A Claude Code skill for deciding whether a product direction is worth building.
Search engines index supply, not demand. Every result you get back is proof somebody already built the thing and wrote about it. So "I can find it with one search" and "the ground is taken" are the same sentence. A need nobody has named yet returns nothing, which is also why you cannot search your way to one.
git clone https://github.com/mifanr/demand-recon.git ~/.claude/skills/demand-reconThen describe a direction you are considering.
Six stages, escalating in cost. Stages 0 through 3 are free and take minutes. Stage 4 is a parallel investigation costing three orders of magnitude more. A document puts those on the same page and people skim documents; a skill enforces the order and stops on a veto.
Stage 0 asks one question before any question about the direction itself:
How many people can you reach directly right now, and who are they?
Every other question is a property of the problem, so you can dodge it by picking a different problem. This one is not, so it runs first. If a need has no name, nobody can search their way to you either, and owned distribution is the only remaining way to be found. It is the premise restated from the other side. It is not the last filter on an idea — it ranks with "what to build" and gets designed alongside it.
Stage 1 is then four questions about the position:
- Can you find it with one search?
- Is the core mechanism one API endpoint belonging to someone else?
- Does switching take under ten minutes?
- Does the free tier cover 80% of what users need?
One more idea does the heavy lifting in the later stages: the value of a public signal equals the delay between it becoming public and being read. A platform changelog is hours, a job posting is weeks, a competitor's two-star reviews are long, and an audience who knows you is not a record at all, so its delay is infinite. Which is why subscribing to changelogs buys you no lead. It buys you arriving at the same moment as everyone else.
One due diligence report on a failed micro-SaaS. Six months live, $310 in monthly subscription revenue, 39 shipped updates, platform review passed, zero of six core keywords on page one. It went up for sale and cut the asking price five times in eight days.
Every individual step was fine. Market data was right about the demand and said nothing about whether anyone could reach him, so every check he ran came back green.
Sample size is one. A checklist, not a law.
SKILL.md is Chinese and canonical. en/SKILL.md is the English translation. MIT.