ContextQMD
Back to Superclaude Framework

SuperClaude Context Architecture Guide

docs/developer-guide/technical-architecture.md

4.3.012.8 KB
Original Source

SuperClaude Context Architecture Guide

Overview

This guide documents how SuperClaude's Context-Oriented Configuration Framework is structured and how Claude Code interprets these context files to modify its behavior.

Important: SuperClaude is NOT standalone software with running processes, execution layers, or performance systems. It is a collection of .md instruction files that Claude Code reads to adopt specialized behaviors.

Table of Contents

  1. Context File Architecture
  2. The Import System
  3. Agent Context Structure
  4. Command Context Structure
  5. Mode Context Structure
  6. MCP Server Configuration
  7. How Claude Code Reads Context
  8. Extending the Framework

Context File Architecture

Directory Structure

~/.claude/ (SuperClaude Framework Files Only)
├── CLAUDE.md                       # Main context file with imports
├── FLAGS.md                        # Flag definitions and triggers
├── RULES.md                        # Core behavioral rules
├── PRINCIPLES.md                   # Guiding principles
├── ZIG.md                          # Zig language integration
├── MCP_Context7.md                 # Context7 MCP integration
├── MCP_Magic.md                    # Magic MCP integration
├── MCP_Morphllm.md                 # Morphllm MCP integration
├── MCP_Playwright.md               # Playwright MCP integration
├── MCP_Sequential.md               # Sequential MCP integration
├── MCP_Serena.md                   # Serena MCP integration
├── MCP_Tavily.md                   # Tavily MCP integration
├── MCP_Zig.md                      # Zig MCP integration
├── MODE_Brainstorming.md           # Collaborative discovery mode
├── MODE_Business_Panel.md          # Business expert panel mode
├── MODE_DeepResearch.md            # Deep research mode
├── MODE_Introspection.md           # Transparent reasoning mode
├── MODE_Orchestration.md           # Tool coordination mode
├── MODE_Task_Management.md         # Task orchestration mode
├── MODE_Token_Efficiency.md        # Compressed communication mode
├── agents/                         # Domain specialist contexts (19 total)
│   ├── backend-architect.md        # Backend expertise
│   ├── business-panel-experts.md   # Business strategy panel
│   ├── deep-research-agent.md      # Deep research expertise
│   ├── devops-architect.md         # DevOps expertise
│   ├── frontend-architect.md       # Frontend expertise
│   ├── learning-guide.md           # Educational expertise
│   ├── performance-engineer.md     # Performance expertise
│   ├── python-expert.md            # Python expertise
│   ├── quality-engineer.md         # Quality assurance expertise
│   ├── refactoring-expert.md       # Code quality expertise
│   ├── requirements-analyst.md     # Requirements expertise
│   ├── root-cause-analyst.md       # Problem diagnosis expertise
│   ├── security-engineer.md        # Security expertise
│   ├── socratic-mentor.md          # Educational expertise
│   ├── spec-panel-experts.md       # Specification review panel
│   ├── system-architect.md         # System design expertise
│   ├── technical-writer.md         # Documentation expertise
│   ├── test-runner.md              # Test execution expertise
│   └── wave-orchestrator.md        # Wave orchestration patterns
└── commands/                       # Workflow pattern contexts
    └── sc/                         # SuperClaude command namespace (25 total)
        ├── analyze.md              # Analysis patterns
        ├── brainstorm.md           # Discovery patterns
        ├── build.md                # Build patterns
        ├── business-panel.md       # Business expert panel patterns
        ├── cleanup.md              # Cleanup patterns
        ├── design.md               # Design patterns
        ├── document.md             # Documentation patterns
        ├── estimate.md             # Estimation patterns
        ├── explain.md              # Explanation patterns
        ├── git.md                  # Git workflow patterns
        ├── help.md                 # Help and command listing
        ├── implement.md            # Implementation patterns
        ├── improve.md              # Improvement patterns
        ├── index.md                # Index patterns
        ├── load.md                 # Context loading patterns
        ├── reflect.md              # Reflection patterns
        ├── research.md             # Deep research patterns
        ├── save.md                 # Session persistence patterns
        ├── select-tool.md          # Tool selection patterns
        ├── spawn.md                # Multi-agent patterns
        ├── spec-panel.md           # Specification review panel
        ├── task.md                 # Task management patterns
        ├── test.md                 # Testing patterns
        ├── troubleshoot.md         # Troubleshooting patterns
        └── workflow.md             # Workflow planning patterns

Note: Other directories (backups/, logs/, projects/, serena/, etc.) are Claude Code
operational directories, not part of SuperClaude framework content.

Context File Types

File TypePurposeActivationExample
CommandsDefine workflow patterns/sc:[command] (context trigger)User types /sc:implement → reads implement.md
AgentsProvide domain expertise@agent-[name] or auto@agent-security → reads security-engineer.md
ModesModify interaction styleFlags or triggers--brainstorm → activates brainstorming mode
CoreSet fundamental rulesAlways activeRULES.md always loaded

