Summary

• When streamMode is "partial", reasoning/thinking block content leaks into the Discord draft preview because the partial text is forwarded to the draft stream without any filtering
• Apply stripReasoningTagsFromText before updating the draft and skip pure-reasoning messages (those starting with "Reasoning:\n") so internal thinking traces never reach the user-visible preview
• This mirrors the approach already used by the Telegram channel (splitTelegramReasoningText in reasoning-lane-coordinator.ts)

Test plan

• Configure Discord channel with streamMode: "partial" and a model with reasoning enabled
• Send prompts that trigger multi-step reasoning
• Verify reasoning blocks no longer appear in the streamed preview messages
• Verify final reply content is unaffected

Fixes #24532

🤖 Generated with Claude Code

Greptile Summary

Fixes reasoning/thinking tag content leaking into Discord draft stream previews when streamMode is "partial" (or "block"). The updateDraftFromPartial function now calls stripReasoningTagsFromText to remove XML-style thinking tags (<think>, <thinking>, etc.) and skips messages prefixed with "Reasoning:\n" that contain only formatted reasoning content.

• Applies stripReasoningTagsFromText with { mode: "strict", trim: "both" } to incoming partial text before updating the draft stream
• Skips empty results and pure-reasoning messages (those starting with "Reasoning:\n")
• All subsequent logic in updateDraftFromPartial correctly uses the cleaned variable instead of raw text
• Mirrors the existing approach used by the Telegram channel (splitTelegramReasoningText in reasoning-lane-coordinator.ts)

Confidence Score: 4/5

• This PR is safe to merge — it adds a well-scoped filter to the Discord draft stream path that mirrors an established pattern from the Telegram channel.
• The change is narrowly scoped to one function (updateDraftFromPartial) in one file, uses an existing shared utility (stripReasoningTagsFromText), and follows the same pattern already validated in the Telegram channel. All references to partial text within the function correctly use the cleaned variable. The only minor concern is the lack of dedicated unit tests for this specific filtering behavior, but the existing test coverage for stripReasoningTagsFromText and the Telegram reasoning coordinator provide indirect validation.
• No files require special attention

_{Last reviewed commit: 93403bf}