# Socratic Check > Pre-decision Socratic checklist — 7 prompts that force the user (and Claude) to challenge a position before committing. Anti-vending-machine habit: bring the thinking, let the prompts attack the premises. Activates when the user says "socratic check", "pre-decision", "before I commit", "stress my reasoning", or presents a decision and asks to pressure-test their own logic (not the idea itself — use devils-advocate for idea attacks). - Skill: `ryankolean/socratic-check` (Agent Skill) - Install (CLI): `npx skillmds@latest add ryankolean/socratic-check` - Raw SKILL.md: https://api.skillmd.com/api/skills/ryankolean/socratic-check/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: ryankolean (https://skillmd.com/u/ryankolean) - Updated: 2026-09-21 - Page: https://skillmd.com/skills/ryankolean/socratic-check --- # Socratic Check A structured 7-prompt pre-decision checklist. Ryan brings a position; each prompt challenges one layer of the reasoning. The output is not a verdict — it is a sharper version of Ryan's own thinking with the weak joints exposed. ## When to Activate **Manual triggers:** - "Socratic check" - "Run the 7 prompts" - "Pre-decision check" - "Before I commit" - "Stress my reasoning" - "Am I thinking about this right?" **Auto-detect:** - User states a decision they are about to make (not an idea to explore) and asks whether they are reasoning about it correctly. - User says "I'm about to X" and signals hesitation. **Do NOT activate when:** - User wants the idea itself attacked → use `devils-advocate`. - User wants requirements discovery → use `interrogate`. - User wants to learn a concept → use `mentor-feynman`. ## The Anti-Pattern This Fixes Vending-machine usage: user asks, Claude answers, user ships. Zero learning, outsourced judgment, unchecked assumptions. Socratic prompting reverses the polarity — the user states a position, Claude attacks premises, the user rebuilds the position stronger or walks away from it. ## The 7 Prompts Run one at a time. Wait for the user's answer before moving to the next. Capture the user's answers verbatim — do not paraphrase. The point is to hear the user's own reasoning, not Claude's summary of it. ### 1. Challenge assumptions > "What are you assuming is true here that you have not verified? For each > assumption, what is the cheapest way to check it before committing?" Why: most failed decisions trace to an unexamined assumption treated as fact. ### 2. Test evidence > "What is the strongest piece of evidence supporting this decision, and > what is the strongest piece against it? If the 'against' evidence is weak > or absent, why is that — have you actually looked?" Why: confirmation bias. Most users can cite pro-evidence and go silent on con. ### 3. Premortem failure analysis > "Fast-forward 6 months. This decision has failed badly. Write the post-mortem: > what went wrong, what we missed, what signal we ignored. Be specific." Why: anticipating failure modes is easier in the future tense than the present. ### 4. Counter-argument strength > "Steelman the opposing position. Not a strawman — the strongest version of > the argument for doing the opposite. Does your position still hold against > that version?" Why: if you can't argue against yourself, you don't understand your own position. ### 5. Bias detection > "What bias might be driving this choice — sunk cost, novelty, recency, > social proof, motivated reasoning? If you stripped the bias out, would you > still make the same call?" Why: biases are invisible until named. Forcing the name surfaces the pull. ### 6. Framing challenge > "How is this decision framed? Who framed it that way? What frame would a > competitor, a skeptic, or someone five years from now use? Does the choice > change under a different frame?" Why: the frame often decides the answer before the analysis begins. ### 7. Consequence evaluation > "What are the second and third-order consequences — not just the direct > outcome but what it enables, forecloses, or signals? What becomes harder > after this? What becomes impossible?" Why: first-order consequences are obvious. Second-order is where the real cost or leverage lives. ## Output Format After all 7 answers are captured, produce: 1. **Position (restated)** — one sentence, Ryan's own words. 2. **Strongest joint** — the part of the reasoning that held up under every prompt. 3. **Weakest joint** — the prompt where the reasoning wobbled most, and what needs to be shored up before committing. 4. **Open verification** — the cheapest check from prompt 1 that is still pending, if any. 5. **Recommendation** — commit / shore up then commit / kill / park. Keep the output to under 150 words. The work is in the 7 prompts; the summary is just the scoreboard. ## Modes - **Fast** (default for Ryan, personal mode) — present all 7 prompts at once as a numbered list, let the user answer inline. Skip per-prompt explainers. - **Guided** (candidate / customer mode) — one prompt at a time, with the "why" line shown. Wait for answer. Move on. ## Cross-skill integration - If prompt 3 (premortem) surfaces a real failure mode, offer to escalate to `devils-advocate` for a full adversarial pass on that mode. - If prompt 1 (assumptions) surfaces an unverified assumption that requires external research, offer to run a web search or pull internal docs. - If the user tries to run this on a vague idea rather than a specific decision, redirect to `interrogate` first to pin down what is actually being decided. ## Source @dailyprompter Instagram post 2026-03-21 — the framing (7 prompts that challenge thinking) was public; the actual 7 prompts were gated behind an engagement wall. Reconstructed from the principle plus related literature: Edward Y. Chang's Socratic Method paper (arxiv 2303.08769), the intellectual-sparring-partner template, and the "10 Prompts That Force AI to Challenge Your Worst Ideas" toolkit. Card: `~/mindspace/library/summit/try/2026-04-17-1146-socratic-prompting-7-prompts.md`.