How to Use an AI “White‑Box” Test to Filter Unviable Startup Ideas

Introduction

Many projects fail not because the team lacks ability, but because they start with the wrong thing. The codex-startup-pressure-test-skill is designed to put a startup idea into a “white‑box” and check whether it can survive early‑stage validation questions.

What problem does it solve?

startup-pressure-test is used before work begins to decide if an idea has real demand, fatal flaws, and a two‑week verifiable minimum path.

The skill examines a concept from an early‑stage perspective, focusing on:

Core assumption : the single most critical thing that must hold for the business.

Fatal flaws : risks most likely to cause failure.

Problem reality : whether the pain is genuine or just “sounds nice”.

Real competition : what users currently use as a substitute.

First 10 customers : where to find the initial users.

Two‑week MVP : the minimal experiment and which features to cut.

The default stance is clear: prove you shouldn’t build first, then see if there’s any reason to continue.

Five best‑fit scenarios

Idea pre‑screen : ask “Is this direction worth pursuing?” – the skill returns Strong / Weak / Pivot required.

Pain validation : ask “Is this a problem users would actually pay to solve?” – the skill returns real pain, early adopters, validation criteria.

Competition mapping : ask “Do we have differentiation?” – the skill returns direct/indirect competitors, switching costs, real enemy.

Cold‑start acquisition : ask “Where are the first customers?” – the skill returns manual acquisition tactics, first‑contact messages, success criteria.

MVP convergence : ask “What should we build in two weeks?” – the skill returns the minimal feature set, what to cut, and a two‑week test plan.

If you only want a quick strength check, use the pressure-test mode; for a full diagnosis, use full.

Installation and basic invocation

npx --yes codex-startup-pressure-test-skill@latest

After installation the skill lives at ~/.codex/skills/startup-pressure-test. Restart Codex and invoke it with $startup-pressure-test in a prompt.

Minimal call

使用 $startup-pressure-test 测试这个创业点子：

Recommended call

使用 $startup-pressure-test 对下面这个创业点子做一次 brutal pressure test：

Input quality matters

The skill fails with overly vague inputs such as “build an AI note‑taking tool” or “create a creator growth platform”. A good input contains four elements:

Idea : a concise description of what you want to build.

Target customer : a narrow, specific persona.

Current behavior : how the persona currently solves the problem.

Target action : the behavior you want them to take (e.g., pay $19/month).

Example of a high‑quality input:

使用 $startup-pressure-test 验证这个点子是否值得做：

点子：一个把本地长视频自动切成带字幕短片的工具。
目标客户：独立开发者和小团队 founder，每周在 X/LinkedIn 发产品 demo。
当前行为：他们用 CapCut 手动剪，耗时 1‑2 小时，很多人因此不发。
目标动作：愿意每月付 19 美元，上传长视频后 10 分钟内得到 3 条可发布短片。
我最担心：用户会觉得剪视频不够痛，仍然手动操作。

Choosing among six modes

pressure-test

: you only want to know if the idea is strong. problem-validation: you doubt the pain is real. competition-map: you need to understand competitors. first-10-customers: you want a manual acquisition plan. mvp-plan: you are ready for a two‑week validation. full: you want a compact combination of the above.

If unsure, use full. For deeper analysis, add “deep / full report” to request expanded assumptions, interview questions, outreach messages, milestones, and pivot options.

Understanding the output

Verdict

The most important part. Possible values:

Strong : continue validation.

Weak : current form is weak; don’t start development.

Pivot required : change the angle before proceeding.

If the verdict is Weak or Pivot required, focus on the “Fatal Flaws” section before building.

Scorecard

Scores across six dimensions: Pain intensity, Buyer clarity, Urgency, Differentiation, Speed to validate, Founder advantage. Low scores indicate where you need more evidence.

Core Assumption

The single sentence that captures the essence of the report, e.g., “Independent developers are willing to pay $19 to save time cutting demo videos.”

Fatal Flaws

Typical fatal flaws include insufficient pain frequency, unclear paying audience, existing solutions being good enough, lack of differentiation, and MVP that cannot test the core hypothesis.

MVP

The MVP is defined as the smallest experiment that tests the most dangerous hypothesis, not a stripped‑down product.

Five ready‑made prompt templates

使用 $startup-pressure-test 用 pressure-test mode 评估这个创业点子：

使用 $startup-pressure-test 用 problem-validation mode 验证这个问题是否真实且足够痛：

使用 $startup-pressure-test 用 competition-map mode 分析这个点子的真实竞争：

使用 $startup-pressure-test 用 first-10-customers mode 制定手工获客计划：

使用 $startup-pressure-test 用 mvp-plan mode 设计 2 周验证计划：

Common misuse and fixes

Misuse 1 : asking the skill to “evaluate my idea” – too vague. Fix: ask for the three most likely failure reasons and quick validation steps.

Misuse 2 : treating “users say they like it” as validation. Fix: ask about past behavior, not hypothetical willingness.

Misuse 3 : building a mini‑product as MVP. Fix: treat MVP as an experiment that validates the riskiest assumption.

Misuse 4 : abandoning a project at the first “Weak” verdict. Fix: interpret Weak as a signal to refine the angle, not to quit outright.

Applying the skill to content teams

Content series, columns, or paid content products can be treated as “startup ideas”. Use the same white‑box logic to avoid self‑indulgent topics.

Content‑topic white‑box template

使用 $startup-pressure-test 测试这个公众号内容产品：

Key outputs for content teams are Problem Reality, Competition, and MVP recommendations.

Troubleshooting

If Codex does not recognize $startup-pressure-test, restart Codex.

Check that the skill directory exists at ~/.codex/skills/startup-pressure-test and contains SKILL.md, agents/, and references/.

Ensure the skill is installed in the correct Codex skills folder.

Final usage advice

Place the skill at the very first step of any project – before writing requirements, not before writing code.

Idea
 ↓
Startup Pressure Test
 ↓
Keep only directions worth validating
 ↓
User interviews / manual acquisition / 2‑week MVP
 ↓
Decide whether to enter full product development

The real efficiency is not “faster completion” but “earlier knowledge that the idea should not be pursued or that the current angle is wrong.”

References

Repository:

https://github.com/Kappaemme-git/codex-startup-pressure-test-skill

README.md – installation and basic usage

Skill definition: startup-pressure-test/SKILL.md Playbooks: startup-pressure-test/references/playbooks.md Codex Skills official docs – Agent Skills – Codex | OpenAI Developers