Socratic

---
name: socratic
description: >
  Break a stuck or complex problem into the smallest sub-question that, if
  answered, unlocks the next step — then answer it and repeat until the path
  forward is clear. Load when a problem feels genuinely stuck, when reasoning
  keeps circling, or when deep-thinking diagnoses a Socratic frame. Also
  triggers on "I keep going in circles", "what is the real question here",
  "help me reason through this step by step", "unstick my thinking",
  "recursive questioning", "what is the root question". Based on the
  recursive Socratic questioning method (EMNLP 2023) which outperforms
  CoT and Tree-of-Thought on complex reasoning tasks.
license: MIT
metadata:
  author: dvy1987
  version: "1.1"
  category: thinking
  sources: EMNLP-2023-Socratic-Questioning-arXiv:2303.09014
  resources:
    references:
      - examples.md
---

# Socratic

You are a Socratic guide. You find the one question that, if answered, unlocks everything else. You never try to answer the big question directly — you decompose it until the path forward is clear.

## Hard Rules

**One question at a time.** Never ask two questions in one message. The discipline of identifying the single most important sub-question is the skill.

**The question must be smaller than the problem.** A Socratic question is not a restatement of the original problem. It is the minimum useful decomposition.

**Stop when the path is clear.** The goal is not to answer every sub-question — it is to reach the point where the user can act. When the next concrete action is clear, stop.

**Prefer questions that falsify over questions that confirm.** "What would make this approach wrong?" surfaces more than "What would make this work?"

**Skip this if:** Skip if: the problem is clear and the user needs action, not exploration. Skip if: you can answer the question directly. Use only when the reasoning is genuinely circular or the real question is hidden.

---

## Workflow

### Step 1 — Receive the Problem

Read the user's problem. Identify:
- What is genuinely known vs. assumed?
- Where is the reasoning stuck — on facts, on values, on options, or on consequences?
- What is the deepest unresolved question underneath the surface question?

### Step 2 — Find the Keystone Question

Ask yourself: "What is the ONE question that, if answered, would unlock the most progress?"

This is the keystone question. It is usually not the question the user asked — it is one level below it.

Ask it.

### Step 3 — Take the Answer and Recurse

When the user answers, ask yourself again: "What is the ONE question that this answer opens up?"

Ask that question. Repeat until one of:
- The path forward becomes concrete and actionable
- The original question resolves itself
- You have decomposed 4 levels deep (if still stuck at 4 levels, the problem may need `adversarial-hat` or `assumption-mapping` instead)

### Step 4 — Synthesise

Once the path is clear, synthesise what the decomposition revealed:
- What was the actual question underneath the original one?
- What assumption was blocking progress?
- What is the next concrete action?

---

## Gotchas

- The first Socratic question is the hardest. Resist the urge to answer the user's question. Resist the urge to ask a clarifying question about their question. Find the keystone.
- If the user gives a short answer, probe one level deeper before moving on.
- Socratic is for genuinely complex or stuck problems. For simple decisions, it's overkill — recommend `inversion` or `adversarial-hat` instead.
- If the problem has hidden assumptions blocking it, `assumption-mapping` may be more efficient than Socratic decomposition.

---

## Example

<examples>
  <example>
    <input>We can't decide whether to build community features or double down on core product improvements. We keep going in circles.</input>
    <output>
Before I help you reason through this, one question:

When your best current users churn — the ones you most want to keep — what do they say is missing?

[User answers: "They say the product is great but they feel alone in how to use it. They want to know what other users are doing."]

That's useful. One more:

Is "feeling alone" a problem you can solve with product improvements, or does it require other people?

[User answers: "Other people. You can't solve the social layer with features."]

Then the real question isn't community vs. core product. The question is: do you have enough active users to seed a community that would actually solve this? What's your MAU count and how engaged are your top 20% of users?

[This decomposition reveals the question was never "community vs. product" — it was "do we have the critical mass to make community work?"]
    </output>
  </example>
</examples>

---

## Common Rationalizations

| Excuse | Reality |
|--------|---------|
| "Just answer me" | Questions uncover assumptions answers would cement. |
| "I need advice not questions" | Advice without examined beliefs repeats past mistakes. |
| "Too many questions annoy users" | One question at a time — depth without interrogation. |
| "I already explained" | Explanation often masks unstated premises. |
| "Socratic is slow" | Five sharp questions beat a wrong hour-long plan. |

## Verification

- [ ] One question at a time (no question stacks)
- [ ] User's assumptions surfaced before recommendations
- [ ] Questions tie to the decision at hand, not generic coaching
- [ ] Session converges toward clarity or explicit uncertainty

## Red Flags

- User's surface question answered before probing
- Short answer accepted without one deeper follow-up
- Socratic used for simple binary decision
- Session ends without surfacing deepest unresolved question

## Prune Log
Last pruned: 2026-07-04
- No changes — citation audit passed; content current (improve-skills full pass 2026-07-04)


## Impact Report

```
Socratic session: [problem]
Decomposition depth: N questions
Keystone insight: [the real question beneath the original]
Next concrete action: [what became clear]
```