AI Coding Prompts

Prompts for implementing a Plan & Execute workflow in Cursor or similar AI coding tool, which is inspired by Cline's built-in Plan & Act modes.

Files

• planner.md - Planning mode prompt
• executor.md - Execution mode prompt
• isolated-executor.md - Isolated execution mode prompt

Setup in Cursor

Planner Mode

1. Create a new custom mode called "Planner"
2. Model: Gemini 2.5 Pro Max
3. Disable edit tools (uncheck "Edit" in tool settings)
4. Enable full folder context in settings
5. Add keybinding for quick access (e.g., ⌘⇧P)
6. Paste the planner.md prompt

Example plan.md file:

# Plan: Refactor User Validation

## Tasks
- [ ] 1.0 Update user schema to allow null values for optional fields
- [ ] 1.1 Add tests for null value handling
- [ ] 1.2 Update API endpoints to handle new validation rules

Usage

1. Switch to Planner mode
2. Include relevant files and folders in the context
3. Describe your idea and create a plan through iterative Q&A
4. Give the agent permission to output plan.md file
5. Create the plan.md file in the root of the project

Executor Mode

1. Create a new custom mode called "Executor"
2. Model: Claude Sonnet 4
3. Enable all tools
4. Add keybinding for quick access (e.g., ⌘⇧E)
5. Enable auto-apply edits, auto-run, and auto-fix errors
6. Paste the executor.md prompt

Usage

1. Switch to Executor mode
2. Add plan.md to the context
3. Type "go" to start execution
4. The executor will implement the task and commit the changes
5. Type "go" to continue to the next task until all tasks are complete

Isolated Executor

The isolated-executor.md prompt is a variant of the executor prompt that is used to run the executor agent in parallel in its own git worktree.

Setup

1. Create a new custom mode called "Isolated Executor"
2. Model: Claude Sonnet 4
3. Install mcp-text-editor MCP server
4. Enable all tools, including the new text edit MCP server
5. Disable native edit tools (uncheck "Edit" in tool settings)
6. Add keybinding for quick access (e.g., ⌘⇧I)
7. Paste the isolated-executor.md prompt

Usage

1. Open 2-3 new chat tabs in Cursor
2. In each tab, switch to Isolated Executor mode
3. Add plan.md to the context in each tab
4. Type the task number you want each agent to work on (e.g., "work on task 8" in tab 1, "work on task 12" in tab 2)
5. Each agent will work on their assigned task in separate worktrees
6. Agents will merge their changes back to the parent branch when complete
7. Monitor progress and assign new tasks as agents finish

Notes

• Make sure to adjust the plan along the way, when spotting missing details or constraints
• Executor prompt currently expects plan.md to be in the root of the project and included in .gitignore. You can change the prompt to also commit the plan.md file.
• Executor prompt expects yarn test for running tests. You can change it to your preferred test command.
• Feel free to adapt the prompts for your specific needs