Docs Your Agent Can Read
Coding agents know Kirby well. They know these plugins less well, and it shows in a way you might have run into: a config block that looks right – right nesting, right keys, one option that doesn't exist.
Everything needed to get it right is on this site. It just wasn't in a form an agent could read, so every page is now published twice – once rendered for you, once as plain Markdown for agents.
Append .md to Any Page
Every documentation page, blog post, changelog and license page has a Markdown twin. Same URL, .md on the end:
curl https://kirby.tools/docs/content-translator/configuration/global.md
No navigation, no components, no markup to work around.
Documentation pages and blog posts also carry a Copy page button. One click puts the Markdown on your clipboard for the next chat; the dropdown beside it opens the page in ChatGPT or Claude with the prompt already written.
An Index, Not a Dump
Your agent's context is better spent on your code than on my documentation. llms.txt is an index: every plugin, every page, one line each. The agent reads that, then fetches the two pages the task needs.
If you would rather hand over everything at once, llms-full.txt is there too.
One Skill per Plugin
The index helps once an agent decides to look something up. A skill loads on its own, whenever the task matches.
Each plugin now ships one: what it does, what it deliberately does not do, and the handful of options that decide the outcome. That Copilot's inline suggestions need the copilot-suggestions mark as soon as a writer field defines its own marks, or Kirby filters them out and nothing appears. That Content Translator falls back to DeepL when strategy is left out, whether or not DeepL is what the project intended.
npx skills add https://kirby.tools
That installs one skill per plugin into whichever agent your project uses. Each of them links to the Markdown pages behind it, so the agent can read the full reference when the summary isn't enough.
Writing them was worth it for the documentation alone. Compressing a plugin into one page of instructions shows you which decisions you never actually wrote down – a few pages here read better now because of it.
Point Your Agent at It
Whichever fits how you work: append .md while you read, hand over llms.txt at the start of a task, or install the skills once and forget about them.
If your agent still gets something wrong, that is worth an issue – it usually means the page it read was unclear.