3: GitHub Issues
gh issue list --state all --limit 30 \
  --search "[PROJECT]" --json number,title,createdAt

### Agent 4: Other Repos
find ~/Code/github.com -maxdepth 3 \
  -iname "*[PROJECT]*" -type d | head -10

### Agent 5: Retrospectives & Docs
grep -ril "[PROJECT]" ψ/memory/retrospectives/ | head -10
grep -ril "[PROJECT]" ψ/memory/learnings/ | head -5
grep -ril "[PROJECT]" ψ/writing/ | head -5

All 5 agents run in parallel!

/trace - Output Format

## 🔍 /trace: [PROJECT]

### 📍 Locations Found
| Source | Path | Status |

### 📅 Timeline (from git)
| Date | Action | Commit | Details |

### 🎫 GitHub Issues
| # | Date | Title |

### 📂 Files (current state)
| Type | Path | Repo |

---
**Current Location**: [where code lives now]
**Recommended Action**: [what to do next]

GitHub Flow

Issue → Branch → Code → PR → Merge

StepCommand
1. Issuegh issue create
2. Branchgit checkout -b feat/xxx
3. CodeEdit files
4. PRgh pr create
5. MergeReview → Merge

ไม่ commit ตรงไป main

Content Iteration

v1 (27 slides) → "too detail!" → v2 (12 slides)

VersionSlidesFeedback
v127Too tight, too detail
v212Good! Overview style

Lesson: Overview > Detail

Deployment Made Simple

// wrangler.jsonc
{
  "name": "claude-code-workshops",
  "assets": {
    "directory": "./public"
  }
}
npx wrangler deploy

No worker.js needed for static sites

Knowledge Organization

vault/
├── 00-MOC.md          # Map of Content
├── 1x-concepts/       # 10, 11, 12...
├── 2x-short-codes/    # 20, 21, 22...
├── 3x-workflows/      # 30, 31...
└── 4x-slides/         # 40, 41, 42...

Numbered prefixes = ordering

Anti-Patterns

Don'tDo Instead
Detail firstOverview first, iterate
Commit to mainBranch → PR → Merge
Guess user wantsAsk, get feedback
Manual searchUse /trace parallel agents

Create Your Own /command

---
description: Your command description
allowed-tools:
  - Bash
  - Task
---

# /your-command

## Step 1: Do something
...

## Step 2: Launch agents
subagent_type: context-finder
prompt: |
  Your prompt here

Save to: .claude/commands/your-command.md

Key Takeaways

  1. Loop: nnn → gogogo → rrr = structured work
  2. Parallel: 5 agents searching = fast results
  3. Iterate: v1 → feedback → v2 = better output

Plan before doing.

Search in parallel.

Learn from feedback.