system-prompts-and-models-of-ai-tools/Cursor Prompts

Cursor AI

Type: IDE (VS Code Fork)
Availability: Free + Pro ($20/month) + Business ($40/user/month)
Website: https://cursor.com
GitHub: cursor-ai/cursor

---

📋 Overview

Cursor is an AI-first code editor built as a fork of VS Code. It features:

• Multi-file editing with AI
• Codebase-aware chat
• Inline code generation
• Terminal integration
• Memory system for persistent context
• Multiple specialized modes (Chat, Agent, Composer)

---

📂 Files in This Directory

System Prompts:

• Prompt.txt - Base chat prompt (generic)
• Chat Prompt.txt - Chat mode system instructions
• Agent Prompt.txt - Agent mode (autonomous task executor)
• Agent Prompt v1.0.txt - First agent mode version
• Agent Prompt v1.2.txt - Updated agent with improvements
• Agent Prompt 2025-09-03.txt - September 2025 iteration
• Agent CLI Prompt 2025-08-07.txt - CLI-optimized variant (August 2025)
• Memory Prompt.txt - Memory system instructions
• Memory Rating Prompt.txt - Memory importance scoring

Tools:

• Agent Tools v1.0.json - Tool definitions for agent mode

---

🔍 Source

• Official Documentation: https://docs.cursor.com
• Public Sources: Community reverse engineering
• Date Captured: Multiple versions from July 2024 - September 2025
• Attribution: Cursor AI, Inc.

---

📊 Key Features in Prompts

1. Conciseness Mandate

Evolution visible across versions:

• v1.0: "Provide helpful explanations"
• v1.2: "Keep your answers short and impersonal"
• CLI 2025-08: "You MUST answer concisely with fewer than 4 lines"

2. Multi-Agent Architecture

Agent mode coordinates between:

• Main agent (planning, coordination)
• Task executors (implementation)
• Search agents (codebase discovery)

3. Memory System

Two-tier memory:

• Short-term: Conversation context
• Long-term: Persistent memories with importance ratings

4. Parallel Execution

Strongly emphasized in later versions:

"Default to PARALLEL for all independent work. Call multiple tools simultaneously."

5. Verification Gates

Built-in quality checks:

• Read before edit
• Verify after changes
• Check for errors
• Test when applicable

---

📈 Evolution Highlights

v1.0 → v1.2 (Major Changes):

1. Added memory system
2. Introduced agent mode
3. More concise communication
4. Better tool organization
5. Parallel execution emphasis

v1.2 → Agent CLI 2025-08 (Refinements):

1. Terminal-specific optimizations
2. Even stricter conciseness
3. Structured tool usage patterns
4. Background process handling
5. CLI workflow awareness

Key Trend:

📉 Verbose explanations → 📈 Terse, action-focused responses

---

🎯 Unique Patterns

1. AGENTS.md Pattern

Cursor pioneered per-project context files:

AGENTS.md contains:
- Common commands (npm test, npm build)
- Code style preferences
- Project structure notes
- Custom instructions

2. Memory Rating System

Assigns importance scores (1-10) to memories:

• 1-3: Trivial (discard)
• 4-6: Moderate (keep briefly)
• 7-10: Important (persist)

3. No Comments Philosophy

Strong stance against code comments:

"IMPORTANT: DO NOT ADD ANY COMMENTS unless asked."

---

🛠️ Tool Architecture

Core Tools:

• File Operations: read_file, write_file, edit_file
• Search: grep_search, file_search, semantic_search
• Execution: run_in_terminal (with background mode)
• Analysis: get_errors, get_diagnostics
• Memory: memory_write, memory_read
• Git: Via terminal commands
• Sub-agents: Task delegation, search agents

Tool Evolution:

• v1.0: ~10 basic tools
• v1.2: ~15 tools + sub-agents
• 2025: 20+ tools + multi-agent orchestration

---

🔐 Security Features

Standard security in all versions:

• Never log secrets or API keys
• Validate file paths
• Warn before destructive operations
• Explain non-trivial bash commands
• Check for secrets before git commits

