[BUG] <thought> tag is not properly rendered

[NikiKeyz]

Problem (one or two sentences)

Gemma 4 31B outputs its thinking process with tags which are not properly rendered in Roo Code UI.

Context (who is affected and when)

Gemma 4 31B outputs its thinking process with tags which are not properly rendered in Roo Code UI.

Reproduction steps

Gemma 4 31B model through OpenAI compatible https://generativelanguage.googleapis.com/v1beta/openai/
Thinking turned on.
Any prompt to reproduce.

Expected result

Thoughts of model rendered as usual

Actual result

in model's answer

Variations tried (optional)

No response

App Version

3.52.0

API Provider (optional)

OpenAI Compatible

Model Used (optional)

Gemma 4 31B

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote-v0]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause: Gemma 4 31B wraps its reasoning in <thought>...</thought> tags, but Roo Code only recognizes <think>...</think> tags for reasoning extraction. There are two places this needs to be fixed:

1. Streaming tag extraction (src/utils/tag-matcher.ts) - The TagMatcher class is instantiated with "think" as the tag name in 4 providers (openai.ts, base-openai-compatible-provider.ts, lm-studio.ts, native-ollama.ts). It parses streaming content character-by-character to separate reasoning from regular text. Currently it only matches <think> tags.

2. Post-processing strip (src/core/assistant-message/presentAssistantMessage.ts) - After streaming, any remaining <thinking> tags are stripped from text output. This also needs to handle <thought> tags.

Proposed changes:

1. Modify TagMatcher to support multiple tag names - Update the constructor to accept tagName: string | string[] and track multiple candidate tag names during character-by-character matching. This is the cleanest approach since string replacement on chunked streams would be fragile.

2. Update all 4 TagMatcher instantiation sites to include both "think" and "thought":

   • src/api/providers/openai.ts (line ~180)
   • src/api/providers/base-openai-compatible-provider.ts (line ~120)
   • src/api/providers/lm-studio.ts (line ~107)
   • src/api/providers/native-ollama.ts (line ~217)

3. Update presentAssistantMessage.ts to also strip <thought> and </thought> tags (line ~291).

4. Add tests for TagMatcher with <thought> tags, and update existing tests as needed.

This approach is backwards-compatible (existing <think> tag handling is unchanged) and extensible for future models that might use other reasoning tag names.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[NikiKeyz]

@roomote, seems to be a nice plan!

[roomote-v0]

Thanks for the approval! I've seen your comment and will proceed with the implementation. I'll open a draft PR to share progress shortly.

[roomote-v0]

Opened a draft PR with my attempt at addressing this: #12096

The changes update TagMatcher to support multiple tag names, so both <think> and <thought> tags are now recognized as reasoning content across all 4 providers. Feedback welcome!