docs: drop the inert section frontmatter and the unpublished root llms.txt

[vpetersson-bot]

Two related things, both consequences of #138 moving the action docs out of /guides/.

The section: key does nothing

53 pages carried it — 19 guides, 19 sbomify-action, 15 compliance. Nothing reads it:

$ grep -rniE '\.params\.section|param "section"' layouts/
$ (no matches)

All 20 matches for "section" in layouts/ are .Section or where … "Section" …, which resolve Hugo's built-in page method — derived from the file's directory, and not settable from frontmatter.

It is not merely dead, it is a trap: it went stale during the /guides/ move, leaving 19 pages declaring section: guides while actually living in sbomify-action. Removed rather than wired up, since .Section is the idiomatic mechanism and already works.

The action pages had silently fallen out of llms.txt

This is the more consequential half, and it is a regression from #138 that nobody would have noticed.

/llms.txt and /llms-full.txt are generated from a fixed list of sections in layouts/home.llms.txt and layouts/home.llmsfull.txt. That list had no sbomify-action entry, so when #138 moved those pages out of guides they landed in no group at all and stopped being emitted:

$ grep -c "sbomify.com/sbomify-action/" public/llms.txt
0
$ grep -c "sbomify.com/sbomify-action/" public/llms-full.txt
0

All 20 pages, gone from both — including their full text in llms-full.txt. For a site whose whole point is being read by machines, the action documentation had become invisible to them.

Added a "sbomify Action" group to both. 18 pages return to llms.txt, 19 to llms-full.txt.

Verification

I diffed the built site file by file against a baseline built from b54f243. After committing — so enableGitInfo reads the same git state CI does — the only differences are:

What	Why
19 guides pages + sitemap.xml	dateModified / lastmod move, because those files genuinely changed
llms.txt, llms-full.txt	gaining the action pages

Nothing else in 493 pages changed, which is the point: removing the key has no rendered effect.

Also confirmed two consecutive builds from identical sources are byte-identical, so that diff is signal rather than noise.

One trap worth recording

Building with an uncommitted tree made the cra taxonomy term render as Cra instead of CRA, across 7 blog posts and 2 tag pages. The tag is spelled CRA in two posts and cra in five, and Hugo takes the display casing from page order — which shifts when enableGitInfo cannot read dates for dirty files.

It is not a real difference and disappears once committed, but it makes a dirty-tree build diff actively misleading. Worth knowing before anyone else diffs a Hugo build here.

(Separately: the tag genuinely is spelled two ways across posts, so the rendered casing is decided by build ordering rather than intent. Normalising it is a small, separate cleanup.)

The root llms.txt is deleted

/llms.txt is generated from layouts/home.llms.txt via the llms output format — as hugo.toml itself documents on line 50. The file at the repository root is not in static/, so Hugo never copies it and it has never been served.

It was not harmless. It looked authoritative, so it got maintained — f473861 "fix: update llms.txt hierarchy description" edited a file nobody could read — and it misled me during #138, where I added the new action pages to it and reported them as registered. The pages that actually needed adding were the ones missing from the generated file, fixed in the previous commit.

Nothing is lost: every section it carried exists in the generated output, and the two that looked unique to it (Content Policy, Contact) are byte-identical there.

Proven by building with and without the file — byte-identical across all 493 pages:

$ diff -rq with-root/ public/ | wc -l
0

🤖 Generated with Claude Code

https://claude.ai/code/session_01UatuV4CjoCupLTRYNbD3Wq