llms-txtlisted
Install: claude install-skill lroolle/skills
# llms-txt
llms.txt is robots.txt inverted: not what a crawler may read, but
the fastest true path through what it should. One markdown file at
the corpus root -- a title, a blockquote summary, and sections of
links with one-line descriptions -- so orientation costs an agent
one fetch instead of a crawl through HTML built for humans.
It is a derived artifact, and stability is the contract: the map is
regenerated from the corpus, never hand-grown. Rerun on an unchanged
corpus, the file is byte-identical; refresh after a change, and only
the entries whose sources changed move. A map that drifts from its
territory is worse than no map -- agents trust the map first.
## Gate
Honesty first: evidence that classic search crawlers fetch llms.txt
is thin. Its real consumers are agents fetching docs on demand and
humans loading docs into context. Sell it as agent-legibility; treat
GEO/SEO as possible upside, never the promise.
| Situation | Move |
|---|---|
| Goal is ranking in classic search | Not this file -- improve the docs and page metadata themselves. |
| Private corpus for your own agents | That is a vault's index.md (wiki-it), not a published map. |
| Whole-repo release polish | burnish's job; it routes back here for this one artifact. |
| Corpus is a single README | The README already is the map. Skip. |
| Site, docs dir, repo, or vault that agents will read | Build the map. |
## The format
The shape, per the llms.txt proposal:
```markdown
# Project Name
> One-paragrap