Summary

• Problem: When thinking=low (or any thinking mode) is enabled, the model's reasoning/thinking blocks were being sent to WhatsApp users as visible messages (issues Reasoning/thinking blocks leak into WhatsApp messages #25214, WhatsApp: Internal reasoning exposed as user message #24328).
• Why it matters: Users were seeing internal model reasoning that should be hidden, causing confusion and poor UX.
• What changed: Added a filter in deliverWebReply to detect and skip messages that start with "Reasoning:" prefix.
• What did NOT change: Normal messages that happen to contain "Reasoning:" in the middle are still delivered. Only messages that START with "Reasoning:" are filtered.

Change Type (select all)

• Bug fix
• Feature
• Refactor
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes Reasoning/thinking blocks leak into WhatsApp messages #25214
• Closes WhatsApp: Internal reasoning exposed as user message #24328

User-visible / Behavior Changes

WhatsApp users will no longer see internal reasoning messages when using models with reasoning capabilities.

Security Impact (required)

• New permissions/capabilities? (No)
• Secrets/tokens handling changed? (No)
• New/changed network calls? (No)
• Command/tool execution surface changed? (No)
• Data access scope changed? (No)

Repro + Verification

Environment

• OS: Linux (tested via unit tests)
• Channel: WhatsApp Web

Steps

1. Enable thinking mode with a reasoning-capable model (e.g., thinking=low)
2. Send a message via WhatsApp
3. The reasoning block should NOT appear in the delivered message

Expected

• Only the final answer is delivered to WhatsApp

Actual (before fix)

• Reasoning block was visible: Reasoning:\n_some internal thinking_

Evidence

• Failing test/log before + passing after

Added 3 new tests:

1. Filters out reasoning messages from delivery
2. Filters out reasoning messages with leading whitespace
3. Delivers normal messages that contain Reasoning: prefix in the middle

All 12 tests pass (9 existing + 3 new).

Human Verification (required)

• Verified scenarios:
  • Reasoning message is filtered and not delivered
  • Message with "Reasoning:" in the middle is still delivered
  • Empty/null text is handled correctly
• Edge cases checked:
  • Leading whitespace before "Reasoning:"
  • Multiple lines of reasoning
• What you did not verify:
  • Actual WhatsApp Web integration (only unit tests)

Compatibility / Migration

• Backward compatible? (Yes)
• Config/env changes? (No)
• Migration needed? (No)

Failure Recovery (if this breaks)

• How to disable/revert this change quickly: Revert commit 122dba75c
• Known bad symptoms reviewers should watch for: Messages that should be delivered are being filtered

Risks and Mitigations

• Risk: Rare false positive where a user message starts with "Reasoning:"
  • Mitigation: The check only matches the exact prefix "Reasoning:" which is an internal formatting used by OpenClaw. User messages are unlikely to start with this exact string.

Greptile Summary

This PR filters out reasoning messages from WhatsApp delivery to prevent internal model thinking from being visible to end users.

• Added isReasoningMessage() helper that checks if text starts with Reasoning: prefix after trimming whitespace
• Early return in deliverWebReply() when reasoning message detected
• Comprehensive test coverage for reasoning filtering, whitespace handling, and false positives

The fix aligns WhatsApp with other channels (Discord, Matrix, Telegram) that already filter reasoning content. The implementation is simple, focused, and well-tested.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• Simple, focused bug fix with excellent test coverage and clear documentation. The logic is straightforward (prefix check), tests cover edge cases (whitespace, false positives), and the implementation follows existing patterns in the codebase for other channels.
• No files require special attention

_{Last reviewed commit: 9e23685}

_{(3/5) Reply to the agent's comments like "Can you suggest a fix for this @greptileai?" or ask follow-up questions!}