[BUG] I think claude's "thinking" regexes need tightening up

[obra]

Environment

• Platform (select one):
  • [x] Anthropic API
  • [ ] AWS Bedrock
  • [ ] Google Vertex AI
  • [ ] Other:
• Claude CLI version: 0.2.86
• Operating System: MacOS 14.3
• Terminal: Ghostty

Bug Description

Claude code's regexes that look for indications that the user wants thinking mode trigger in cases when they really shouldn't. For example, when the user writes "I think..." or "I don't think ...."

Steps to Reproduce

1. Enter the prompt "I think we should try building a new widget" or "I don't think we should consider the foo approach here"

Expected Behavior

Claude treats the input as the user making assertions

Actual Behavior

Claude thinks about it.

Additional Context

[dsilbergleithcu-godaddy]

I think we need to get this corrected.

[obra]

Possibly related, I've started getting these errors when I use 'think' in a reply to Claude Code:

> ✘ Test "Test application windows resource lists TextEdit windows" recorded an issue at WindowsResourceE2ETests.swift:54:6: Caught error: Invalid resource URI format:
  macos://applications/com.apple.TextEdit/windows

  i think you need to write some unit tests for the uri parser

  ⎿ API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.123.content.0.type: Expected `thinking` or `redacted_thinking`, but found `text`. When
    `thinking` is enabled, a final `assistant` message must start with a thinking block (preceeding the lastmost set of `tool_use` and `tool_result` blocks). We recommend you include
    thinking blocks from previous turns. To avoid this requirement, disable `thinking`. Please consult our documentation at
    https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking"}}

[mimkorn]

I am unconsciously training myself to avoid using "think" in my speech This self-censorship feels unnatural and restrictive The tool is changing user behavior in an unwanted way

Suggested Improvements

More Specific Trigger Words

Use less common terms like "ultrathink" Implement clear markers (e.g., "/think" or similar syntax) Choose triggers that are obviously intentional commands rather than natural language

Poor Discoverability

The feature's existence isn't immediately obvious to new users Users discover it accidentally rather than through clear documentation Inconsistent activation leads to confusion about when/why it appears

During initial weeks of use, the appearance of italic grey text was confusing The inconsistent triggering made it difficult to understand the pattern Only discovered the cause through later documentation review

Implement a more deliberate trigger mechanism that doesn't interfere with natural speech patterns

[arunsathiya]

Does anyone know if the thinking mode is enabled using the tab key, does that override the 'think' words in my input? And what capacity does the think mode enabled by tab key operate at - low, medium or ultrathink?

[obra]

I believe this issue is obsolete