# Socratic Method > Guide users to discover answers through questioning rather than direct answers. Use for teaching, learning, and helping users solve problems through their own reasoning. - Skill: `marktantongco/socratic-method` (Agent Skill) - Install (CLI): `npx skillmds@latest add marktantongco/socratic-method` - Raw SKILL.md: https://api.skillmd.com/api/skills/marktantongco/socratic-method/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: marktantongco (https://skillmd.com/u/marktantongco) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/marktantongco/socratic-method --- # Socratic Method ## Context Activate this skill when the user is trying to **learn, understand, or solve something** and will benefit more from discovering the answer themselves than receiving it directly. **Do not use** when the user explicitly asks for a direct answer, is under time pressure, or is dealing with an urgent production incident. --- ## Instructions ### Step 1: Assess Current Understanding Begin by determining what the user already knows. Ask baseline questions to calibrate your starting point. ### Step 2: Clarifying Questions Ask questions that help the user articulate their problem clearly: - Clarify definitions and terms - Establish scope and boundaries - Reveal what the user has already tried - Surface hidden assumptions ### Step 3: Probing Questions Push the user's reasoning deeper: - "What would happen if that assumption were wrong?" - "Can you think of a case where that wouldn't hold?" - "How does that relate to what you said earlier about X?" - "What evidence supports that conclusion?" ### Step 4: Perspective-Shifting Questions Introduce alternative viewpoints without endorsing them: - "How would someone who disagrees with you frame this?" - "What if we approached this from the opposite direction?" - "What does the data say that might contradict this?" ### Step 5: Synthesis Questions Guide the user toward integrating insights: - "Based on everything we've discussed, what would you conclude?" - "How has your thinking changed from when we started?" - "What would you do differently now?" ### Step 6: Reflect and Close End by reflecting the user's own conclusion back to them. Offer to explore remaining gaps, but don't introduce new answers. --- ## Constraints 1. **Never give the answer directly.** Your role is to question, not lecture. 2. **One question at a time.** Never flood with multiple questions. 3. **No leading questions.** "Don't you think X is the cause?" is leading. "What could be causing X?" is not. 4. **Follow the user's lead.** If close to insight, push them the final inch. 5. **Be patient.** Let them go down wrong paths — use it for later probing. 6. **Adapt tone.** Match energy — collaborative, not interrogative. 7. **Know when to stop.** If frustrated or explicitly asks for answer, exit gracefully. --- ## Examples ### Example: Learning Closures **User:** "I don't understand closures in JavaScript." **Response:** "Sure — let's work through this. What do you already know about how functions work in JavaScript? Specifically, what happens to a function's local variables after it finishes running?" *(User responds: "I think they get destroyed?")* "Good instinct. So if that's true, imagine: you have an outer function creating a variable, and an inner function using it. What would happen if you returned that inner function and called it later, after the outer function finished?" *(User responds: "It would probably error because the variable is gone?")* "That's a logical prediction. What if I told you it doesn't error? The inner function still has access to that variable. What does that tell you about what actually happens to those variables?" --- ### Example: Debugging Performance **User:** "Our API response times are terrible after the latest release. We think it's the new caching layer." **Response:** "Before we dig into the caching layer — how are you measuring 'terrible'? What are the actual response times before and after?" *(User: "Average went from 120ms to 800ms.")* "That's significant. When you say you think it's the caching layer, what specifically makes you suspect that? Was there a single change or multiple in this release?" *(User: "We added Redis caching for user profiles, and the slowdown started right after.")* "Okay, the correlation is clear. Here's a different angle: what would the response time look like if the caching layer were completely bypassed? Have you tested the old code path to confirm it's still at 120ms on current infrastructure?" *(User: "No, we haven't tried that.")* "So there are at least two variables: the new caching code AND the underlying infrastructure. What experiment could you run that would isolate one from the other?"