cursor-ide-agent_skill

Overview

This skill exposes Cursor Agent usage in two practical paths: a Local CLI for fast, general-purpose coding and a VS Code Node path for full IDE intelligence via the OpenClaw Node protocol. The CLI is the default for speed and broad tasks; use the Node path when you need diagnostics, references, debugging, or precise workspace context. It provides invocation patterns, session controls, and examples to integrate Cursor into automation and remote IDE workflows.

How this skill works

The CLI runs Cursor Agent locally as an interactive TUI or in non-interactive automation modes (plan, ask, agent) and supports session management, sandboxing, and cloud handoff. The Node path uses the OpenClaw Node protocol to control a remote Cursor/VS Code instance and exposes commands for file operations, language features, diagnostics, tests, git, and debugging. Use CLI for quick edits and automation; switch to Node for IDE-level operations like diagnostics.get, lang.definition, test.run, and vscode.debug.*.

When to use it

• Quick bug fixes, refactors, or one-off code generation — prefer CLI for speed
• Automated, non-interactive tasks (CI steps, scripts) — use CLI with -p and JSON output
• When you need real type/lint diagnostics, go to the Node path
• Navigating definitions, references, and precise renames — use Node
• Running tests, inspecting results, or debugging with breakpoints — Node is required

Best practices

• Default to CLI for most everyday tasks; reserve Node for workspace-specific intelligence
• Run Node diagnostics.get before edits to surface real type and lint errors
• Use pty:true when invoking the CLI from exec to avoid hangs; use background + poll for long jobs
• Use non-interactive CLI flags (--plan, -p, --output-format json) for automation and parsing
• Follow the Fix → Verify → Commit loop on Node: diagnostics → agent.run → diagnostics → test.run → git.commit

Example use cases

• Locally generate a new feature and commit it quickly with agent -p and --force (CLI)
• Detect and fix TypeScript errors using vscode.diagnostics.get then vscode.agent.run (Node)
• Run the project's test suite through the IDE and iterate until tests pass (Node: test.run + test.results)
• Automate scans for TODOs or unused imports and produce machine-readable output (CLI with JSON)
• Remote-edit a repository through a connected Cursor/VS Code node, then stage and commit changes (Node git commands)