Create Any‑Domain MVP in 30 Minutes with the Diverge‑Converge Skill

01 Diffusion and Convergence

The author first tried to design an MCP that could fetch logs with a single natural‑language command, but the idea was vague. By telling the AI "first diffuse, then converge," a one‑hour dialogue generated a 10,000‑word implementation plan covering architecture layers, technology choices, directory structure, code skeleton, token signing, and log‑path handling, with each decision explained.

02 The Double‑Diamond Model

The method draws on the Double Diamond design framework (British Design Council) and Guilford’s psychological theory of divergent and convergent thinking. The first diamond discovers the problem (diffuse then converge), the second solves it (diffuse then converge). The author extracts these proven concepts, adds AI‑agent semantics, and packages them as a reusable skill.

03 Perfect Dialogue

For each AI turn the author requires: (1) a confidence score (e.g., "suggest solution B with 85% confidence"), (2) a side‑by‑side table comparing alternative options, (3) only one critical question per round, focusing on the decision point that determines subsequent design choices. The AI expands the map, the human selects direction, and the process repeats until the MVP is fully specified.

04 The Diverge‑Converge Skill

This skill is a generic questioning framework, not tied to any specific software. It can be used by graduate students to map literature and hypotheses, entrepreneurs to explore market, competitors, risks, and resources, or writers to outline articles. It accelerates framing but never replaces human judgment; the final manuscript is a starting point, not a finished product.

## What this skill is
It defines a **protocol**, not a script: when a user brings an immature idea, the skill tells the agent how to act as a thinking partner, using controlled diffusion and convergence to turn the vague idea into a concrete, implementable plan, finally crystallized as a manuscript.

## Why "first diffuse, then converge"
- Skipping diffusion leads to premature commitment and costly rework.
- Diffusion expands unknown‑unknowns.
- Convergence locks decisions one by one until a complete path emerges.
- The two phases form a dual‑engine that gradually shifts focus from expansion to narrowing.

The skill is named diverge-converge and hosted at https://github.com/linshidream/skill-hub . Installing it under ~/skills/ enables a user to say "I have an idea, first diffuse then converge" and let the agent iteratively clarify the concept. When the user commands "generate implementable document," the agent compiles all decisions, rationales, rejected alternatives, and open questions into a manuscript that can be handed off to another agent or to a human for execution.

05 Final Thoughts

The author, a software engineer, treats the diverge‑converge method as a daily thinking model for any complex problem. While AI can quickly build a framework, the ultimate value lies in human aesthetic sense and judgment.