fix(openai_compat): strip <think> tags from content before sending to…

[denysvitali]

📝 Description

Some models (e.g. MiniMax M2.5, DeepSeek) embed chain-of-thought in the content field using tags rather than a dedicated reasoning_content field. Move these blocks into ReasoningContent so channels don't render raw thinking output to users.

🗣️ Type of Change

• 🐞 Bug fix (non-breaking change which fixes an issue)
• ✨ New feature (non-breaking change which adds functionality)
• 📖 Documentation update
• ⚡ Code refactoring (no functional changes, no api changes)

🤖 AI Code Generation

• 🤖 Fully AI-generated (100% AI, 0% Human)
• 🛠️ Mostly AI-generated (AI draft, Human verified/modified)
• 👨‍💻 Mostly Human-written (Human lead, AI assisted or none)

🔗 Related Issue

📚 Technical Context (Skip for Docs)

• Reference URL:
• Reasoning:

🧪 Test Environment

• Hardware:
• OS:
• Model/Provider:
• Channels:

📸 Evidence (Optional)

Click to view Logs/Screenshots

☑️ Checklist

• My code/docs follow the style of this project.
• I have performed a self-review of my own changes.
• I have updated the documentation accordingly.

[CLAassistant]

All committers have signed the CLA.

[nikolasdehor]

This addresses the same problem as PR #1237 (leaked thinking tags) but takes a different approach: extracting <think> content into ReasoningContent rather than stripping all reasoning/thinking tags.

Comparison with PR #1237:

• #1233 (this PR): only handles <think>...</think>, moves content to ReasoningContent
• #1237: handles <think>, <thinking>, <thought>, <reasoning>, <final>, escaped unicode, unclosed blocks -- strips all of them

Feedback on this PR:

1. Only <think> is handled -- some providers use <thinking>, <thought>, or <reasoning> tags. PR #1237 is more comprehensive in this regard.

2. Good idea to preserve reasoning -- moving thinking content to ReasoningContent rather than discarding it is arguably better than pure stripping. This lets channels/UI that support reasoning display optionally show it. However, the condition `if msgReasoning == "" && strings.Contains(msgContent, "")" means this only fires when no dedicated reasoning field exists.

3. Regex concern -- (?s)<think>(.*?)</think> with the lazy quantifier is correct for matching the innermost <think> block. However, this won't handle nested tags (unlikely but possible) or malformed HTML.

4. No test coverage -- PR #1237 includes tests; this PR does not. Tests are important for a parsing change like this.

5. Redundant regex compilation inside callback -- thinkTagRe.FindStringSubmatch(m) is called inside ReplaceAllStringFunc, which already matched the regex. The submatch is available if you use ReplaceAllStringSubmatchFunc (not in stdlib, but the pattern can be restructured to use FindAllStringSubmatch + manual reconstruction).

I would recommend consolidating this with PR #1237. The "move to ReasoningContent" behavior from this PR is superior to pure stripping, while #1237 has better tag coverage and tests.

[sipeed-bot]

@denysvitali Hi! This PR has had no activity for over 2 weeks, so I'm closing it for now to keep things organized. Feel free to reopen anytime if you'd like to continue.