claude-mods/rules/thinking.md

Extended Thinking

Claude Code supports extended thinking mode for deeper analysis. Trigger with these phrases:

Thinking Triggers

Phrase	Budget	Use When
think	Low	Quick analysis, simple decisions
think hard	Medium	Multiple options to evaluate
think harder	High	Complex problems, architectural decisions
ultrathink	Maximum	Critical decisions, security review

Usage Examples

"think about how to structure this feature"
"think hard about the trade-offs here"
"think harder about potential edge cases"
"ultrathink about the security implications"

When to Use Each Level

think - Quick Analysis

• Choosing between 2-3 simple options
• Reviewing straightforward code
• Making minor decisions

think hard - Deeper Analysis

• Planning a new feature
• Evaluating multiple approaches
• Debugging non-obvious issues

think harder - Comprehensive Evaluation

• Architectural decisions
• Complex refactoring plans
• Performance optimization strategies
• Trade-off analysis with many variables

ultrathink - Maximum Depth

• Security-critical code review
• Production incident investigation
• Major architectural changes
• Risk assessment for breaking changes

Best Practices

1. Use during planning - Think before implementing
2. Match complexity - Don't ultrathink simple tasks
3. Combine with subagents - "think harder, then use subagents to verify"
4. Be specific - "think hard about the authentication flow" > "think hard"

Example Workflow

1. "Read the authentication module"
2. "think harder about how to add OAuth support"
3. [Claude provides detailed analysis]
4. "Now implement it following the plan"

Anti-Patterns

BAD:  "ultrathink about this typo fix"
GOOD: "fix the typo in line 42"

BAD:  "think" (with no context)
GOOD: "think about the best data structure for this use case"

BAD:  Jumping straight to implementation
GOOD: "think hard about the approach, then implement"