taskie

Taskie

This is a simple framework of reusable prompts that makes it easy to use LLMs to implement features & projects of almost any level of complexity while maintaining a high level of code quality.

Works well with Anthropic Claude Sonnet 3.7, 4.0 and 4.5. Tested using the Cursor IDE (VSCode), Cursor Background Agents, Claude Code CLI and Claude Code Web.

Packaged and distributed as a Claude Code plugin. Also works with OpenAI Codex CLI and other LLM tools. For usage outside of Claude Code and Codex CLI, please refer to PROMPTS.md.

Heavily inspired by Taskmaster and wbern/claude-instructions.

Installation

For Claude Code

Add the Taskie marketplace

/plugin marketplace add ipletnjov-tw/taskie

Install the plugin

/plugin install taskie@taskie

Latest version: v2.3.0

For OpenAI Codex CLI

Run the installation script from the Taskie directory:

./install-codex.sh

This copies all prompts to ~/.codex/prompts/ with taskie- prefix. Restart Codex CLI or start a new session to load the prompts.

Codex CLI Technical Details

The Codex installation includes 19 files:

• 18 user-invocable prompts (taskie-new-plan.md, taskie-continue-plan.md, etc.)
• 1 shared ground rules file (taskie-ground-rules.md)

All prompts reference ~/.codex/prompts/taskie-ground-rules.md to load shared ground rules at runtime. This design:

• Maintains DRY: Ground rules exist in one place
• Enables updates: Change ground rules once, affects all prompts
• Reduces file size: Each prompt is ~90 lines smaller

Note: Task-specific workflow instructions (like the phases in complete-task prompts) remain inlined since they're unique to those workflows, not cross-cutting concerns.

CODEX_HOME Limitation: If using a custom CODEX_HOME environment variable, you must manually edit all prompt files to update the ground rules path from ~/.codex/prompts/taskie-ground-rules.md to your custom location.

Available Commands

Command Syntax by Tool

Tool	Syntax Pattern	Example
Claude Code	/taskie:command-name	/taskie:new-plan
Codex CLI	/prompts:taskie-command-name	/prompts:taskie-new-plan

Planning Commands

• new-plan - Create a new implementation plan
• continue-plan - Continue an existing plan from git history
• plan-review - Review and critique the current plan
• post-plan-review - Address plan review comments

Task Management Commands

• create-tasks - Generate tasks from the current plan
• add-task - Add a new task to an existing implementation plan
• tasks-review - Review the task list and task files
• post-tasks-review - Address task review comments
• next-task - Start implementing the next task
• continue-task - Continue working on the current task

Code Review Commands

• code-review - Critically review implemented code
• code-review-conventions - Code review focused on project conventions and patterns
• post-code-review - Apply code review feedback
• all-code-review - Review ALL code across ALL tasks in the plan
• post-all-code-review - Apply complete implementation review feedback

TDD Commands

• next-task-tdd - Implement next task using strict TDD (red-green-refactor)
• complete-task-tdd - TDD implementation with automatic review cycle

Unified Workflow Commands

• complete-task - Implement + review + fix in one command
• complete-task-tdd - TDD variant of complete-task

Usage

Claude Code

All commands support appending additional instructions. Most of the time, your prompts will look exactly like this:

/taskie:command-name

Or with additional context:

/taskie:command-name Additional instructions here

Codex CLI

For Codex CLI, use the /prompts:taskie- prefix:

/prompts:taskie-command-name

Or with additional instructions:

/prompts:taskie-command-name Additional instructions here

The workflow is identical to Claude Code. Examples below use Claude Code syntax; for Codex CLI, replace /taskie: with /prompts:taskie-.

Kick off a new implementation plan

Your first prompt should look like this:

/taskie:new-plan I need to implement feature X, it needs to be A, B, and C. You will have to use Y and please make sure you don't forget X.

You will want to read the generated plan and refine the details with the LLM. In addition, you can prompt /taskie:plan-review to get the LLM to critically review and scrutinize the plan.

After the review is written, use /taskie:post-plan-review for the LLM to address the review comments. You may want to add what review comments it should leave unaddressed, in case you disagree with any of them.