A Slice Gate That Stops Weekend MVPs From Growing Phantom Features

Description and use

The slice gate limits AI-generated code to a defined envelope, preventing phantom features and keeping the MVP minimal.

Problem

Solo founders often add unintended features when using AI code generators, causing launch delays and bloated code.

Outcome

The slice gate enforces a strict boundary around the MVP, preventing accidental feature creep and ensuring a clean, deployable codebase.

Workflow

• Define slice.json with name, stories, allowed_paths, allowed_env, forbidden strings, max_files, and start_command.
• Create unknowns.md as a checklist to record assumptions and missing features.
• Write the application code only within the allowed paths listed in slice.json.
• Run gate.py before committing to verify the working tree matches the slice constraints.
• If gate passes, start the app with the specified start_command and test the MVP.
• Iterate: update unknowns.md with new guesses, adjust slice.json if necessary, and re-run the gate.
• Deploy the verified slice to a free host.

Evidence

Every coding session is bound to a slice.json that lists allowed files, allowed environment variables, forbidden vendor strings, and the one command that must start the app.

Source

Unknowns go in unknowns.md. The model records guesses instead of silently turning them into features.

Source