Claude Code Prompts — Template Categories
System Prompts (1 template)
The core agent identity prompt that defines:
- Agent role and behavioral rules
- Safety constraints and risk assessment
- Permission boundaries
- Communication style
Tool Prompts (11 templates)
| Template | Purpose |
|---|---|
| Shell Execution | Safe command execution with sandboxing rules |
| File Read | Context-aware file reading with line ranges |
| File Write | New file creation with safety checks |
| File Edit | Surgical string replacement in existing files |
| Search (Glob) | Fast pattern-based file discovery |
| Search (Grep) | Content search with regex support |
| Web Fetch | URL content retrieval and processing |
| Web Search | Internet search with source attribution |
| Agent Launch | Spawning sub-agents for parallel tasks |
| User Query | Interactive question-asking with context |
| Planning Mode | Task decomposition and plan approval |
Agent Prompts (5 templates)
| Template | Purpose |
|---|---|
| General Purpose | Default agent for multi-step tasks |
| Code Explorer | Fast codebase navigation and search |
| Architecture Planner | System design and implementation planning |
| Verification Tester | Testing and quality assurance |
| Documentation Agent | Code documentation generation |
Memory Prompts (4 templates)
| Template | Purpose |
|---|---|
| Conversation Summarizer | Compress long conversations |
| Session Tracker | Track progress across context windows |
| Memory Extractor | Extract persistent learnings |
| Memory Consolidator | Merge and deduplicate memories |
Coordination (1 template)
Multi-agent orchestrator that manages task delegation, progress tracking, and result synthesis across parallel agent teams.
Utility Prompts (4 templates)
Session titling, action summaries, conversation recaps, and next-step suggestions.
Design Principles
The templates follow key architectural patterns:
- Safety-First — Every tool prompt includes risk assessment and confirmation rules
- Minimal Context — Prompts are optimized to use minimum tokens while maintaining behavior
- Composable — Mix and match templates to build custom agent configurations
- Observable — Built-in action logging and explanation patterns
FAQ
Q: What is Claude Code Prompts? A: A collection of 26 independently authored prompt templates that recreate the architectural patterns of AI coding agents — system prompts, tool routing, agent delegation, memory management, and multi-agent coordination.
Q: Is this the same as leaked system prompts? A: No. These are clean-room recreations — original implementations inspired by observed behavior patterns, designed to be understood and modified. Not copies of proprietary prompts.
Q: Can I use these in my own agent? A: Yes, MIT license. Copy, modify, and use in any project.