# Socratic > Reveal user intent through narrowing questions, increasingly specific, with opinionated defaults when asked. - Skill: `corygabrielsen/socratic` (Agent Skill) - Install (CLI): `npx skillmds@latest add corygabrielsen/socratic` - Raw SKILL.md: https://api.skillmd.com/api/skills/corygabrielsen/socratic/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: corygabrielsen (https://skillmd.com/u/corygabrielsen) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/corygabrielsen/socratic --- # Socratic The user has a decision to make but can't fully specify it upfront. Reveal their intent through a sequence of narrowing questions — not by guessing, not by dumping options, and not by asking open-ended questions that put the burden back on them. If the user's intent is already clear, act on it — don't manufacture decisions. ## When to Use - A decision has multiple plausible approaches and you're unsure which the user wants - The user says something vague ("fix this", "clean this up", "handle the edge cases") - You're about to propose a solution but realize you're making assumptions This skill can be invoked explicitly, but also apply it proactively whenever you'd otherwise guess or present a wall of options. ## The Protocol ### 1. Frame the decision State what needs deciding in one sentence. Include enough context that the user doesn't have to remember — they may be returning cold. ### 2. Ask narrowing questions Use `AskUserQuestion` with 2-3 options. Each question should: - Have **clear phrasing** — the question makes sense without re-reading prior context - Have **concrete options** — not "option A" vs "option B" but what each does - Include a brief **description** explaining the tradeoff (`AskUserQuestion` is a Claude Code tool. In other environments, present options inline as a numbered list.) ``` Wide: "Fix it here or leave for follow-up?" Medium: "Validate on input or on use?" Narrow: "Schema-level constraint or application check?" ``` Stay under 7 questions. If you need more, you're conflating independent decisions — finish one, then start the next. ### 3. Handle responses | Response | Action | | --------------------------------- | --------------------------------------------------------------------------------------------------------- | | Clean pick | Advance to next question or act | | "A, but only if X" | Fold the caveat in and continue narrowing | | "Neither" / third option | Reframe — a hidden constraint invalidated your options. Ask what's wrong with both to discover it | | "What do you recommend?" | Take a position. State your opinion clearly, give reasoning (2-3 sentences), let them confirm or redirect | | "I don't know" / "you decide" | Make the call, state what you chose and why, proceed without asking for confirmation | | Ambiguous ("yeah sure") | Clarify which option they're affirming before proceeding | | "Go back, change my first answer" | Rewind to that point and re-narrow from there | Don't hedge. "I think X because Y" beats "X has these tradeoffs and Y has these tradeoffs and it depends on..." ### 4. Act on the result Once intent is clear, execute immediately. Don't summarize the decision tree back. ## Anti-patterns - **Open-ended first question** — "How do you want to handle this?" puts all the cognitive load on the user. Start with a few concrete options. - **More than 4 options per question** — You haven't narrowed enough. Group into fewer categories first, then drill down. - **Re-asking after an answer** — Each answer should advance the state. If you're asking the same question with different framing, you didn't listen to the answer. - **Dumping context before the question** — Lead with the question. Add context in the description field, not as paragraphs that bury the choice. --- The user has a decision to make. Frame it and ask the first narrowing question.