claude-expert-skill-creator

### Skill Name: **claude-expert-skill-creator**

**Purpose**: Automate the creation of expert-level AI skills by extracting domain knowledge through guided interviews, separating deterministic logic from flexible interpretation, and packaging the result for reuse.

**Interview Transcript (Excerpt)**:
*AI*: "What specific problem does this skill solve?"
*User*: "I want to create an AI skill that automates the generation of customer support responses for a SaaS company's help desk."

*AI*: "Define 'customer support response' for this context."
*User*: "A response to a user's query about billing, account access, or feature usage. It should include: (1) a clear answer, (2) relevant links, and (3) a tone matching the user's sentiment (e.g., frustrated vs. neutral)."

*AI*: "Provide 3 examples of edge cases where the response might need to deviate from the standard template."
*User*: "1) A user threatens to cancel due to billing issues. 2) A user reports a critical bug affecting their workflow. 3) A user asks about a feature not yet released."

**Deterministic Logic Identified**:
- Script to fetch user sentiment from ticket metadata (e.g., 'frustrated' if ticket tags include 'angry').
- Script to pull relevant knowledge base articles based on keywords (e.g., 'billing' → link to billing FAQ).
- Script to route responses to human agents if sentiment = 'critical' or ticket priority = 'high'.

**Flexible Interpretation Identified**:
- LLM guidance to generate the response body, incorporating user sentiment and edge cases.
- LLM guidance to suggest follow-up actions (e.g., 'Would you like to schedule a call?').

**Packaged Skill Structure**:
```
skill/
├── __init__.py
├── deterministic/
│   ├── sentiment_analyzer.py
│   ├── kb_router.py
│   └── escalation_handler.py
├── llm/
│   ├── response_generator.py
│   └── followup_suggester.py
├── templates/
│   ├── response_base.md
│   └── edge_case_handlers.json
└── examples/
    ├── billing_query.json
    ├── critical_bug_report.json
    └── feature_request.json
```

**Usage Example**:
```python
from skill.deterministic.sentiment_analyzer import analyze_sentiment
from skill.llm.response_generator import generate_response

# Input: User ticket with metadata
ticket = {
    "id": "TKT-12345",
    "query": "Why was I charged twice this month?",
    "tags": ["billing", "angry"],
    "kb_articles": ["KB-456", "KB-789"]
}

# Step 1: Analyze sentiment
sentiment = analyze_sentiment(ticket["tags"])

# Step 2: Generate response
response = generate_response(
    query=ticket["query"],
    sentiment=sentiment,
    kb_links=ticket["kb_articles"]
)

# Output: 
# {
#   "response": "We apologize for the double charge... [KB-456 link]...",
#   "followup": "Would you like us to refund the duplicate charge?"
# }
```