cursor-agent-development_skill

Overview

This skill guides developers through using the cursor-agent CLI for interactive development, session management, and supervised development-verification loops. It explains core commands, session patterns, configuration options, and how to structure agent-led projects for predictable outcomes. It focuses on practical steps to create, resume, and manage chat sessions and to implement dev/QA workflows.

How this skill works

The skill inspects common cursor-agent commands and workflow patterns and explains how to use them to create persistent chat sessions, run interactive or non-interactive prompts, and switch models or output formats. It covers session lifecycle commands (create, resume, list), advanced flags (model, sandbox, output-format, browser), and integration points such as MCP servers, rules, skills, subagents, hooks, and AGENTS.md. It shows how to wire API keys and custom base URLs for agent connectivity.

When to use it

• When starting an interactive agent-driven coding session or CLI experiment
• When you need persistent context across terminals and days for a project
• When implementing a supervised development-verification (dev/QA) loop
• When integrating external MCP servers or custom model endpoints
• When creating reusable skills, rules, or subagents for project automation

Best practices

• Keep prompts explicit and include acceptance criteria for QA sessions
• Use separate sessions for development and verification to isolate concerns
• Persist chat IDs in files to resume work across machines and terminals
• Iterate in small steps and verify after each change with a QA session
• Use sandbox mode for risky operations and enable force only when necessary

Example use cases

• Generate a feature implementation in a dev session, write tests or checks in a QA session, and loop until acceptance criteria pass
• Create reusable skills describing company standards and expose them to subagents for consistent behavior
• Automate browser-driven tasks using the --browser flag combined with session-managed scripts
• Configure MCP servers to route tasks to specialized tooling or internal models
• Switch models or output formats to produce JSON payloads or streamed outputs for downstream pipelines