Cursor Agent Development

Cursor Agent Development Workflow

This skill guides developers through effective use of cursor-agent (or agent command) for interactive development, project management, and implementing supervised development-verification loops.

Quick Start

# Interactive mode
agent

# Non-interactive mode
agent --print "your prompt"

# Create and resume session
CHAT_ID=$(agent create-chat)
echo $CHAT_ID > .chat_id
agent --resume $(cat .chat_id)

For detailed command reference, see reference/basic-commands.md.

Development-Verification Loop Pattern

Supervised development with quality gates:

Setup

1. Create two sessions (dev + QA)
2. Define requirements and acceptance criteria

Loop

1. Development: agent --resume $DEV_CHAT --print "开发任务" > output.py
2. Verification: agent --resume $QA_CHAT --print "验收检查" > 验收报告.md
3. Decision: Pass → End | Issues → Continue loop

For complete workflow details, see reference/development-verification-loop.md.

Session Management

• Create: CHAT_ID=$(agent create-chat)
• Resume: agent --resume <chatId> or agent resume
• List: agent ls

Sessions persist across terminals and days. For detailed session management, see reference/session-management.md.

Advanced Features

• Model selection: agent --model <model> "prompt"
• Output formats: --output-format json|stream-json
• Sandbox mode: --sandbox enabled|disabled
• Browser automation: --browser

For complete feature reference, see reference/advanced-features.md.

MCP Integration

Configure MCP servers in ~/.cursor/mcp.json or .cursor/mcp.json:

{
  "mcpServers": {
    "server-name": {
      "url": "server-url",
      "headers": {"API_KEY": "your-key"}
    }
  }
}

Manage with: agent mcp list|enable|disable|list-tools

For detailed MCP guide, see reference/mcp-integration.md.

Rules, Skills, and Subagents

Rules

Persistent context for coding standards. Location: .cursor/rules/ or ~/.cursor/rules/

Create with: agent rule or manually in .cursor/rules/<name>/RULE.md

Skills

Reusable knowledge modules. Location: .cursor/skills/ or ~/.cursor/skills/

Create in .cursor/skills/<name>/SKILL.md with YAML frontmatter.

Subagents

Specialized AI assistants. Location: .cursor/agents/ or ~/.cursor/agents/

Create as .md files with YAML frontmatter and system prompt.

AGENTS.md

Simple project instructions. Location: AGENTS.md (project root or subdirectories)

Plain Markdown, no YAML required. Auto-loaded by Cursor IDE.

Hooks

Intercept and modify agent behavior. Location: .cursor/hooks/ or ~/.cursor/hooks/

Hook types: beforeFileEdit, afterFileEdit, beforeShellCommand, afterShellCommand, stop

For detailed guides, see reference/rules-skills-subagents.md and reference/agents-md-hooks.md.

Best Practices

1. Clear prompts: Be specific with requirements
2. Iterative development: Break tasks into steps
3. Session persistence: Save IDs, resume across sessions
4. Separation: Use different sessions for dev and verification

For workflow patterns and examples, see reference/workflows-patterns.md.

Troubleshooting

• Session not found: Check ~/.cursor/chats/*/<chatId>/ or recreate
• Permission issues: Use --sandbox enabled or --force (careful)
• MCP issues: Check with agent mcp list and re-enable servers

Configuration

API Key

# Environment variable
export CURSOR_API_KEY="your-api-key"

# Command line
agent --api-key "your-api-key" "prompt"

# From .env file
set -a && source .env && set +a
export CURSOR_API_KEY="$OPENAI_API_KEY"  # If using OPENAI_API_KEY

Custom Base URL

# For OpenAI-compatible endpoints
export OPENAI_BASE_URL="https://api.your-endpoint.com/v1"
export OPENAI_API_KEY="your-api-key"

# Load from .env
set -a && source .env && set +a
export OPENAI_BASE_URL
export OPENAI_API_KEY

Note: cursor-agent primarily uses CURSOR_API_KEY. For custom endpoints, you may need to set both CURSOR_API_KEY and OPENAI_API_KEY.

For complete configuration guide with .env examples, see reference/configuration.md.

Reference Documentation

All detailed documentation is in the reference/ directory:

• basic-commands.md - Command reference
• configuration.md - API key and base URL configuration
• development-verification-loop.md - Dev-QA loop pattern
• session-management.md - Session operations
• advanced-features.md - Model selection, output formats, etc.
• mcp-integration.md - MCP server configuration
• rules-skills-subagents.md - Rules, Skills, Subagents
• agents-md-hooks.md - AGENTS.md and Hooks
• workflows-patterns.md - Common patterns and workflows

Key Takeaways

1. Sessions enable context continuity - Use for long-term projects
2. Development-Verification loops - Separate concerns for quality
3. Clear prompts - Specific requirements yield better results
4. Iterative approach - Build incrementally, verify continuously
5. Progressive disclosure - Core info in SKILL.md, details in reference/