Typed Natural Language FAQs

Question 1

What is Typed Natural Language (TNL)?

Accepted Answer

TNL establishes structured, persistent English contracts for AI coding agents. It prevents design drift and context loss by creating per-feature specifications that are approved by you, saved on disk, and re-read across multiple AI agent sessions.

Question 2

What are the core components of a TNL contract?

Accepted Answer

A TNL contract consists of seven fixed fields: `id`/`title`, `scope`, `paths`, `surfaces`, `behaviors` (MUST/SHOULD/MAY clauses), `non-goals`, and `rationale`. These fields precisely define the feature's design, implementation requirements, and constraints.

Question 3

Does TNL improve the accuracy and reliability of AI coding agents?

Accepted Answer

Yes, controlled A/B tests showed TNL significantly improved AI agent accuracy by +4 to +6 scenarios across different models compared to baseline prose instructions. Its structured contracts, persistence, and enforcement mechanisms lead to more reliable code generation and fewer rework cycles.

Question 4

How does TNL ensure design consistency and context retention for AI agents?

Accepted Answer

TNL utilizes a fixed, seven-field schema for contracts, persistently stored on disk. This structure, combined with machine-checkable verification of clauses and a PreToolUse hook to re-inject the contract mid-edit, prevents agents from silently drifting and retains crucial design context.

Question 5

Can I integrate Typed Natural Language (TNL) with my existing AI coding agent?

Accepted Answer

Yes, TNL is designed for seamless integration. It provides first-class `tnl init` support for agents like Claude Code, Codex, and Gemini. For other agents, a minimal manual setup involving a Markdown instruction file and a `tnl/` directory is all that's required.

Typed Natural Language

Typed Natural Language

主要功能

使用案例

主要功能

使用案例