claude-code-system-prompts/architecture/1-high-level-system-architecture.md
2026-01-11 09:30:06 -07:00

78 lines
4.0 KiB
Markdown

# High-Level System Architecture
This diagram shows the overall structure of Claude Code's prompt system.
```mermaid
flowchart TB
subgraph Core["Core System"]
Main["<b>Main System Prompt</b><br/>Behavior, tone, policies,<br/>task handling"]
click Main href "../system-prompts/system-prompt-main-system-prompt.md" "Main System Prompt"
end
subgraph Conditional["Conditional Additions"]
Security["Security Policy"]
Learning["Learning Mode"]
Git["Git Status"]
Scratchpad["Scratchpad Directory"]
MCP["MCP CLI"]
Chrome["Chrome Browser"]
click Learning href "../system-prompts/system-prompt-learning-mode.md" "Learning Mode"
click Git href "../system-prompts/system-prompt-git-status.md" "Git Status"
click Scratchpad href "../system-prompts/system-prompt-scratchpad-directory.md" "Scratchpad Directory"
click MCP href "../system-prompts/system-prompt-mcp-cli.md" "MCP CLI"
click Chrome href "../system-prompts/system-prompt-claude-in-chrome-browser-automation.md" "Chrome Browser Automation"
end
subgraph Tools["Built-in Tools"]
direction LR
FileTools["File Operations<br/>Read, Write, Edit,<br/>Glob, Grep"]
ExecTools["Execution<br/>Bash"]
WebTools["Web<br/>Search, Fetch"]
PlanTools["Planning<br/>TodoWrite,<br/>Enter/Exit PlanMode"]
InteractTools["Interaction<br/>Task, Skill,<br/>AskUserQuestion"]
end
subgraph Agents["Sub-agents"]
direction LR
Explore["Explore<br/><i>Read-only search</i>"]
Plan["Plan<br/><i>Read-only design</i>"]
TaskAgent["Task Agent<br/><i>Autonomous work</i>"]
click Explore href "../system-prompts/agent-prompt-explore.md" "Explore Agent"
click Plan href "../system-prompts/agent-prompt-plan-mode-enhanced.md" "Plan Mode Agent"
click TaskAgent href "../system-prompts/agent-prompt-task-tool.md" "Task Tool Agent"
end
subgraph Utilities["Utility Agents"]
direction LR
Summary["Conversation<br/>Summarization"]
Session["Session Title<br/>& Branch Gen"]
Hooks["Hooks"]
Sentiment["User Sentiment<br/>Analysis"]
click Summary href "../system-prompts/agent-prompt-conversation-summarization.md" "Conversation Summarization"
click Session href "../system-prompts/agent-prompt-session-title-and-branch-generation.md" "Session Title Generation"
click Hooks href "../system-prompts/agent-prompt-agent-hook.md" "Agent Hook"
click Sentiment href "../system-prompts/agent-prompt-user-sentiment-analysis.md" "User Sentiment Analysis"
end
Main --> Conditional
Main --> Tools
Main --> Agents
Main --> Utilities
style Core fill:#e1f5fe
style Tools fill:#f3e5f5
style Agents fill:#e8f5e9
style Utilities fill:#fff3e0
style Conditional fill:#fce4ec
```
## Components
| Component | Description | Key Prompts |
|-----------|-------------|-------------|
| **Core System** | The main system prompt that defines Claude Code's behavior, tone, and policies | [Main System Prompt](../system-prompts/system-prompt-main-system-prompt.md) |
| **Conditional Additions** | Context-dependent prompt sections added based on environment and configuration | [Learning Mode](../system-prompts/system-prompt-learning-mode.md), [Git Status](../system-prompts/system-prompt-git-status.md), [MCP CLI](../system-prompts/system-prompt-mcp-cli.md) |
| **Built-in Tools** | Tool descriptions injected when tools are available | See [Tool Ecosystem](./3-tool-ecosystem.md) |
| **Sub-agents** | Specialized agents spawned for complex tasks | [Explore](../system-prompts/agent-prompt-explore.md), [Plan](../system-prompts/agent-prompt-plan-mode-enhanced.md), [Task](../system-prompts/agent-prompt-task-tool.md) |
| **Utility Agents** | Background agents for session management and analysis | [Summarization](../system-prompts/agent-prompt-conversation-summarization.md), [Session Title](../system-prompts/agent-prompt-session-title-and-branch-generation.md) |