Every page advertises llms.txt and llms-full.txt, but they are only generated for projects with an API reference — docs-only sites ship two dead links

Summary

The AI / Agents margin section appends links to llms.txt and llms-full.txt unconditionally. _generate_llms_txt and _generate_llms_full_txt both return early when _quarto.yml has no api-reference key.

For any project without an API reference — a pure documentation site — the files are never written and the links 404.

Root cause

core.py, building the margin section. The Skills item is gated; the two below it are not:

if self._config.skill_enabled:
    ...
    ai_items.append(f'<a href="skills.html">Skills{_sparkle_svg}</a><br>')
ai_items.append("[llms.txt](llms.txt)<br>")
ai_items.append("[llms-full.txt](llms-full.txt)<br>")

margin_sections.append(f"\n#### {get_translation('ai_agents', lang)}\n")

core.py, _generate_llms_txt (and identically _generate_llms_full_txt):

quarto_yml = self.project_path / "_quarto.yml"
if not quarto_yml.exists():
    return
...
if "api-reference" not in config:
    return

The generator is called unconditionally during the build and the step is logged as Created llms.txt + llms-full.txt, so the build reports success while having written nothing.

Minimal repro

Any great-docs.yml with sections: and no api-reference in the generated _quarto.yml. Build, then:

$ find . -name 'llms*.txt'
(nothing)

$ grep -c 'llms.txt' _site/index.html
2

Confirmed on our own site: index.html and skills.html each carry both links, and no llms*.txt exists anywhere in the project or the built output.

Impact

Two dead links on the front page of every documentation-only site, in a section specifically aimed at agents — which are the readers least likely to report a 404 and most likely to be confused by one.

It is also quietly misleading: the build log says the files were created.

Suggested fix

Gate the links on the same condition the generators use — or, better, have the generators produce a section index for non-API projects rather than returning. The sections: structure already carries titles and page lists, which is most of what llms.txt wants.

If gating is the preferred route, the minimal version is to have _generate_llms_txt report whether it wrote anything and append the links only when it did, so the two cannot drift apart again.

Workaround

We generated the two files ourselves for several days, then deleted that and now strip the two links from the built HTML instead — our site is not published, so an index for crawlers had no reader either way.

Version

great-docs 0.17.0 (latest on PyPI at time of writing), Python 3.13, Windows.

Start in core.py by reading the margin-section construction alongside _generate_llms_txt and _generate_llms_full_txt, then reproduce with a great-docs.yml containing sections but no api-reference. Check _site/index.html, generated llms*.txt files, and the build log. Done means documentation-only builds no longer advertise unavailable files, and the reported generation result matches what was written.