AI Assistance Disclosure 🤖

Per CONTRIBUTING.md requirements:

• AI-assisted: Built with Claude Code (Opus 4.5)
• Testing level: Fully tested - ran test suite (182 tests passed), type-check, and manual verification sending messages through WhatsApp
• I understand what this code does:
  • tryFlattenLiteralAnyOf() detects TypeBox's Type.Union([Type.Literal(...)]) pattern (which compiles to anyOf with const values) and converts it to a flat enum array
  • The BrowserActSchema discriminated union was flattened to a single object with all properties optional (except the discriminator kind)
  • Runtime validation still happens in the execute functions; the schema changes only affect what's sent to the LLM provider

Response to GPT-5.2-codex review

High (text for wait): The text field IS present in the flattened schema - it's shared between the type and wait actions. Verified:

Request properties: kind, targetId, ref, doubleClick, button, modifiers, text, submit, slowly, key, startRef, endRef, values, fields, width, height, timeMs, textGone, fn

Medium (relaxed validation): Acceptable tradeoff. The browser control server validates requests and returns clear error messages for invalid combinations. The LLM uses the schema as guidance, and runtime errors are handled gracefully.

Low (dropped metadata): Fixed in e9c834f - tryFlattenLiteralAnyOf now preserves title, default, and examples metadata in addition to description.

Thanks ! Welcome to the Clawtributors! Say hi on Discord!

Landed on main as c3b3f57 + a2b3f2c. Added a README co-author credit commit (a882beb) so @carlulsoe shows up in the contributors graph. Closing PR.