superthink
Deep reasoning framework for complex tasks with confidence validation, context gathering, and systematic planning
When & Why to Use This Skill
SuperThink is a sophisticated deep reasoning framework designed to enhance AI agent performance on complex engineering tasks. By implementing a confidence-driven workflow that requires a 90% certainty threshold before execution, it ensures architectural alignment and significantly reduces token waste. This skill excels at systematic planning, deep context gathering, and multi-step validation, making it an essential tool for high-stakes software development and system integration.
Use Cases
- Complex Feature Engineering: Managing intricate code changes across multiple files and legacy systems while maintaining architectural integrity.
- Strategic Architectural Decisions: Evaluating long-term system implications and ensuring pattern compliance before implementing structural changes.
- Ambiguous Requirement Resolution: Conducting deep context gathering and dependency mapping to clarify vague project specifications.
- Security-Sensitive Implementations: Executing high-risk code modifications with a rigorous five-phase validation process to prevent vulnerabilities.
- Multi-System Integration: Orchestrating complex data flows and API interactions between diverse platforms through structured planning and tracking.
| name | superthink |
|---|---|
| description | Deep reasoning framework for complex tasks with confidence validation, context gathering, and systematic planning |
| version | 1.0.0 |
SuperThink
Confidence-driven workflow preventing misaligned execution: invest 100-200 tokens upfront to save 5,000-50,000 tokens.
When to Use
USE for:
- Complex features requiring changes across multiple files/systems
- Architectural decisions with long-term implications
- Unclear requirements needing exploration
- Multi-system integration
- Breaking changes
- Security-sensitive implementations
SKIP for:
- Simple, well-defined single-file changes
- Obvious bug fixes with clear solutions
- Routine tasks (formatting, linting, etc.)
Core Philosophy
Confidence-driven workflow: Requires ≥90% confidence score before proceeding to implementation.
- Context Gathering (30% effort) - Understand what exists
- Confidence Assessment (20% effort) - Validate readiness
- Structured Planning (30% effort) - Break down approach
- Implementation (20% effort) - Execute with tracking
Five-Phase Workflow
Phase 1: Initial Context (30 tokens)
Parse requirements | Identify systems | Create TodoWrite tasks
Phase 2: Deep Context Gathering (60 tokens)
Search codebase (Grep/Glob) | Read existing code | Check docs | Analyze dependencies
Phase 3: Confidence Assessment (40 tokens)
| Component | Weight | Validates |
|---|---|---|
| No Duplicate Implementations | 25% | Existing functionality check |
| Architecture Compliance | 25% | Pattern alignment |
| Documentation Verified | 20% | API/library validation |
| Data Model Understood | 20% | Schema/type confirmation |
| Dependencies Mapped | 10% | Side effects understood |
Requires ≥90% score to proceed.
Phase 4: Structured Planning (50 tokens)
Order by dependency | Follow existing patterns | Include validation | Expand TodoWrite
Phase 5: Execution with Validation (Variable tokens)
Mark in_progress (ONE at a time) | Execute | Validate immediately | Mark completed immediately | Handle errors at root cause
TodoWrite Hygiene
- Create tasks at Phase 1
- Expand at Phase 4
- Mark completed after each step
- ONE in_progress at a time
Success Metrics
60-70% fewer tokens | Zero failures | First-try success