anomalyco / anomalyco/opencode

[Bug] Empty tool_calls arrays split continuous reasoning into separate parts

Open
#47,875 0 comments 0 reactions 1 assignee View on GitHub

@nexxeln is already working on this.

Since Sep 8, 2026.

Dominant language
TypeScript
Stars
209k
Forks
27.5k
PR merge metrics
PR metrics pending

Description

Description

OpenCode's pinned @ai-sdk/openai-compatible@2.0.41 closes active reasoning whenever a streamed delta contains tool_calls, even when the array is empty. Consecutive reasoning deltas with tool_calls: [] therefore produce separate persisted reasoning parts instead of one continuous part.

This is the SDK bug tracked in vercel/ai#20203, fixed by vercel/ai#20253 and published in compatible 3.0.44. OpenCode's pinned version still needs the fix. A narrow backport would avoid a major SDK migration.

Related: #40865 describes fragmented Thought blocks, but also content-field behavior outside this report's scope. This report only concerns consecutive reasoning_content deltas with empty tool-call arrays, not merging across real text or tool boundaries.

Plugins

None required; reproduced with local SSE fixtures through the SDK and real session processor.

OpenCode version

dev at ecbc6ccac85b3e8087b6445e584318419b9e2b34; ai@6.0.168, @ai-sdk/openai-compatible@2.0.41, Bun 1.3.14.

Steps to reproduce
  1. Configure an OpenAI-compatible test endpoint to return this chat-completion SSE response:
data: {"id":"repro","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"reasoning_content":"r1","tool_calls":[]},"finish_reason":null}]}

data: {"id":"repro","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"reasoning_content":"r2","tool_calls":[]},"finish_reason":null}]}

data: {"id":"repro","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"done"},"finish_reason":null}]}

data: {"id":"repro","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]
  1. Send a prompt and inspect the assistant's persisted reasoning parts.

Expected: one completed reasoning part, r1r2, followed by text done.

Actual: two completed reasoning parts, r1 and r2; text done remains intact. Omitting tool_calls produces the expected single part.

Screenshot and/or share link

Not applicable; reproduced with headless SDK/session tests, not a live provider account.

Operating System

Linux.

Terminal

Not applicable to the headless reproduction.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.