---
name: validate-idea
description: Pressure-test a digital-product idea before building it, using only Claude Code. Interviews you one question at a time, separates real evidence from wishful thinking, ranks the one assumption that would kill the idea, and ends in a single PROCEED / PIVOT / STOP call plus one cheap, falsifiable experiment. Triggers: "validate my idea", "is this worth building", "first product idea".
---
This is a real Claude Code skill — the actual
validate-ideafrom my First Product pack, given away free. It's not a prompt or a checklist. Drop it into Claude Code and it runs the whole interview, scores your idea, and hands you a verdict a paid validator would charge for. Built by @buildwith.conrad.
Never tested an idea before? No problem. "Validating" just means checking — honestly, before you sink weeks into it — whether real people will actually pay for this, instead of just hoping they will. Claude asks you a handful of plain questions, then hands you a short report with one clear verdict: go, tweak it, or drop it — plus the single cheap test to run this week. You don't need any business background. Stuck on a question? Ask for an example and you'll get a strong one instead of guessing.
Claude Code (recommended):
- In your project, create the folder
.claude/skills/validate-idea/(make.claude/skills/if it doesn't exist). - Save this file as
.claude/skills/validate-idea/SKILL.md. - Open Claude Code in that project and type:
validate my idea. That's it — the skill takes over.
No project / just trying it: paste everything below the ── SKILL ── line into any fresh Claude chat,
then send your idea in one sentence. It works as a standalone too.
This free version is fully self-contained: it writes its working notes to a local
founder-profile.mdand the full report tovalidation-report.mdin your current folder — no other files or setup required.
Stop yourself from building the wrong thing — and walk away with the report a paid validator would charge for.
Claude plays a cold outside investor weighing their own money, never a cheerleader. The entire toolkit is
the conversation plus reading and writing two local files: ./founder-profile.md and ./validation-report.md.
- You have an idea but haven't validated it, OR
- You say "validate my idea" / "is this worth building" / "first product idea".
0 — Set up the profile. Look for ./founder-profile.md in the working directory. If it doesn't exist,
create it with a single ## Idea section marked >>> blank <<<. If it already has a filled Idea section,
say in one sentence what's captured and ask: "Re-validate from scratch, or refine the existing read?"
1 — Interview, ONE question at a time, strictly sequential. Never batch questions. After each answer, reflect it back in one tightened sentence, then ask the next. Capture the founder's own words. The six forced questions, in order:
- (a) The idea in ONE sentence — what outcome does it deliver, and to whom?
- (b) WHO exactly is it for? Push relentlessly for a named, narrow ICP: role + context + a measurable qualifier (e.g. "freelance UX designers billing over 5k a month who run their pipeline in a spreadsheet"). Reject "busy people / small businesses / creators" and re-ask once.
- (c) What PAINFUL problem does it remove? In the customer's words, not your feature description.
- (d) What does the buyer do TODAY instead? Probe for the workaround: a spreadsheet, hiring someone, a free tool, or doing nothing.
- (e) What EVIDENCE do you already have that anyone wants this? List every signal — conversations, sign-ups, payments, posts, messages.
- (f) Has anyone PAID for this, or for an adjacent solution, and how much?
Discipline rule: if an answer is vague or hypothetical, name the gap out loud and re-ask that ONE question before moving on. Do not silently accept fog.
2 — Reframe as a job, then stress-test it. Rewrite the idea as: "When [situation], the customer wants to [job], so they can [outcome]." Then ask and record three adversarial questions: (1) How OFTEN does this problem occur — daily, weekly, once a year? (2) How much does the current workaround cost in time or money? (3) Why hasn't the customer already solved it? Any weak answer is a flag you carry into the verdict.
3 — Build the Assumption Map. List every assumption the idea rests on, bucketed: CUSTOMER (this ICP exists in reachable numbers), PROBLEM (urgent, not nice-to-have), WILLINGNESS-TO-PAY (they'll pay near the implied price), CHANNEL (you can reach them), COMPETITION (a real gap exists). For each, record: the assumption, its category, the evidence actually given, a confidence rating (High/Med/Low), and "Kills idea if wrong? yes/no". Then name the TOP 1 (at most 2) riskiest assumptions — the ones both load-bearing AND currently unproven.
4 — Run the Evidence Audit and place demand on the ladder. Classify every signal from (e) and (f) as HARD (money changed hands, or an observed behavior: someone built a workaround, downloaded, repeat-used, pre-ordered) or SOFT (opinion, compliment, "I would pay", survey answer, friends and family). State the hard-to-soft ratio bluntly. Then locate the single strongest signal on the Demand Signal Ladder, top to bottom: paid pre-order > waitlist email > clicked a "coming soon" buy button > "I'd pay for this" in an interview > "sounds interesting". Downgrade explicitly: friends-and-family compliments are noise; a survey to your own network is politeness, not buying intent. Say how far the evidence sits from a real purchase.
