AI context
The whole documentation as one Markdown file, with the source of every example, for coding agents.
A coding agent writes better table code from this documentation than from what it remembers about data tables in general. So the site also publishes itself as Markdown: one file with every page and every example, an index of the pages, and each page on its own.
Prerequisites
- A coding agent that can read files in your repository or fetch a URL: Claude Code, Cursor, GitHub Copilot, Codex or similar.
@kingtech/react-tablekitinstalled, or about to be. See Installation.
Installation
Save the full file next to your code, where the agent can read it:
curl -o docs/react-tablekit.md https://react-tablekit.vercel.app/llms-full.md
Then tell the agent it is there. One line in the instructions file your tool reads at the start of
every session is enough: CLAUDE.md, AGENTS.md, a rule in .cursor/rules/, or
.github/copilot-instructions.md.
For anything involving data tables, read docs/react-tablekit.md first, and prefer it to what you remember.
The file describes one version of the package. When you upgrade, download it again.
What is in the file
Every page of this documentation in reading order: the getting-started pages, every feature page, the demos, the guides and the whole API reference. Where the site shows a running example, the file has its source; where the site shows a reference table, the file has the table. The examples import a few site modules (sample data, shared columns and the mocked API they fetch from), and those are included once at the end, so every example is complete.
Each page starts with its title and its URL, so an agent can say which page an answer came from. The file is generated from the same Markdown as the site, in the same build, so it cannot say anything the site does not.
Which file to use
| File | What it holds | Use it when |
|---|---|---|
| llms-full.md | Every page and the source of every example | the agent should know the whole library |
| llms-full.txt | The same file, under the name tools look for | a tool asks for a documentation URL |
| llms.txt | An index: one line per page, each linking its Markdown | the agent should fetch only what it needs |
A page's URL + index.md | That one page, with its examples | you are working on a single feature |
The full file is large, but it fits whole in the context window of a current frontier model. If
your tool's window is smaller, give it llms.txt and let it fetch the pages it needs. Any page
works on its own the same way: the sorting page is
/react-tablekit/sorting/ with index.md added.
Minimal working example
With the file in docs/, a request like this has everything it needs to be answered from the
documentation rather than from guesswork:
Using docs/react-tablekit.md, add an orders table to OrdersPage. Load it from GET /api/orders with
server-side paging and sorting, filter by status (open, paid, refunded), and let the user export
the current page as CSV.
Verify
Ask the agent something only the documentation answers: for example, which CSS variable sets the
background of striped rows. An agent reading the file answers --tk-row-striped-bg and can name
the page it came from, Theme tokens. One that is not reading it guesses.
Next steps
- Installation: the package, its peers and the stylesheet.
- Quickstart: a complete first table.
- API reference: generated from the TypeScript declarations.