# Decide > Break a stuck decision into options, criteria that reflect the user's actual priorities, and a clear recommendation with reasoning. Use whenever the user types /decide, says they are stuck, torn, going back and forth, can't choose, or "have been putting this off", or lays out two or more options and asks which to pick — career moves, pricing, tools, hires, projects, offers. Also use when a user keeps re-litigating the same choice across a conversation. - Skill: `rohankhullar24-oss/decide` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add rohankhullar24-oss/decide` - Raw SKILL.md: https://api.skillmd.com/api/skills/rohankhullar24-oss/decide/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: rohankhullar24-oss (https://skillmd.com/u/rohankhullar24-oss) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/rohankhullar24-oss/decide --- # Decide Turn a fuzzy dilemma into a structured choice, then actually take a position. The user is stuck because the options are entangled in their head — the job is to disentangle them and have the spine to recommend one. ## Procedure ### Step 1 — Pin down the real decision Restate the decision in one sentence and get it confirmed. Watch for the classic trap: the stated decision ("which CRM should I buy?") often hides the real one ("do I need a CRM at all?"). If you spot a hidden upstream decision, surface it before scoring anything. ### Step 2 — Gather only what's missing Ask at most 3 questions, and only ones whose answers would change the recommendation. Almost always the right three are drawn from: - What are you actually optimizing for right now — money, time, growth, peace of mind? - What's the real deadline, and what happens if you decide nothing? - Is there a constraint you've already accepted that rules something out (budget, location, a person)? If the user has already given enough context, skip questions entirely and proceed. Do not interrogate someone who came ready. ### Step 3 — Lay out the options honestly - List every live option, and always include the one people forget: **do nothing / keep the status quo**, with its true cost. - If the user gave two options, check for a third they've dismissed too quickly or a hybrid ("take the job AND negotiate a later start"). - Kill zombie options explicitly: if one option is dominated (worse on every criterion), say so and remove it rather than politely scoring it. ### Step 4 — Score against THEIR criteria - Derive 3-5 criteria from what the user said matters — not a generic list. Weight them (e.g. 40/30/20/10) and show the weights so the user can object. - Score each option per criterion in plain language, then roughly overall. Keep numbers coarse (1-5). Precision theater ("Option A: 7.35") is dishonest — the inputs are gut feelings. - The table is a thinking tool, not the verdict. If the weighted winner contradicts obvious judgment, say so and explain which weight is probably wrong. ### Step 5 — Apply the three tests Run these and report what each one says: 1. **Reversibility.** Is this a one-way door or a two-way door? Two-way doors (reversible) deserve a fast decision and a bias toward action. One-way doors deserve the full analysis. 2. **The regret test.** Which option would the user most regret NOT trying in 5 years? (For personal decisions this often outweighs the scores.) 3. **The friend test.** If a friend described this exact situation, what would the user tell them in ten seconds? People usually already know. ### Step 6 — Recommend Take a position. Name one option, give the 2-3 reasons that actually drive it, and state: - **Confidence:** high / moderate / low - **What would change my mind:** the single piece of information that would flip the recommendation - **First step:** the smallest concrete action to execute the decision within 48 hours (stuck decisions die without a first step) ## Output format Keep the whole response scannable: a short options list, a compact criteria table, the three tests in one line each, then the recommendation block. No walls of prose. ## Boundaries - For decisions with legal, medical, or major financial consequences, still run the full framework — it is genuinely useful — but recommend the relevant professional as part of the first step, and present the output as structured thinking, not professional advice. - If the user is deciding for someone else ("should my cofounder..."), redirect: the framework only works on decisions the user owns. - If after Step 2 the options are genuinely tied AND the door is two-way, say the most honest thing: "this decision doesn't deserve more analysis — pick A and calendar a review in 30 days." Deciding fast is sometimes the recommendation.