Conversation
Add a Hugo LLMS output format that renders /llms.txt from the homepage package front matter, plus curated links to other major scverse packages and project resources. Also fix mudata's dead docs URL. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
✅ Deploy Preview for jade-cajeta-1bcca0 ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
There was a problem hiding this comment.
Currently the core package list depends on code (I guess javascript of similar) execution to get rendered in the txt file. Do we know if an agent will see the rendered list at runtime, or just the:
Foundational packages maintained by the scverse core team.
{{ range .Params.packages }}
- [{{ .name }}]({{ .url }}): {{ .description }}
{{- end }}
This is important because if it does not render properly the agent will be quite lost.
There was a problem hiding this comment.
Hmm, I guess agents that look at the code will see this. Only if they access the rendered website, they'll see what we expect. So hardcoding this would be the right thing to do? Or will agents simply attempt to look at these parameters to figure out what should be there? Sure - less efficient but then we don't need to hardcode things.
There was a problem hiding this comment.
So I've seen that it depends a lot on the agent capability. High-end models (such as opus or fable) they will go through the trouble of jumping to other pages and following links if they are provided, while lower-end models (haiku, sonnet) they'll "speed-run" and not bother visiting other pages, so I would say it is safer to hardcode this just in case.
|
Please see also #200 where the plan was to replace the manual list of core packages with information from the package registry as the single point of truth. For agents, we could just point them at this json which contains all core and ecosystem packages: https://scverse.org/ecosystem-packages/packages.json |
Zethson
changed the title
Three things had gone stale in the month this sat open. The core package loop read .Params.packages, which main deleted when it moved core packages to the ecosystem registry; the section would have rendered empty. It reads the registry now, the same way the packages page does, so it covers all twelve rather than the eight that were listed by hand. The outputs config collided with site search: this branch set home to HTML, RSS and LLMS, main had set it to html, rss and entities. Taking either side alone drops an output — llms.txt or the search index. Both are listed now. The mudata documentation URL fix is already upstream in the registry, and the front matter it patched is gone, so those hunks drop.
✅ Deploy Preview for jade-cajeta-1bcca0 ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
The section was hand-maintained with a note that new packages need a manual line. It now comes from registry entries whose project_home is in the scverse organisation, minus the core packages listed above it and minus core-infrastructure, which is the org's own repos rather than packages. No package name is written by hand in the template any more. What is left is site pages, the forums and the YouTube channel, none of which have a registry to read. This shrinks the section from eleven entries to two, because eight of the eleven are absent from the ecosystem registry: gget, spatialdata-io, napari-spatialdata, spatialdata-plot, annbatch, pytometry, rustar-aligner and anndata-rs. They are missing from the package index and from site search too, so the hand-written list was hiding that rather than fixing it. Adding them to the registry restores them here automatically.
3 participants
Fixes #216. Adds a generated
/llms.txtgiving AI agents a curated map of scverse — core packages come from the ecosystem registry, so they stay in sync, while the## More packageslist of other scverse-org repos is hand-maintained.