claude4spec is a planning layer for software you build with AI coding agents. Instead of planning ad-hoc and letting the agent generate against a vague idea, you structure the whole system first — endpoints, DTOs, database tables, UI views, and acceptance criteria — as a queryable, version-controlled spec that lives in your repo.
The plan is the source your agent reads from. You write it once; the agent queries it instead of you re-explaining the system in every prompt. When you change a DTO, you can see every endpoint that uses it before you touch the code.
What it is not
- Not a wiki. Notion and Confluence hold prose; claude4spec holds structured entities with typed relations you can query.
- Not an IDE or a code generator. It does not write code, run tests, or open a terminal. It is the plan; your agent writes the code against it.
- Not a chat transcript. The spec is a deliberate plan of the system, not a log of past conversations.
Your working spec stays local — plain markdown plus a SQLite database inside your repo, so it diffs in git and works offline. Publishing a snapshot to the hub is opt-in.