184 lines
5.4 KiB
Markdown
184 lines
5.4 KiB
Markdown
# Sequential Thinking Agent Skill
|
|
|
|
Structured, reflective problem-solving methodology converted from the sequential-thinking MCP server into a native Agent Skill.
|
|
|
|
## Overview
|
|
|
|
This skill teaches Claude to apply systematic sequential thinking methodology for complex problem-solving, without relying on external MCP tools. It enables:
|
|
- Breaking down complex problems into manageable thought sequences
|
|
- Dynamic adjustment of thought count as understanding evolves
|
|
- Revision of previous thoughts when new insights emerge
|
|
- Branching into alternative reasoning paths
|
|
- Hypothesis generation and verification
|
|
|
|
## Skill Structure
|
|
|
|
```
|
|
sequential-thinking/
|
|
├── SKILL.md (105 lines)
|
|
│ Core methodology, when to apply, scripts usage
|
|
│
|
|
├── package.json
|
|
│ Test dependencies (jest)
|
|
│
|
|
├── .env.example
|
|
│ Configuration options
|
|
│
|
|
├── scripts/
|
|
│ ├── process-thought.js (executable)
|
|
│ │ Validate and track thoughts deterministically
|
|
│ │
|
|
│ └── format-thought.js (executable)
|
|
│ Format thoughts for display (box/simple/markdown)
|
|
│
|
|
├── tests/
|
|
│ ├── process-thought.test.js
|
|
│ │ Validation, tracking, history tests
|
|
│ │
|
|
│ └── format-thought.test.js
|
|
│ Formatting tests (all formats)
|
|
│
|
|
└── references/
|
|
├── core-patterns.md (95 lines)
|
|
│ Essential revision & branching patterns
|
|
│
|
|
├── examples-api.md (88 lines)
|
|
│ API design example walkthrough
|
|
│
|
|
├── examples-debug.md (90 lines)
|
|
│ Performance debugging example
|
|
│
|
|
├── examples-architecture.md (94 lines)
|
|
│ Architecture decision example
|
|
│
|
|
├── advanced-techniques.md (76 lines)
|
|
│ Spiral refinement, hypothesis testing, convergence
|
|
│
|
|
└── advanced-strategies.md (79 lines)
|
|
Uncertainty management, revision cascades, meta-thinking
|
|
```
|
|
|
|
**Documentation**: 627 lines across 7 files (all under 100 lines)
|
|
**Scripts**: 2 executable Node.js scripts with tests
|
|
|
|
## Key Features
|
|
|
|
### Progressive Disclosure Design
|
|
Each file focuses on specific aspects, loaded only when needed:
|
|
- **SKILL.md**: Quick reference with core methodology
|
|
- **core-patterns.md**: Common patterns for everyday use
|
|
- **examples-*.md**: Real-world walkthroughs for learning
|
|
- **advanced-*.md**: Sophisticated techniques for complex scenarios
|
|
|
|
### Token Efficiency
|
|
- Concise explanations sacrifice grammar for brevity
|
|
- Examples demonstrate patterns without verbose explanation
|
|
- Cross-references between files avoid duplication
|
|
|
|
### Methodology Conversion
|
|
Extracted from MCP server's approach and converted to instructions:
|
|
- MCP tool provided **interface** for sequential thinking
|
|
- Agent skill provides **methodology** to think sequentially
|
|
- No dependency on external tools—pure instructional approach
|
|
|
|
## Usage Modes
|
|
|
|
**Explicit Mode**: Use visible thought markers
|
|
```
|
|
Thought 1/5: [Analysis]
|
|
Thought 2/5: [Further analysis]
|
|
```
|
|
|
|
**Implicit Mode**: Apply methodology internally without cluttering output
|
|
|
|
## When Claude Should Use This Skill
|
|
|
|
Automatically activated for:
|
|
- Complex problem decomposition
|
|
- Adaptive planning with potential revisions
|
|
- Debugging and root cause analysis
|
|
- Architecture and design decisions
|
|
- Problems with unclear or emerging scope
|
|
- Multi-step solutions requiring context
|
|
|
|
## Scripts Usage
|
|
|
|
### Process Thought (Validation & Tracking)
|
|
|
|
```bash
|
|
# Process a thought
|
|
node scripts/process-thought.js --thought "Initial analysis" --number 1 --total 5 --next true
|
|
|
|
# Process with revision
|
|
node scripts/process-thought.js --thought "Corrected analysis" --number 2 --total 5 --next true --revision 1
|
|
|
|
# Process with branching
|
|
node scripts/process-thought.js --thought "Branch A" --number 2 --total 5 --next true --branch 1 --branchId "branch-a"
|
|
|
|
# View history
|
|
node scripts/process-thought.js --history
|
|
|
|
# Reset history
|
|
node scripts/process-thought.js --reset
|
|
```
|
|
|
|
### Format Thought (Display)
|
|
|
|
```bash
|
|
# Box format (default)
|
|
node scripts/format-thought.js --thought "Analysis" --number 1 --total 5
|
|
|
|
# Simple text format
|
|
node scripts/format-thought.js --thought "Analysis" --number 1 --total 5 --format simple
|
|
|
|
# Markdown format
|
|
node scripts/format-thought.js --thought "Analysis" --number 1 --total 5 --format markdown
|
|
|
|
# With revision
|
|
node scripts/format-thought.js --thought "Revised" --number 2 --total 5 --revision 1
|
|
|
|
# With branch
|
|
node scripts/format-thought.js --thought "Branch" --number 2 --total 5 --branch 1 --branchId "a"
|
|
```
|
|
|
|
### Running Tests
|
|
|
|
```bash
|
|
# Install dependencies (first time only)
|
|
npm install
|
|
|
|
# Run all tests
|
|
npm test
|
|
|
|
# Run tests in watch mode
|
|
npm run test:watch
|
|
|
|
# Run with coverage
|
|
npm run test:coverage
|
|
```
|
|
|
|
## When to Use Scripts
|
|
|
|
**Use scripts when**:
|
|
- Need deterministic validation of thought structure
|
|
- Want persistent thought history tracking
|
|
- Require formatted output for documentation
|
|
- Building tools that integrate with sequential thinking
|
|
|
|
**Don't use scripts when**:
|
|
- Applying methodology directly in responses
|
|
- Want lightweight, inline thinking
|
|
- No need for validation or tracking
|
|
|
|
Scripts are **optional tooling** - the methodology can be applied without them.
|
|
|
|
## Source
|
|
|
|
Converted from: https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking
|
|
|
|
Original MCP server by Anthropic (MIT License).
|
|
Skill conversion:
|
|
- Extracts methodology as instructions
|
|
- Adds executable scripts for deterministic validation
|
|
- Makes tool-independent while preserving functionality
|