clarification-protocol

Clarification Protocol

Generates focused, challenging questions to extract missing context and clarify vague requirements before routing to specialist agents.

When to Use

• After request-analyzer identifies vague terms or missing context
• When requirements are ambiguous and could lead to wrong solutions
• Before delegating to cto-architect or strategic-cto-mentor
• When buzzwords need to be translated into specific requirements

Core Principles

1. Maximum 2-3 Questions Per Round

Users lose patience with long questionnaires. Prioritize ruthlessly:

• Ask only what's blocking progress
• Combine related questions
• Defer nice-to-have information

2. Challenge Mode, Not Interview Mode

Don't just ask—challenge assumptions:

• Bad: "What scale do you need?"
• Good: "You mentioned 'scalable'—are we designing for 10K users or 10M? That changes the architecture significantly."

3. Provide Example Answers

Help users understand what you're looking for:

• Bad: "What's your timeline?"
• Good: "What's your timeline? For context, a robust MVP typically takes 8-12 weeks with a team of 4."

Question Generation Framework

Step 1: Prioritize Missing Information

Rank by impact on routing and design:

Priority	Category	Examples
P0	Blocking	Can't proceed without this (e.g., "What problem does AI solve here?")
P1	High Impact	Significantly changes approach (e.g., "1K or 100K users?")
P2	Medium Impact	Affects details but not direction (e.g., "Budget range?")
P3	Nice to Have	Can be discovered later (e.g., "Preferred cloud provider?")

Ask P0 first, then P1. Defer P2/P3.

Step 2: Select Question Type

Type	When to Use	Template
Scope	Vague feature description	"When you say [X], do you mean [A] or [B]?"
Scale	Missing numbers	"What scale are we designing for? [X] users? [Y] requests/second?"
Timeline	Vague deadlines	"What's the actual deadline? Is there flexibility if scope changes?"
Constraint	Unknown limitations	"Are there constraints I should know about? Budget, team size, existing systems?"
Success	Unclear goals	"How will we know this succeeded? What metrics matter?"

Step 3: Frame as Challenge

Transform neutral questions into challenging ones:

Neutral (Weak)	Challenge (Strong)
"What do you mean by AI-powered?"	"'AI-powered' can mean many things—are you trying to classify data, generate content, or make predictions? What specific problem are you solving?"
"What's your budget?"	"Knowing budget helps me avoid over-engineering. Are we optimizing for < $5K/month or is $50K/month acceptable for the right solution?"
"How many users?"	"You mentioned 'scale'—I want to avoid designing for problems you don't have. How many users today, and what's realistic in 12 months?"

Output Format

## Clarification Needed

Before I route this to the appropriate specialist, I need to understand a few things:

### Challenge 1: [Vague Term/Concept]
[Direct challenge statement]

**Question**: [Specific question]

**Why this matters**: [Impact on design/routing]

**Example answers**:
- Option A: [example] → leads to [approach]
- Option B: [example] → leads to [different approach]

### Challenge 2: [Missing Context]
[...]

---

Once you clarify these points, I'll route to [suggested agent] with a clear brief.

Question Templates by Category

See question-templates.md for comprehensive templates.

See challenge-patterns.md for examples of turning neutral questions into challenges.

Examples

Example 1: AI Feature Request

User Request: "We want to add AI capabilities to improve customer experience"

Clarification Output:

## Clarification Needed

Before I route this to cto-architect, I need to understand what "AI capabilities" means for your use case:

### Challenge 1: Vague AI Requirement
"AI capabilities" could mean anything from a chatbot to predictive analytics. I don't want to design the wrong solution.

**Question**: What specific customer problem are you trying to solve with AI?

**Why this matters**: A recommendation engine has completely different architecture than a support chatbot.

**Example answers**:
- "Customers can't find products" → Recommendation engine / search improvement
- "Support tickets are overwhelming" → AI-assisted ticket routing or chatbot
- "Churn is high" → Predictive analytics for at-risk customers

### Challenge 2: Missing Scale
**Question**: How many customers are we talking about? And how many interactions per day would this AI feature handle?

**Why this matters**: 1,000 daily interactions vs 1,000,000 changes whether we use a simple API call or need dedicated ML infrastructure.

---

Once you clarify these points, I'll route to cto-architect with a clear brief for the appropriate AI architecture.

Example 2: Scaling Concerns

User Request: "Our app needs to scale better"

Clarification Output:

## Clarification Needed

"Scale better" is a symptom—I need to understand the disease before prescribing treatment.

### Challenge 1: What's Actually Broken?
**Question**: What specific scaling issue are you experiencing? Slow page loads? Database timeouts? High infrastructure costs? All of the above?

**Why this matters**: Scaling a slow database is different from scaling a CPU-bound API.

**Example answers**:
- "Page loads take 5+ seconds at peak" → Likely caching/CDN solution
- "Database queries timing out" → Database scaling or query optimization
- "AWS bill doubled last month" → Cost optimization vs actual scaling

### Challenge 2: Current vs Target Scale
**Question**: What's your current traffic, and what do you need to support? Be specific: requests per second, concurrent users, data volume.

**Why this matters**: "Scale" at 10K users looks very different from "scale" at 10M users. I don't want to over-engineer.

---

Once I understand the specific bottleneck, I'll route to the appropriate specialist.

Anti-Patterns to Avoid

1. The Interrogation

Bad: Asking 10 questions at once Good: Max 2-3 targeted questions

2. The Open-Ended Trap

Bad: "Tell me more about your requirements" Good: "Is this for internal users (hundreds) or external customers (thousands+)?"

3. The Assumption Question

Bad: "What microservices architecture do you want?" (assumes microservices) Good: "What's your current architecture, and what's driving the need to change?"

4. The Jargon Barrier

Bad: "What's your CAP theorem preference for the distributed system?" Good: "If the system goes offline briefly, should it prioritize consistency (everyone sees the same data) or availability (the system stays up)?"

References

• Question Templates - Ready-to-use question templates by category
• Challenge Patterns - Examples of effective challenge framing