fix!: build judge input as string and strip legacy judge config messages

[jsonbailey]

Summary

• Aligns the JS AI SDK with python-server-sdk-ai#165 and sdk-specs#160.
• Judges now pass a single formatted string (MESSAGE HISTORY:\n...\n\nRESPONSE TO EVALUATE:\n...) to the runner instead of an interpolated message list.
• Legacy judge configs with {{message_history}} / {{response_to_evaluate}} placeholder messages are stripped at config construction so old and new flag values continue to work.
• Runner.run is narrowed to (input: string, outputType?) everywhere — server-ai, server-ai-openai, server-ai-langchain, server-ai-vercel.

Breaking changes

• Runner.run(input: string | LDMessage[], outputType?) → Runner.run(input: string, outputType?). Provider runners (OpenAI, LangChain, Vercel) no longer accept a pre-built LDMessage[]. They always prepend any config messages and append the prompt as a user turn.
• Judge no longer Mustache-interpolates {{message_history}} / {{response_to_evaluate}} into config messages. The SDK builds the input string and passes it verbatim to the runner.
• Judge.evaluate no longer early-returns with "Judge configuration must include messages" — a judge with no messages now proceeds to invoke the runner with the formatted input string.

Test plan

• yarn workspace @launchdarkly/server-sdk-ai test — 239 passing
• yarn workspace @launchdarkly/server-sdk-ai-openai test — 39 passing
• yarn workspace @launchdarkly/server-sdk-ai-langchain test — 38 passing
• yarn workspace @launchdarkly/server-sdk-ai-vercel test — 20 passing
• lint clean for all four packages
• yarn workspaces foreach -pR --topological-dev --from '@launchdarkly/server-sdk-ai' run build — all pass
• Run chat-judge example end-to-end and confirm judge events emit correctly

🤖 Generated with Claude Code

---

Note

Medium Risk
Medium risk due to a breaking API change (Runner.run no longer accepts LDMessage[]) and behavioral changes in judge evaluation/config processing that could affect downstream integrations.

Overview
Breaking API change: Runner.run is narrowed from (string | LDMessage[]) to (string) across server-ai and provider packages, and provider runners (OpenAI/LangChain/Vercel) now always build the message list by prepending config messages and appending the prompt as a user turn.

Judge evaluation update: Judge stops Mustache-interpolating legacy templates into message arrays and instead builds a single evaluation string (MESSAGE HISTORY... / RESPONSE TO EVALUATE...) that it passes to the runner; it also no longer errors when judgeConfig.messages is missing.

Backward compatibility: LDAIClientImpl.judgeConfig now strips legacy judge placeholder messages containing {{message_history}}/{{response_to_evaluate}} before returning configs, with tests updated/added to cover the new input format and stripping behavior.

^{Reviewed by Cursor Bugbot for commit f4960cc. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

@launchdarkly/js-sdk-common size report
This is the brotli compressed size of the ESM build.
Compressed size: 26281 bytes
Compressed size limit: 29000
Uncompressed size: 128971 bytes

[github-actions]

@launchdarkly/js-client-sdk size report
This is the brotli compressed size of the ESM build.
Compressed size: 31867 bytes
Compressed size limit: 34000
Uncompressed size: 113634 bytes

[github-actions]

@launchdarkly/browser size report
This is the brotli compressed size of the ESM build.
Compressed size: 179543 bytes
Compressed size limit: 200000
Uncompressed size: 830815 bytes

[github-actions]

@launchdarkly/js-client-sdk-common size report
This is the brotli compressed size of the ESM build.
Compressed size: 38473 bytes
Compressed size limit: 39000
Uncompressed size: 211104 bytes

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit f4960cc. Configure here.}