customer-support

# Customer Support — Customer Support System Building Pipeline

An agent team collaborates to generate FAQ, response manual, escalation policy, and analytics all at once.

## Execution Mode

**Agent Team** — 5 members communicate directly via SendMessage and perform cross-validation.

## Agent Composition

| Agent | File | Role | Type |
|-------|------|------|------|
| faq-builder | `.claude/agents/faq-builder.md` | FAQ construction, search optimization | general-purpose |
| response-specialist | `.claude/agents/response-specialist.md` | Response scripts, tone & manner | general-purpose |
| escalation-manager | `.claude/agents/escalation-manager.md` | Escalation tiers, SLA | general-purpose |
| cs-analyst | `.claude/agents/cs-analyst.md` | Metrics, VOC, improvement proposals | general-purpose |
| cs-reviewer | `.claude/agents/cs-reviewer.md` | Cross-validation, customer journey simulation | general-purpose |

## Workflow

### Phase 1: Preparation (Performed directly by orchestrator)

1. Extract from user input:
    - **Service Information**: Product/service name, type, customer base size
    - **Support Channels**: Which channels are in use — phone/chat/email/social media
    - **Current State**: Whether an existing CS system exists, team size
    - **Existing Materials** (optional): Existing FAQs, manuals, data
2. Create `_workspace/` directory at the project root
3. Organize the input and save to `_workspace/00_input.md`
4. If existing files are provided, copy them to `_workspace/` and skip the corresponding Phase

### Phase 2: Team Assembly and Execution

| Order | Task | Assigned To | Dependencies | Deliverable |
|-------|------|-------------|-------------|-------------|
| 1a | FAQ Construction | faq-builder | None | `_workspace/01_faq.md` |
| 1b | Response Manual | response-specialist | None | `_workspace/02_response_manual.md` |
| 2 | Escalation Policy | escalation-manager | Tasks 1a, 1b | `_workspace/03_escalation_policy.md` |
| 3 | CS Analytics Framework | cs-analyst | Tasks 1a, 1b, 2 | `_workspace/04_cs_analytics.md` |
| 4 | CS Review | cs-reviewer | Tasks 1a, 1b, 2, 3 | `_workspace/05_review_report.md` |

Tasks 1a (FAQ) and 1b (Manual) are **executed in parallel**. Both can start simultaneously as they have no initial dependencies.

**Inter-team Communication Flow:**
- faq-builder completes → delivers out-of-FAQ-scope scenarios to response-specialist, delivers self-service boundary to escalation-manager
- response-specialist completes → delivers escalation trigger conditions to escalation-manager
- escalation-manager completes → delivers SLA metrics to cs-analyst
- cs-analyst completes → delivers all documents to cs-reviewer
- cs-reviewer cross-validates all deliverables. If 🔴 Must Fix items are found, sends revision requests to the relevant agent → rework → re-validate (up to 2 times)

### Phase 3: Integration and Final Deliverables

1. Verify all files in `_workspace/`
2. Confirm that all 🔴 Must Fix items from the review report have been addressed
3. Report the final summary to the user

## Execution Modes by Request Scale

| User Request Pattern | Execution Mode | Agents Deployed |
|---------------------|----------------|-----------------|
| "Build a customer support system", "Full CS setup" | **Full Pipeline** | All 5 |
| "Just create FAQs" | **FAQ Mode** | faq-builder + reviewer |
| "Write a response manual" | **Manual Mode** | response-specialist + reviewer |
| "Create an escalation policy" | **Escalation Mode** | escalation-manager + reviewer |
| "Design CS metrics" | **Analytics Mode** | cs-analyst + reviewer |
| "Review this FAQ" | **Review Mode** | reviewer only |

## Data Transfer Protocol

| Strategy | Method | Purpose |
|----------|--------|---------|
| File-based | `_workspace/` directory | Storing and sharing main deliverables |
| Message-based | SendMessage | Real-time key information transfer, revision requests |

## Error Handling

| Error Type | Strategy |
|-----------|----------|
| Insufficient service info | Build a generic CS system based on industry type, tag as "requires service customization" |
| No existing CS data | Set targets based on industry benchmarks, propose initial measurement methods |
| Web search failure | Work from general CS best practices, note "data limited" |
| Agent failure | Retry once → if still fails, proceed without that deliverable, note the omission in the review report |
| 🔴 found in review | Send revision request to the relevant agent → rework → re-validate (up to 2 times) |

## Test Scenarios

### Normal Flow
**Prompt**: "Build a customer support system for a B2C SaaS project management tool. We have 10,000 monthly active users and operate chat and email channels. The CS team has 5 members."
**Expected Results**:
- FAQ: Sign-up/billing/features/integration categories, 30+ Q&As
- Manual: 5+ scenario scripts, channel-specific guides for chat + email
- Escalation: L1-L3 tiers, SLA matrix, crisis response protocol
- Analytics: CSAT/NPS/FCR/AHT metrics + weekly/monthly report templates
- Review: Includes 3 customer journey simulations

### Existing File Utilization Flow
**Prompt**: "I have this FAQ; create a response manual and escalation policy" + FAQ file attached
**Expected Results**:
- Copy existing FAQ to `_workspace/01_faq.md`
- Skip faq-builder and deploy response-specialist + escalation-manager + cs-analyst + reviewer

### Error Flow
**Prompt**: "Create a customer support FAQ"
**Expected Results**:
- Switch to FAQ Mode (faq-builder + reviewer)
- If service info is insufficient, ask follow-up questions about industry/product type
- If only minimal info is provided, generate a generic SaaS FAQ template

## Agent Extension Skills

Extension skills that enhance agent domain expertise:

| Skill | File | Target Agents | Role |
|-------|------|---------------|------|
| csat-analyzer | `.claude/skills/csat-analyzer/skill.md` | cs-analyst, cs-reviewer | CSAT/NPS/CES design, operational metrics, VOC analysis, CS dashboard design |
| escalation-flowchart | `.claude/skills/escalation-flowchart/skill.md` | escalation-manager, response-specialist | L1/L2/L3 structure, severity classification, trigger conditions, SLA matrix, crisis response |