---

💡 Best Practices Extracted

From Cursor Prompts:

1. Be Concise: Minimize token usage, respect user time
2. Parallel by Default: Independent operations run simultaneously
3. Verify Changes: Always check after editing
4. No Comments: AI explanations belong in chat, not code
5. Use Memory: Persist important context across sessions
6. Read Before Edit: Understand before modifying
7. Fail Fast: Stop after 3 failed attempts, ask user

---

📊 Comparison to Other Tools

vs. GitHub Copilot:

• Cursor: Full IDE with multi-file editing, agent mode
• Copilot: Plugin for existing IDEs, inline completions

vs. Windsurf:

• Cursor: More mature, larger user base
• Windsurf: Cascade architecture, newer patterns

vs. Claude Code:

• Cursor: Visual IDE with GUI
• Claude Code: Terminal-focused, CLI tool

---

🎓 Learning Resources

From Cursor Prompts:

• Context Management: AGENTS.md pattern for project-specific context
• Memory Systems: Importance-based persistence
• Agent Architecture: Coordination + delegation patterns
• Tool Design: Parallel-first, verification gates
• Communication: Concise, action-oriented

Use Cases:

1. Study conciseness evolution (v1.0 → CLI 2025)
2. Learn memory system design (Memory Prompt.txt)
3. Understand agent architectures (Agent Prompt.txt)
4. See tool orchestration (Agent Tools v1.0.json)

---

🔄 Version Comparison

Feature	v1.0	v1.2	CLI 2025
Response Length	Moderate	Short	Very Short
Agent Mode	❌	✅	✅
Memory System	❌	✅	✅
Parallel Execution	Implicit	Explicit	Emphasized
Tool Count	~10	~15	~20
Sub-agents	❌	✅	✅
AGENTS.md Support	❌	✅	✅

---

🎯 Key Takeaways

What Makes Cursor Unique:

1. AI-first IDE (not just a plugin)
2. Mature agent mode with multi-agent orchestration
3. Memory system for context persistence
4. Rapid iteration visible through versions
5. Strong emphasis on speed (conciseness, parallel execution)

Prompt Engineering Insights:

• Conciseness is a learned optimization (got stricter over time)
• Memory systems require importance scoring to scale
• Agent modes need clear delegation patterns
• Tool design matters: parallel-first architecture
• User feedback drives evolution (comments → no comments)

---

📅 Version History

Version	Date	Major Changes
v1.0	July 2024	Initial agent mode
v1.2	August 2024	Memory system, refinements
CLI 2025-08	August 2025	Terminal optimizations
2025-09	September 2025	Latest iterations

---

🔗 Related Tools

Similar Architecture:

• Windsurf - Cascade architecture, similar multi-agent design
• Amp - Oracle + Executor separation
• Claude Code - Terminal-focused, concise prompts

Complementary:

• GitHub Copilot - Can use alongside Cursor
• Tabnine - Alternative autocomplete

---

📚 Academic Interest

Research Topics:

1. Multi-agent coordination in coding assistants
2. Memory system design for AI assistants
3. Evolution of conciseness in AI prompts (token economics)
4. Tool architecture for parallel execution
5. Context management strategies (AGENTS.md pattern)

Relevant Papers:

• "Multi-Agent Systems for Code Generation" (cite if available)
• "Context Management in Large Language Models"
• "Tool Use in AI Assistants"

---

🤝 Community

• Discord: https://discord.gg/cursor
• Forum: https://forum.cursor.com
• Twitter: @cursor_ai
• Documentation: https://docs.cursor.com

---

⚖️ License

Cursor is proprietary software. These prompt files are documented for:

• Educational purposes
• Research
• Understanding AI assistant design
• Comparative analysis

Source Attribution: Cursor AI, Inc.
Documentation Purpose: Fair use for education and research

---

🙏 Acknowledgments

• Cursor Team for building an innovative AI-first IDE
• Community for sharing insights and reverse-engineering efforts
• Contributors to this repository

---

Last Updated: 2025-01-02
Cursor Version: Latest as of September 2025
README Version: 1.0