Liminal Spec

Liminal Spec

A spec-driven development skill pack for AI coding assistants.

What This Is

Liminal Spec is a set of skills that guide AI agents through a structured development process: define what you're building, specify the requirements in detail, design the implementation, and break it into implementable stories. Each phase produces a document that the next phase reads cold, with no shared conversation and no accumulated assumptions.

The core idea is a traceability chain: every acceptance criterion gets test conditions, every test condition maps to a test, every test drives implementation. When tests pass, you can follow the chain from any test back to the requirement it verifies. You know the build matches what was specified because the chain makes it auditable.

Is This For You?

Liminal Spec works best when you can describe most of what you want to build before you start building it. The sweet spot is a feature or epic that would take a team 1-4 weeks to build traditionally: standard database patterns, known integrations, established UI frameworks, clear user workflows. The kind of work where you can spend a few hours thinking through what users need and what the system should do, and come away with a substantial picture of the requirements.

That describes most enterprise business applications, many startup products, and a lot of hobbyist projects with clear goals.

**Not the right tool for:**

• Quick bug fixes or just a few small file changes. Use your AI coding assistant's plan mode feature.
• Experimental work where you're discovering what works through iteration, like a novel algorithm, an unfamiliar protocol, or something where the hard part is figuring out whether the approach even works. Use small slices of plan mode and execution iterations.
• Situations where you genuinely don't know what you want yet and need to build small pieces to find out

You can also mix approaches. The surrounding system benefits from specs even when one component