The Import System

How CLAUDE.md Works

The main CLAUDE.md file uses an import system to load multiple context files:

markdown
# CLAUDE

*MANDATORY*
@FLAGS.md                  # Flag definitions and triggers
@RULES.md                  # Core behavioral rules
@PRINCIPLES.md             # Guiding principles
*SECONDARY*
@MCP_Context7.md           # Context7 MCP integration
@MCP_Magic.md              # Magic MCP integration
@MCP_Morphllm.md           # Morphllm MCP integration
@MCP_Playwright.md         # Playwright MCP integration
@MCP_Sequential.md         # Sequential MCP integration
@MCP_Serena.md             # Serena MCP integration
@MCP_Tavily.md             # Tavily MCP integration
@MCP_Zig.md                # Zig MCP integration
*CRITICAL*
@MODE_Brainstorming.md     # Collaborative discovery mode
@MODE_Business_Panel.md    # Business expert panel mode
@MODE_DeepResearch.md      # Deep research mode
@MODE_Introspection.md     # Transparent reasoning mode
@MODE_Task_Management.md   # Task orchestration mode
@MODE_Orchestration.md     # Tool coordination mode
@MODE_Token_Efficiency.md  # Compressed communication mode
*LANGUAGE SPECIFIC*
@ZIG.md                    # Zig language integration

Import Processing

  1. Claude Code reads CLAUDE.md
  2. Encounters @import statements
  3. Loads referenced files into context
  4. Builds complete behavioral framework
  5. Applies relevant contexts based on user input

Agent Context Structure

Anatomy of an Agent File

Each agent .md file follows this structure:

markdown
---
name: agent-name
description: Brief description
category: specialized|architecture|quality
---

# Agent Name

## Triggers
- Keywords that activate this agent
- File types that trigger activation
- Complexity thresholds

## Behavioral Mindset
Core philosophy and approach

## Focus Areas
- Domain expertise area 1
- Domain expertise area 2

## Key Actions
1. Specific behavior pattern
2. Problem-solving approach

Agent Activation Logic

  • Manual: User types @agent-python-expert "task"
  • Automatic: Keywords in request trigger agent loading
  • Contextual: File types or patterns activate relevant agents

Command Context Structure

Anatomy of a Command File

markdown
---
name: command-name
description: Command purpose
category: utility|orchestration|analysis
complexity: basic|enhanced|advanced
mcp-servers: [context7, sequential]
personas: [architect, engineer]
---

# /sc:command-name

## Triggers
- When to use this command
- Context indicators

## Usage
/sc:command-name [target] [--options]

## Workflow Pattern
1. Step 1: Initial action
2. Step 2: Processing
3. Step 3: Validation

## Examples
Practical usage examples

Command Processing

When user types /sc:implement "feature" in Claude Code conversation:

  1. Claude reads commands/sc/implement.md
  2. Adopts implementation workflow pattern
  3. May auto-activate related agents
  4. Follows defined workflow steps

Mode Context Structure

Behavioral Modes

Modes modify Claude's interaction style:

markdown
# MODE_[Name].md

## Activation Triggers
- Flag: --mode-name
- Keywords: [triggers]
- Complexity: threshold

## Behavioral Modifications
- Communication style changes
- Decision-making adjustments
- Output format modifications

## Interaction Patterns
- How to respond
- What to prioritize

MCP Server Configuration

Configuration Location

MCP servers are configured in ~/.claude.json (NOT part of SuperClaude context):

json
{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"]
    },
    "sequential-thinking": {
      "command": "npx",
      "args": ["-y", "sequential-thinking-mcp@latest"]
    }
  }
}

MCP Integration

  • MCP Servers: Actual software providing tools
  • SuperClaude: Context that tells Claude when to use them
  • Activation: Flags or keywords trigger MCP usage

How Claude Code Reads Context

Context Loading Sequence

User Input (in Claude Code): "/sc:analyze src/ --focus security"
                    ↓
1. Parse Command: identify 'analyze' command
                    ↓
2. Load Context: read commands/sc/analyze.md
                    ↓
3. Check Flags: --focus security
                    ↓
4. Auto-Activation: load security-engineer.md
                    ↓
5. Apply Patterns: follow analysis workflow
                    ↓
6. Generate Output: using loaded contexts

Context Priority

  1. Explicit Commands: /sc: commands take precedence
  2. Manual Agents: @agent- override auto-activation
  3. Flags: Modify behavior of commands/agents
  4. Auto-Activation: Based on keywords/context
  5. Default Behavior: Standard Claude Code

Extending the Framework

Adding New Commands

  1. Create ~/.claude/commands/sc/new-command.md
  2. Define metadata, triggers, and workflow
  3. No code changes needed - just context

Adding New Agents

  1. Create ~/.claude/agents/new-specialist.md
  2. Define expertise, triggers, and behaviors
  3. Agent becomes available

Adding New Modes

  1. Create ~/.claude/MODE_NewMode.md
  2. Define activation triggers and modifications
  3. Mode activates based on triggers

Best Practices

  • Keep Context Focused: One concept per file
  • Clear Triggers: Define when context activates
  • Workflow Patterns: Provide step-by-step guidance
  • Examples: Include practical usage examples
  • Metadata: Use frontmatter for configuration

Important Clarifications

What SuperClaude Is NOT

  • ❌ No Execution Engine: No code runs, no processes execute
  • ❌ No Performance System: No optimization possible (it's just text)
  • ❌ No Detection Engine: Claude Code does pattern matching
  • ❌ No Orchestration Layer: Context files guide, not control
  • ❌ No Quality Gates: Just instructional patterns

What SuperClaude IS

  • ✅ Context Files: .md instructions for Claude Code
  • ✅ Behavioral Patterns: Workflows and approaches
  • ✅ Domain Expertise: Specialized knowledge contexts
  • ✅ Configuration: Settings for actual tools (MCP)
  • ✅ Framework: Structured prompt engineering

Summary

SuperClaude's architecture is intentionally simple: it's a well-organized collection of context files that Claude Code reads to modify its behavior. The power comes from the careful crafting of these contexts and their systematic organization, not from any executing code or running processes.

The framework's elegance lies in its simplicity - by providing Claude Code with structured instructions through context files, we can achieve sophisticated behavioral modifications without any software complexity.