ultrathink-trigger - SKILL.md Agent Skill

name: ultrathink-trigger description: Auto-escalate reasoning depth for complex problems. Use proactively when encountering architecture, debugging, or unfamiliar code.

Ultrathink Trigger

Automatically invoke maximum reasoning depth for complex problems.

Complexity Indicators

Signal	Weight	Example
Files affected > 5	+2	Refactoring across modules
Unfamiliar language/framework	+3	First time with codebase
Architecture decision	+4	Choosing patterns
2+ failed fix attempts	+3	Root cause unclear
Multi-domain interaction	+2	Frontend + Backend + DB
Security implications	+3	Auth, permissions, secrets
Performance-critical code	+2	Hot paths, algorithms
Breaking change potential	+3	API changes, schema migrations

Threshold

If total weight >= 5: Trigger ultrathink mode

Invocation

Prefix complex analysis with one of:

"Think harder about this:"
"ultrathink:"
"Take your time to deeply analyze:"

These phrases trigger Claude's extended thinking mode for more thorough analysis.

Integration Points

With /fix command

After 2 failed attempts, automatically invoke:

ultrathink: What are all the possible root causes?
What assumptions am I making? What haven't I checked yet?

With /brainstorm command

Auto-triggers when topic complexity >= 5 (or --deep flag). Applied to EVALUATE and SYNTHESIZE phases:

ultrathink: Deeply analyze these options considering trade-offs,
edge cases, maintainability, and long-term implications.

Flags:

--deep - Force ultrathink regardless of complexity
--quick - Suppress auto-detect, stay fast

With orchestrator INNOVATE phase

For architecture decisions:

ultrathink: Evaluate these approaches considering
maintainability, performance, security, and complexity.

With sequential-thinking

For multi-step complex analysis:

mcp__sequential_thinking__sequentialthinking({
  thought: "ultrathink: [complex problem]",
  thoughtNumber: 1,
  totalThoughts: 10, // More steps for deep analysis
  nextThoughtNeeded: true
})

When to Use Proactively

Architecture Decisions: Before proposing patterns or structure
Debugging Deadlock: After 2+ failed fix attempts
Security Analysis: Any auth, permission, or secret handling
Performance Optimization: Before optimizing hot paths
Breaking Changes: Before proposing API/schema changes
Unfamiliar Code: First encounter with a new codebase area

Example Workflow

1. Encounter complex problem
2. Calculate complexity weight:
   - Multi-file change: +2
   - Architecture decision: +4
   - Total: 6 (>= 5, triggers ultrathink)
3. Invoke: "ultrathink: Design the data flow for..."
4. Extended analysis produces thorough solution
5. Proceed with implementation

Anti-Patterns

Using ultrathink for simple, well-understood tasks
Skipping ultrathink for complex problems to "save time"
Not recalculating complexity when scope changes