Publish AI reference discovery and embedding instructions

docs/ai_agents.md

@@ -0,0 +1,24 @@
+# Lyng instructions for AI agents
+
+Before writing Lyng, read the [current language reference](ai_language_reference.md)
+and [compact standard-library reference](ai_stdlib_reference.md). These are the
+canonical implementation-oriented references; older specifications and proposals
+must not override them. Discover them from <https://lynglang.com/llms.txt>.
+
+Use the language reference for syntax, types, generic bounds, imports, and current
+limitations. Use the stdlib reference to distinguish default globals from explicit
+imports and optional host modules. Follow links to individual API pages when you
+need more detail. Do not guess Kotlin/Java method names or assume optional modules
+are installed. Check the host runtime version and validate uncertain constructs
+with a small read-only script before changing user data.
+
+For offline embedding, copy both compact references from the same committed Lyng
+revision, retain their paths and commit ID, and split long results at section
+boundaries to fit the host's tool output limit. Keep host-specific capability and
+permission documentation alongside them. In Tetradion, call `agent.docsList()` or
+`agent.docsSearch("query")`, then `agent.docs("reference.syntax")`,
+`agent.docs("reference.flow")`, `agent.docs("reference.types")`,
+`agent.docs("reference.modules")`, or `agent.docs("reference.stdlib")`.
+
+The site publishes these Markdown files directly under `/docs/`, without requiring
+JavaScript or browser automation.