I built an open-source agent skill called Validator. It validates a product or project idea before you build it. It researches the market, finds every competitor, destroys your assumptions, and gives you a clear GO, CONDITIONAL GO, or NO GO verdict backed by sourced evidence.
Here is exactly what it does, why I built it, and how it works.
The Problem
Most people build things nobody wants.
They get an idea. They get excited. They spend three weeks coding it. Then they show it to the world, and nobody cares. Not because the code was bad. Because the problem was already solved, or the audience was wrong, or the moat was zero.
The real issue is that people skip validation. They validate with vibes. They say things like "I think people would want this" or "I could not find a competitor, so there must be none."
That is not validation. That is wishful thinking.
The other problem is that AI tools make it worse, not better. They can build your product faster, so you skip the thinking step entirely and go straight to shipping. You build the wrong thing in three hours instead of three weeks. The speed does not fix the problem. It makes the problem cheaper to repeat.
What Validator Does
Validator is an agent skill. You install it into your AI coding agent. Then you tell the agent:
"I have an idea for X. Validate it."
The agent then runs a 12-phase research process. It does not build anything. It does not write code. It researches.
When it is done, you get a single markdown file with a verdict.
The verdict is one of three things:
- GO — build it. The evidence supports it.
- CONDITIONAL GO — build it only if specific conditions are met. Those conditions are listed. A falsification test is included.
- NO GO — do not build it. The evidence says it will fail.
There is no "it depends." There is no "maybe you could try." There is a decision.
The 12 Phases
Here is what Validator does, step by step:
- Problem validation — does this problem actually exist? Is it painful enough?
- Market research — how big is the market? Is it growing or shrinking?
- Every single competitor — not just the famous ones. Enterprise, mid-size, startup, open source, side project. All of them.
- Head-to-head feature comparison — what does each competitor have? What do they lack?
- Destroying your own assumptions — "Our X feature is unique" gets checked against the market. If it is not unique, the skill says so in writing.
- Gap analysis — where are the real openings? Are they worth pursuing?
- Moat evaluation — what stops someone from copying you in a week? Rated objectively.
- Outcome risk — what happens if you build it and nobody pays?
- Platform policy check — does your product depend on a platform API or ToS that could break or get you banned?
- Technical feasibility — can the core mechanic actually be implemented right now?
- Final decision — GO, CONDITIONAL GO, or NO GO.
- Pivot recommendations — if the idea fails, what adjacent thing might work?
The Rules That Make It Different
Validator has strict rules that most idea validation tools do not enforce.
One audience, forced.
"All businesses" is rejected. "Everyone" is rejected. "Small and medium companies" is rejected. Validator pushes until you pick ONE segment. Because if you build for five different buyers with five different budgets and five different pains, you solve nothing for anyone.
Every competitor, not just the famous ones.
Most people check for three competitors and stop. Validator checks enterprise tools, startups, open source projects, side projects, and anything else that solves the same problem. If a competitor exists, it gets listed with funding, revenue, pricing, employees, business model, and launch year.
Every claim carries a source URL.
No "trust me." No vibes. Every fact in the report has a source link on the line below it. Unverified assumptions are forbidden. If the agent cannot find evidence, it says so.
Data source priority.
When sources conflict, Validator uses a hierarchy: Government data wins over industry reports. Industry reports win over company filings. Company filings win over academic papers. Academic papers win over surveys. Surveys win over official docs. Official docs win over reviews. Reviews win over social media. Social media wins over blogs.
Falsification test for every CONDITIONAL GO.
If the verdict is CONDITIONAL GO, the report must state a specific, observable condition that would flip it to NO GO. Example: "NO-GO if a funded competitor launches a $0 version within 6 months." This is a written exit gate. Not a vibe.
Evidence saturation, not endless browsing.
Validator knows when it is done. It stops searching when results become repetitive, after 3 consecutive empty searches, or when no new high-value evidence appears. It does not waste time browsing forever.
What the Output Looks Like
The final report is a single markdown file called MAIN.md. It contains:
- Market demand metrics with sources
- Competitor tiers (enterprise, mid-size, startup, open source)
- Problem-already-solved analysis
- Head-to-head feature table
- Moat table with objective ratings
- Verdict (GO / CONDITIONAL GO / NO GO)
- Build conditions (if CONDITIONAL GO)
- Falsification test
- Pivot recommendations
- Full source list
The Honest Catch
Validator is not a magic oracle. It is a research process. It forces you to look at evidence instead of feelings. But the quality of the output depends on the quality of the AI agent running it and the search tools available to that agent.
If your agent has bad web search, the research will be shallow. If your agent hallucinates sources, the report will be unreliable. Validator enforces the process, but it cannot guarantee the agent's search results are perfect.
Also, Validator will kill your idea if the evidence says it should die. That is the point. But if you are emotionally attached to your idea, you will hate the output. That is also the point.
How to Install It
Validator is an Agent Skills-format skill. It is a folder with a SKILL.md file. It works in any tool that supports the Agent Skills format.
Clone it:
git clone https://github.com/Sumama-Jameel/validator.git
cd validator
Copy it into your agent's skills directory:
cp -r SKILL.md examples ~/.claude/skills/validator
Or install with the skills CLI:
npx skills add Sumama-Jameel/validator -g
Or install as a Claude Code plugin:
/plugin marketplace add Sumama-Jameel/validator
/plugin install validator@sumama-validator
Then just say:
"I have an idea for X. Validate it."
The skill triggers automatically when you propose a new project or ask whether an idea is worth building.
Who It Is For
Validator is for builders who want to stop wasting time on dead ideas.
It is for solo developers, startup founders, and engineers who build fast but need to check if the direction is right before they commit weeks of work.
It is not for people who want validation in the emotional sense. It will not tell you your idea is nice. It will tell you if it should exist.
It is open source under the MIT license.
GitHub link: https://github.com/Sumama-Jameel/Validator
Top comments (0)