llms.txt / llms-full.txt

This site is LLM-native. Point a model at one URL and it learns the entire Itera lesson-authoring surface — the 21 tools, the model, the auth, the results API.

/llms.txt A curated index — section + page titles with links. The map.

/llms-full.txt The full concatenated corpus — every page's content in one plain-text file. The territory.

What they are

llms.txt is an emerging convention for exposing a site’s content in a form optimised for LLM ingestion. Itera docs ship two files, both generated at build time from the same content you’re reading (via the starlight-llms-txt plugin) — so they never drift from the docs.

File	Shape	Use it when
/llms.txt	Curated index (titles + links, grouped)	You want the model to know what exists and fetch pages on demand.
/llms-full.txt	Full corpus (all page bodies concatenated)	You want the model to learn everything at once — paste it into context.

Teach an LLM to author

# Give a model the whole authoring surface in one shot:
curl -s https://docs.iterahq.dev/llms-full.txt

Then ask it to author — it has the tool names, inputs, the Program → Unit → Lesson → Activity model, the ten activity kinds, and the auth flow. A good smoke test: “Using the Itera authoring tools, emit the create_program → create_unit → create_lesson → create_activity call sequence for a two-activity lesson.” The shapes should match the reference.

Parity check

The tool reference is hand-authored, so a CI check keeps it honest. On every build/PR, scripts/check-docs-sync.mjs asserts that the 21 canonical itera-mcp tool names (the source of truth is ITERA_TOOL_NAMES in iterahq/itera-mcp, itself count-gated in that repo’s tests/registration.test.ts) each have a documented entry under reference/, and that no undocumented tool name sneaks in. It exits non-zero (red CI) on any drift.

Why this matters

The llms-full.txt corpus is only as trustworthy as the docs. The parity check guarantees a model reading it sees the real tool set — not a stale or invented one.