Fix duplicate assistant messages during overlapping submits

[whoiskatrin]

Summary

• Keep the active local assistant message pinned as the internal tail entry while an overlapping submit is sent, so AI SDK stream writes keep replacing the in-flight assistant instead of appending a duplicate
• Preserve that protected assistant when CF_AGENT_CHAT_MESSAGES broadcasts arrive mid-stream, then restore the original message ordering once the stream completes
• Use an anchor-based restore (the ID of the preceding message) instead of a numeric index, so the assistant returns to the correct position even if the server injects or reorders messages during the stream
• Clear protection state in the useEffect cleanup to prevent leaks on unmount or effect re-runs
• Add six React regression tests covering the core overlap scenario, multiple server broadcasts, CF_AGENT_CHAT_CLEAR mid-protection, anchor-based restore with server-injected messages, done:true without a prior start chunk, and the not-streaming guard

Closes #1231

Testing

npm run test:react -w @cloudflare/ai-chat -- --run
npx oxfmt --check packages/ai-chat/src/react.tsx packages/ai-chat/src/react-tests/use-agent-chat.test.tsx
npx oxlint packages/ai-chat/src/react.tsx packages/ai-chat/src/react-tests/use-agent-chat.test.tsx

Reviewer Notes

• The root issue is in AI SDK chat state: stream updates replace the last message only while the in-flight assistant is still the array tail. An overlapping submit inserts a user message after it, so later chunks append a second assistant with the same ID.
• This fix stays entirely in useAgentChat; it does not change server persistence or messageConcurrency semantics.
• The protection is intentionally temporary: we only keep the streaming assistant at the tail while the local request is active, then restore the original order so the visible conversation stays user, assistant, user between turns.
• Restore uses an anchor message ID (the message before the assistant) rather than a numeric index. This is robust against insertions, deletions, or reorderings that happen between protection and restore (e.g., server-injected system messages or cross-tab broadcasts).
• Complementary to fix(ai-chat): strip messageId from continuation start chunks (#1229) #1235 (4f79280) which fixes duplicate assistants during continuations server-side; this PR fixes duplicate assistants during overlapping submits client-side. No overlap — different files, different layers, different triggers.

---

[changeset-bot]

🦋 Changeset detected

Latest commit: 19aaf9c

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@cloudflare/ai-chat	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[pkg-pr-new]

Open in StackBlitz

agents

npm i https://pkg.pr.new/agents@1232

@cloudflare/ai-chat

npm i https://pkg.pr.new/@cloudflare/ai-chat@1232

@cloudflare/codemode

npm i https://pkg.pr.new/@cloudflare/codemode@1232

hono-agents

npm i https://pkg.pr.new/hono-agents@1232

@cloudflare/shell

npm i https://pkg.pr.new/@cloudflare/shell@1232

@cloudflare/think

npm i https://pkg.pr.new/@cloudflare/think@1232

@cloudflare/voice

npm i https://pkg.pr.new/@cloudflare/voice@1232

@cloudflare/worker-bundler

npm i https://pkg.pr.new/@cloudflare/worker-bundler@1232

commit: 19aaf9c

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.