Startup Idea Machine

A Codex skill that asks three focused questions, searches current public signals, and starts firing startup ideas shaped around the founder.

It is deliberately interactive. The first idea burst is broad. The founder marks ideas as love, maybe, or no. Codex learns from those reactions, generates a sharper second round, validates the finalists, and saves the result as a native Markdown report that opens directly in Codex.

What it does

• Asks only three useful questions before starting
• Searches live public sources instead of inventing demand
• Generates a first burst of 12 concise startup ideas
• Supports open-web, niche, boring-business, new-tech, unbundle, GitHub-demand, and manual-work hunts
• Learns from love, maybe, and no reactions
• Produces a second round tailored to the founder's taste
• Checks competitors, workarounds, spending signals, reachability, and timing
• Shows disconfirming evidence and unknowns
• Scores selected finalists with a consistent framework
• Creates a session file and a native Markdown decision report

The interaction

Codex begins with three questions:

1. What are you unusually good at, curious about, or experienced in?
2. What customers or business styles attract you, and what should be avoided?
3. What constraints matter: time, budget, technical ability, team, and ambition?

It then returns 12 rapid idea cards. Reply with reactions such as:

Love: 2, 7
Maybe: 4, 9
No: 1, 3, 5

Codex uses those reactions to create a more precise second burst.

Install

npx --yes startup-idea-machine@latest

The command installs the generate-startup-ideas skill for Codex.

Example prompts

Use $generate-startup-ideas and find startup ideas for me.

Use $generate-startup-ideas in boring-business mode. Ask me the three questions, then start the first idea burst.

Use $generate-startup-ideas to find opportunities enabled by recent AI tools and APIs. Keep the first round fast and varied.

Use $generate-startup-ideas to refine the ideas I liked in the previous round and validate the best three.

Outputs

The skill can save:

• startup-idea-session.json — profile, sources, rounds, reactions, and validation state
• startup-idea-machine-report.md — finalists, evidence, scoring, winner, risks, and a seven-day validation plan

All generated content is in English. Public complaints and trends are treated as signals, not proof of demand or guaranteed business opportunities.

License

MIT