@llui/compiler-introspection

Opt-in compiler module that emits runtime-introspection metadata. Consumed by @llui/agent (for end-user agent surfaces) and @llui/mcp (for the dev MCP server).

When enabled, the compiler emits:

• Msg schemas — discriminated-union JSON Schema for every component's Msg, used by llui_validate_message and llui_get_message_schema.
• Msg annotations — @intent, @should, @emits, etc. surfaced as runtime metadata.
• State schemas — JSON Schema for every component's State.
• Binding descriptors — per-binding kind/mask/path data used by llui_get_bindings.
• Schema hash — a stable content hash so consumers can detect schema changes across reloads.

Why it's separate

These payloads add bundle weight that production apps without agent/MCP integration don't need. Splitting them into an opt-in package keeps the @llui/compiler engine lean.

Constants

introspectionFactory

Builds the introspection module set for a single source file. Activation order matches the v2c/decomp-7 design:

1. bindingDescriptorsModule — preTransform fires first so the universal handler-tagger + scope-variant-registration runs before any visitor or emit phase sees the file.
2. msgSchemaModule, stateSchemaModule, msgAnnotationsModule — producer modules populate the schema-hash inputs slot.
3. schemaHashModule — emit reads the populated slot. Per-module activation gates mirror what transformLlui did inline before decomp-26:

• msgSchemaModule when at least one of Msg / Effect schemas extracted
• stateSchemaModule when State schema extracted
• msgAnnotationsModule when annotation map non-null (includes empty)
• schemaHashModule always (well-defined hash over null inputs)
• bindingDescriptorsModule always when introspection is on The "always-on schemaHash" comes from spec §7.4: the hash ships in prod too, used by HMR re-send gating regardless of agent mode.

const introspectionFactory: IntrospectionFactory