Summary

• When includeReasoning is active (or reasoningLevel falls back to the model default after being deleted rather than set to "off"), the agent emits reasoning blocks as separate reply payloads prefixed with "Reasoning:\n"
• Matrix has no dedicated reasoning lane (unlike Telegram which splits reasoning into a separate stream), so these internal thinking traces leak into the chat room as regular user-visible messages
• Filter out pure-reasoning payloads (those starting with "Reasoning:\n" or a <thinking> tag) before delivery so internal reasoning never reaches the Matrix room

Test plan

• Enable a reasoning-capable model (e.g. Claude Opus) with extended thinking
• Send messages via Matrix channel
• Verify reasoning blocks (e.g. "Reasoning: Planning change summary...") no longer appear as separate chat messages
• Verify normal assistant replies are unaffected

Fixes #24411

🤖 Generated with Claude Code

Greptile Summary

Adds filtering to prevent reasoning-only messages from appearing in Matrix chat rooms. When reasoning is enabled, the agent emits internal thinking traces prefixed with "Reasoning:\n" or wrapped in <thinking> tags. Since Matrix lacks a dedicated reasoning channel (unlike Telegram), these would leak as user-visible messages. The fix detects and skips pure reasoning payloads before delivery while preserving normal assistant replies.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The change is well-scoped, defensive, and follows established patterns from the Telegram implementation. The filtering logic is conservative (only removes reasoning-only messages), preserves all normal replies, and adds verbose logging for observability. No breaking changes or edge cases identified.
• No files require special attention

_{Last reviewed commit: e5dfbdd}