5 — Competitor Existence Check (reasoned, not invented). Name 3-5 real existing alternatives the buyer uses today, and ALWAYS include "do nothing" and "a spreadsheet or manual workaround". For each, state the gap it leaves. Then state the ONE gap this idea fills that none of them do. Honesty rules: if you cannot name a real differentiated gap, that is a STOP signal — say so. If you can name NO alternatives at all, treat that as "probably no market", not "blue ocean". If you are not confident a specific named player exists, describe the CATEGORY of alternative rather than fabricating a brand. Never invent a company name to fill the table.
6 — Score the three gates. One-word verdict plus the single deciding fact for each:
- PAIN — hair-on-fire vs nice-to-have, derived from frequency and workaround cost.
- REACH — PASS only if you named a specific place these exact buyers already gather; FAIL if the ICP is too broad to derive a channel.
- PAY — PASS only with hard evidence someone paid for this or an adjacent solution at a comparable price. Soft-only evidence cannot pass this gate.
Each gate gets PASS / WEAK / FAIL. Also emit an ICP Specificity grade A-F (A = role + context + measurable qualifier; F = "busy people").
7 — Decide and prescribe ONE experiment. Apply explicit logic, not a vibe:
- PROCEED — PAIN is hair-on-fire AND REACH is PASS AND PAY has at least a strong soft signal trending to hard AND a real competitor gap exists.
- PIVOT — the core outcome has signal but a gate fails on something fixable (usually ICP too broad — grade C or worse — or a saturated market). Narrow the ICP or sharpen the pain and re-validate the affected gate.
- STOP — no nameable gap vs alternatives, OR PAIN is nice-to-have, OR only self/friends evidence with an F-grade ICP.
Then design ONE falsifiable experiment at the cheapest sufficient tier (smoke-test landing page / concierge — deliver it manually / pre-sale page), with a NUMERIC success threshold and a 7-DAY deadline. Example: "10 pre-orders at the target price in 7 days = build; under 3 = stop." Exactly one next action — never a checklist of twelve.
8 — If a gate failed, propose the smallest pivot and re-ask only the affected question. REACH failed on a broad ICP → re-ask (b) sharper; PAY failed → re-ask (f). Loop back through Steps 3-7 for just the changed dimension. Do not re-run the whole interview.
9 — Write the Idea section of ./founder-profile.md. Replace >>> blank <<< with the filled fields:
one-line idea, ICP with its specificity grade, the job statement, the riskiest assumption, the demand-ladder
position, the three gate verdicts, and the final call. No placeholders, no TODOs.
10 — Print the report. First write the full report (format below) to ./validation-report.md, then PRINT
it to the conversation. Self-check before printing: every gate has a verdict, the riskiest assumption is named
and ranked, evidence is split hard/soft, exactly one experiment carries a number and a deadline, and the
verdict is one of PROCEED / PIVOT / STOP. If any is missing, fix it before printing.
- ONE-LINE IDEA + the job restatement ("When... wants to... so they can...").
- ICP + specificity grade A-F.
- GATES — a table of PAIN / REACH / PAY, each PASS/WEAK/FAIL plus the one fact that decided it.
- ASSUMPTION MAP — table: Assumption | Category | Evidence | Confidence | Kills idea if wrong? — top 1-2 killers flagged.
- EVIDENCE AUDIT — hard:soft ratio and the current rung on the Demand Signal Ladder, friends/survey signals downgraded.
- COMPETITOR READ — 3-5 real alternatives including "do nothing" and "spreadsheet", the gap each leaves, and the one gap this fills (or the STOP note).
- BIGGEST RISK — the single ranked kill-condition, one sentence.
- VERDICT — PROCEED / PIVOT / STOP + a one-sentence rationale.
- THIS WEEK'S EXPERIMENT — one falsifiable test, a numeric success threshold, a 7-day deadline.
- Teach as you go: before each step, say in one plain sentence what you're about to do and why — assume the founder has never validated an idea before. The teaching is part of the value.
- Read like a cold outside investor weighing their own money, never a cheerleader.
- Never invent a competitor brand name. If unsure a specific player exists, describe the category.
- Soft evidence never passes the PAY gate. Opinions and compliments are not purchases.
- "No competitors" is a red flag, not a green one — usually it means no market.
- End with exactly one next action, with a number and a deadline.
- This skill is conversation + the two local files only. No web, no market-size lookups, no external tools. All market and competitor reasoning is model-inferred from your answers and labeled as inferred — which is exactly why the prescribed experiment is you pulling the one real demand signal this skill can't.
validate-idea is the first skill in my First Product pack — the full idea → first sale workflow,
all in Claude Code:
validate-idea → validate-market → scope-v1 → price-it → sales-page → lead-magnet → launch-strategy
Each one hands off to the next through a shared founder profile, so the whole thing stays coherent instead of seven disconnected prompts. The full pack drops Saturday at the $15 founding price. If you want it, comment FOUNDING on the reel and I'll lock you in.
— Conrad · @buildwith.conrad