add llms.txt and llm-full.txt docs

[pablofullana]

User story / Problem statement

Currently docs.dappbooster.cc serves only the typedoc HTML reference. A coding agent, or an LLM-backed editor that wants the kit's docs, has to scrape the pages or be pointed at them by hand. M1 asks for agentic-ready documentation, and there is no plain-text entry a model can read.

Expected outcome

docs.dappbooster.cc serves two files a model can read as-is: /llms.txt, a short markdown index of the docs with links, and /llms-full.txt, the whole reference in one file to paste at once. The build writes both, so they cannot fall behind the reference.

Acceptance criteria

• docs:build writes /llms.txt and /llms-full.txt into the typedoc output
• /llms.txt carries the kit blurb and links to getting started, hooks, components, config, and coming-from-wagmi
• /llms-full.txt carries the full reference text, no HTML
• Both answer at docs.dappbooster.cc/llms.txt and /llms-full.txt after a deploy
• The same writer runs under docs:check in a compare mode: it regenerates both files in memory and fails when either is missing or differs from what is committed or built
• Links in /llms.txt are absolute

Technical notes

docs:build runs typedoc into the typedoc/ folder Vercel serves; write both files in that step so they track the reference. The JSDoc already holds the per-symbol text that /llms-full.txt flattens.

Typedoc emits HTML only and has no built-in llms.txt output. Three routes were checked:

• typedoc-plugin-llms-txt (v0.1.2, peer typedoc ^0.28) writes the index file only, so it cannot cover /llms-full.txt.
• typedoc-plugin-markdown (v4.13.0, peer typedoc 0.28.x) emits markdown beside the HTML through the outputs array, but joining it into one file is still ours to write.
• typedoc --json is already in the core and dumps the whole reflection tree, comments included. One script in kit/ reads that and writes both files, which matches kit/docs-check.mjs and kit/check-anatomy.mjs and adds no plugin to track.

Additional context

M1 deliverable: agentic-ready documentation. /llms.txt follows llmstxt.org. /llms-full.txt is not in that spec: it is the common convention for a whole docs site in one file, and the hyphenated spelling is the one to use. Sits with #207 (docs site) and BootNodeDev/dappbooster-canton-landing#14 (landing).

[gabitoesmiapodo]

Closing. Reasons below.

Are the txt files needed at all for LLMs to use what's in this repo?

Mostly no. Two of the three cases are already covered.

An agent working in this repo reads the source and the CLAUDE.md files.

An agent in a consumer project has it too.

The case it does serve: someone with neither, asking a chat window about the kit, or an agent pointed at docs.dappbooster.cc.

What it will not do: No major provider commits to reading llms.txt in production. Google has said it will not support it. A study of 500 million AI bot events found a few hundred requests for the file; the bots crawl HTML. So do not expect it to change whether anything cites the kit.