Handling "agent friendly" websites

[skatkov]

I'm curious, lightpanda browser is advertised as "ai agent" browser. And some websites try to signal "agent" friendly content.

Some projects do it through the llms.txt file. Some do it through Accept: text/markdown, see: https://acceptmarkdown.com/

But some other pages might have alternative links for markdown format, like my own blog:

<link rel="alternate" type="text/markdown" href="https://www.skatkov.com/posts/2026-07-15-snooping-on-githits.md">

In my case, .md url provide a stripped down content, without all the clutter.

I've been playing around, and it doesn't seem that LightPanda prioritizes or even uses these?

Other agents are doing that (not all, though):
https://acceptmarkdown.com/status

Browser doesn't send "Accept: text/markdown" either.

[arrufat]

Thanks! Three separate things here — this is my understanding, curious what @krichprollsch thinks:

• Accept: text/markdown on navigations — no: we deliberately send Chrome's Accept header, and if a server returned markdown for the main document there'd be no DOM, which everything (CDP, selectors, agent tools) depends on. It does make sense in fetch --format markdown, where we could send it and dump the body raw if honored.
• <link rel="alternate" type="text/markdown"> — yes, we'd like this in the markdown read paths (fetch --format markdown, MCP markdown tool). Author-curated markdown beats our DOM→markdown conversion. Since the alternate can diverge from the rendered page, we'd surface or opt-in follow it rather than silently swap sources.
• llms.txt — out of scope for the browser; site-level discovery is an agent-layer concern at most.