Durasi: 30 menit | Block: 2

---

🅰️ Project A: Asisten Operasional UMKM ⭐⭐

Scenario

Agent yang "kenal" bisnis kamu — bisa jawab pertanyaan operasional dan generate daily brief.

Deliverables

SOUL.md:

Nama: [Asisten Bisnis]
Role: Operations Assistant
Tone: Ramah, helpful, data-driven

Context Files:

• products.md — catalog produk
• faq.md — FAQ operasional

Cron Job:

- name: "daily-brief"
  schedule: "0 8 * * *"
  task: "Baca data kemarin, buat morning brief"

Test Scenarios

1. FAQ: "Berapa harga produk X?" → cek context
2. Boundary: "Bisa kasih diskon 50%?" → respect boundaries
3. Brief: Trigger cron → generate summary

---

🅱️ Project B: Social Media Manager Agent ⭐⭐⭐

Scenario

Agent yang bisa generate content ideas, check competitor, dan schedule content.

Deliverables

SOUL.md:

Nama: [Social Media Manager]
Role: Content Strategist
Tone: Creative, trendy, on-brand

Context Files:

• brand-guidelines.md — brand voice, do/don't
• content-calendar.md — schedule template
• products.md — product info

Tools:

• Web search (competitor check, trend research)
• File write (save content drafts)

Cron:

- name: "competitor-check"
  schedule: "0 10 * * 5"  # Jumat
  task: "Check 3 competitor social media, summarize"

Test Scenarios

1. Content: "Generate 3 post ideas untuk promo minggu depan"
2. Research: "Cek competitor X di Instagram, apa konten terbaru mereka?"
3. Boundary: "Post ini tanpa review" → harus minta approval

---

🅲️ Project C: Customer Service Agent ⭐⭐⭐

Scenario

Agent yang handle customer inquiry, FAQ, dan escalate kalau perlu.

Deliverables

SOUL.md:

Nama: [CS Agent]
Role: Customer Service
Tone: Empathetic, solution-oriented
Boundaries: 
- JANGAN janji kompensasi
- JANGAN handle refund > 500K
- ESCALATE kalau customer marah

Context Files:

• faq.md — lengkap (produk, pembayaran, pengiriman)
• products.md — catalog
• sop-complaint.md — complaint handling procedure

Evaluator-Optimizer:

Quality Rules:
- Self-evaluate setiap response
- Check: empathetic? accurate? within boundaries?
- Re-generate kalau score < 8/10

Test Scenarios

1. FAQ: "Berapa ongkir ke Surabaya?" → answer from context
2. Complaint: "Produk saya rusak!" → empathize + escalate SOP
3. Boundary: "Mau refund 1 juta" → flag for human review

---

🅳️ Project D: Research & Report Agent ⭐⭐⭐⭐

Scenario

Agent yang bisa research mandiri, compile data, dan deliver report otomatis.

Deliverables

SOUL.md:

Nama: [Research Analyst]
Role: Market Research & Analysis
Tone: Analytical, data-driven, objective

Context Files:

• market-overview.md — market context
• competitors.md — competitor database
• research-template.md — report format template

Tools:

• Web search (market research)
• Browser (scrape specific sites)
• File read/write (compile & save reports)

Cron:

- name: "weekly-market-brief"
  schedule: "0 9 * * 1"
  task: >
    Research market updates:
    1. Industry news
    2. Competitor changes
    3. Trend analysis
    Compile into formatted report.
  output: "output/market-brief-[date].md"

Evaluator-Optimizer:

Quality Rules:
- Verify facts from 2+ sources
- Include data/citations
- Self-evaluate: comprehensive? actionable?
- Max 2 improvement cycles

Test Scenarios

1. Research: "Cari trend kopi specialty 2026 di Indonesia"
2. Report: "Compile weekly report dan simpan"
3. Quality: Output harus melewati evaluation criteria

---

🎯 Build Strategy

1. Start with SOUL.md (5 min) — define persona dulu
2. Create context files (10 min) — isi knowledge base
3. Configure tools (5 min) — enable yang diperlukan
4. Setup cron (5 min) — kalau applicable
5. Test 3 scenarios (5 min) — FAQ, boundary, tool usage