llms.txt

The hub publishes two machine-readable files so an AI agent can consume this documentation as text instead of scraping HTML: an index and a full-text corpus. Both are generated from the same route registry that renders the site, so they can never describe pages the site does not stand behind.

Purpose

• https://hub.aggregator.gg/llms.txt is the index: one line per approved page with its title, canonical URL, and description, following the llms.txt convention.
• https://hub.aggregator.gg/llms-full.txt is the corpus: the complete markdown body of every approved page in one file, ready to load into an agent’s context without any crawling.

If you are building an agent that integrates against Aggregator.gg, fetch llms-full.txt once and treat it as the documentation corpus; the working rules for such an agent live in Agent guidelines.

Generated files

Both files are rebuilt from the route registry on every site build; they are artifacts of the same commit that produced the pages.

llms.txt is a markdown list under a single header:

# Aggregator.gg Developer Hub

## Documentation

- [Getting started](https://hub.aggregator.gg/getting-started/): Register, integrate, verify and launch with human control...
- [Callbacks](https://hub.aggregator.gg/callbacks/): The full wallet callback contract...

llms-full.txt concatenates every approved page. Each page is delimited by an HTML comment naming its route id, then its title as a heading, then the body reduced to plain markdown: component imports and JSX scaffolding are stripped, tabbed code groups are flattened into labelled blocks, and asides become blockquotes, so the text needs no MDX runtime to read:

# Aggregator.gg Developer Hub - full documentation

Single-file concatenation of every approved page on https://hub.aggregator.gg.

<!-- getting-started -->

# Getting started
...

---

<!-- callbacks -->

# Callbacks
...

Inclusion policy

A page enters the LLM surfaces only when all of the following hold in the route registry:

1. publication.public is true;
2. publication.approved is true;
3. llm.include is true;
4. provenance status is approved;
5. a content digest is recorded.

Draft and review-only routes never leak into either file: a route failing any gate is withheld from both surfaces, and when no route qualifies at all, the files carry an explicit review-build notice rather than partial content. The same registry drives search and SEO flags, so a page’s visibility to crawlers, to site search, and to agents is one decision, not three.

Provenance

Every included page is pinned: the registry records a SHA-256 digest of the page’s authored source, and the machine-checked content ledger records, per page, the live-docs or code sources it was ported from, its money-path claims with evidence for each, and every deliberate delta against the source material. The canonical URLs inside both files point at hub.aggregator.gg, so content you fetched can always be traced back to the page, and the page to its sources.

For an agent this means the corpus is not just text: any money-affecting statement in it (units, retries, declines) was individually evidenced against production documentation or platform code before the page was approved.

Validation

The repository test suite enforces the gate mechanically: draft routes must stay off both surfaces, approved routes must carry matching digests, and a registry that tried to include an unapproved page fails the build. The built-site check then verifies the generated artifacts before deploy.

Step check: fetch https://hub.aggregator.gg/llms.txt and confirm every listed URL responds with a live page; fetch https://hub.aggregator.gg/llms-full.txt and confirm each <!-- route-id --> marker matches a page listed in the index, no more, no fewer.