---

name: analysis-iterative description: This skill should be used when the user says "brainstorm", "deep analysis", "let's think through", "analyze this with me", or "help me think through". Also used by /humaninloop:specify for input enrichment when feature descriptions lack Who/Problem/Value clarity. Provides progressive deep analysis through one-by-one questioning with 2-3 options per question and clear recommendations. Challenges disagreement to strengthen thinking and concludes with a synthesis document.

Iterative Analysis

Purpose

Guide deep thinking through progressive, one-at-a-time questioning. Each question builds on previous answers, creating a collaborative exploration that concludes with a structured synthesis document.

Workflow Phases

Phase 1: Opening

When triggered, acknowledge the topic and set expectations:

CopyI'll help you think through [topic] with a series of focused questions.

For each question, I'll present options with my recommendation. Feel free to
disagree—I'll push back to help strengthen your thinking.

Let's begin.

Then ask the first question.

Phase 2: Iterative Questioning

Core Rules:

1. ONE question per turn—never multiple questions
2. Always provide 2-3 concrete options
3. Always state your recommendation with reasoning
4. After each answer, briefly show how it affects the analysis before the next question

Question Format:

Copy[Brief context showing current understanding and how previous answer shaped thinking]

**Question [N]**: [Clear, focused question]

**Options:**
- **A) [Option name]**: [What this means and its implications]
- **B) [Option name]**: [What this means and its implications]
- **C) [Option name]**: [What this means and its implications]

**My Recommendation**: Option [X] because [clear reasoning based on what we know so far]

After receiving an answer:

Copy[Acknowledge the choice]

[Show how this shapes the analysis—what it opens up, what it rules out,
what becomes more important to explore]

[Transition to next question]

Phase 3: Handling Disagreement

When the user picks differently than recommended:

1. Explore their reasoning: "What's drawing you to that direction?"
2. Present counterarguments: Share your concerns respectfully but directly
3. If they maintain their choice: Accept it, integrate it, and proceed

Example:

CopyInteresting—you're leaning toward [their choice] over [your recommendation].

Before we lock that in, let me push back: [specific concern or trade-off they
may not have considered].

What's your thinking on that?

If they confirm after the challenge, acknowledge and integrate:

CopyFair enough. That's a deliberate choice with eyes open to the trade-offs.
Let me factor that in...

[Show how this affects the analysis]

Phase 4: Conclusion

When core decisions are made or user signals completion, generate the synthesis.

Transition to synthesis:

CopyI think we've covered the key decision points. Let me synthesize what we've
worked through...

Then generate the synthesis document. See SYNTHESIS.md for the template.

Adaptive Depth

Do NOT force a fixed number of questions. Continue until:

• Core decisions are made
• The user signals they're ready to conclude
• A natural stopping point is reached

Signs to wrap up:

• Answers are becoming more confirmatory than exploratory
• Key trade-offs have been explicitly addressed
• The user's confidence in direction is increasing

Examples

Good Question

CopyBased on your focus on rapid iteration, we need to decide on the feedback mechanism.

**Question 3**: How quickly do users need to see the impact of their changes?

**Options:**
- **A) Real-time (< 100ms)**: Changes reflect instantly. Highest technical complexity
  but most responsive feel.
- **B) Near-real-time (1-3 seconds)**: Brief processing delay. Simpler to implement,
  still feels responsive.
- **C) On-demand refresh**: User explicitly triggers update. Simplest, but breaks flow.

**My Recommendation**: Option B because your team is small and time-to-market
matters. Real-time adds significant complexity for marginal UX gain at this stage.
You can always upgrade later.

Good Disagreement Handling

CopyYou're leaning toward real-time over my near-real-time recommendation.

I want to make sure you've weighed this: real-time means WebSocket infrastructure,
handling reconnection logic, and conflict resolution—easily 2-3x the implementation
work. Your team of three would be stretched thin.

Is the instant feedback worth that trade-off, or is there a specific user flow
where real-time is non-negotiable?

Good Transition After Answer

CopyGot it—you're prioritizing simplicity with the on-demand refresh approach.

That's a pragmatic call. It means we can skip the real-time infrastructure entirely
and focus engineering time on the core value proposition. Users will need a clear
"refresh" affordance, but that's a UI detail we can nail.

This actually opens up our next question about data freshness expectations...

Anti-Patterns to Avoid

• Asking multiple questions in one turn
• Presenting options without a clear recommendation
• Accepting user's choice without exploring disagreement
• Asking questions that don't build on previous answers
• Rushing to conclusion before core decisions are made
• Making the synthesis before explicitly transitioning

---

Modes

This skill supports specialized modes for specific use cases.

Specification Input Enrichment

When invoked with mode:specification-input, this skill runs a focused variant for enriching sparse feature descriptions. See SPECIFICATION-INPUT.md for details.