flower

CLI tool that scaffolds structured development workflows in any project using AI agents.

flower is a CLI (@flowrr/cli) that copies workflow skills to a target project's .agents/skills/ directory. These skills guide AI agents through a structured, software engineering-based development process: propose → design → plan → implement → verify → review.

Features

• ✅ Standardized AI Workflows: Brings structured software engineering practices to AI coding agents.
• ✅ Phase-by-Phase Execution: Breaks down complex tasks into propose, design, plan, implement, verify, and review phases.
• ✅ Local Skill Management: Installs standard workflow skills directly into your project's .agents/skills/ folder.
• ✅ Project Agnostic: Works with any codebase or language.

Quick Start

Prerequisites

• Node.js 18+
• An AI coding assistant that supports the agentskills.io specification (like Crush).

Installation & Setup

Install the CLI globally or run it via npx to scaffold the workflow skills into your current project:

# Initialize flower skills in your current project
npx @flowrr/cli setup

This will create an .agents/skills/ directory in your project containing the necessary workflow skills.

The Workflow

The flower workflow consists of six sequential skills designed to produce high-quality, verified code.

| Phase | Skill | Purpose | Output | | ----- | ------------------ | -------------------------------------------------------------- | ---------------- | | 1 | flower-propose | Capture and clarify requirements from user input | requirement.md | | 2 | flower-design | Create technical architecture and approach | design.md | | 3 | flower-plan | Break down work into ordered tasks with acceptance criteria | plan.md | | 4 | flower-implement | Execute tasks from the plan and write code | Code changes | | 5 | flower-verify | Verify implementation completeness, correctness, and coherence | verify.md | | 6 | flower-review | Review code quality, security, and project conventions | review.md |

Note: All output markdown files are generated in .agents/flower/{YYMMDD-HHMM}--{short-desc}/.

Basic Usage

Once installed, you can trigger the workflow by interacting with your AI agent using the trigger keywords defined in each skill.

1. Start a new task: Ask your agent: "I want to add user authentication. Let's start the proposal phase." (Triggers flower-propose)

2. Generate a design (for complex tasks): Ask your agent: "Create a technical design for the authentication requirement." (Triggers flower-design)

3. Plan the implementation: Ask your agent: "Create an implementation plan based on the design." (Triggers flower-plan)

4. Write the code: Ask your agent: "Implement the tasks from the plan." (Triggers flower-implement)

5. Verify and Review: Ask your agent: "Verify the implementation." and then "Review the code quality."

Architecture

The project is structured as a monorepo containing the CLI tool and the source of truth for the skills.

flower/
├── packages/cli/              # CLI package (@flowrr/cli)
│   ├── src/                   # CLI source code (citty framework)
│   └── skills/                # Bundled skills (copied during build)
└── skills/                    # Source skills (source of truth)
    ├── flower-propose/
    ├── flower-design/
    ├── flower-plan/
    ├── flower-implement/
    ├── flower-verify/
    └── flower-review/

CLI Commands

• setup: Copies the flower skills into your project's .agents/skills/ directory. Use -f or --force to overwrite existing skills.
• doctor: Checks your project to ensure skills are correctly installed and haven't been maliciously modified.

# Check installation health
npx @flowrr/cli doctor

Troubleshooting

Common Issues

Error: Skills directory already exists

If you try to run setup but the .agents/skills directory already contains files, the CLI will warn you to prevent overwriting custom changes. Fix: Run the setup command with the force flag:

npx @flowrr/cli setup --force

Doctor Command Warnings

If doctor reports modified files or missing SKILL.md files, it means the installed skills differ from the official package. Fix: Re-run the setup command with --force to restore the official skills.

Development

See AGENTS.md for detailed development workflow, build processes, and architecture documentation.

1. Source of Truth: Always modify skills in the root /skills/ directory.
2. Testing: Run bun run dev in packages/cli to test CLI changes locally.
3. Building: The CLI is built targeting Node.js using bun run build.

License

MIT