agentic-coding

Agentic Coding CLI

A CLI tool to set up Cursor rules and tasks for AI-augmented engineering workflows

---

Overview

This CLI tool helps you quickly set up a comprehensive set of Cursor rules and task templates for agentic coding. Instead of manually creating rules files, simply run one command to copy a battle-tested collection of rules, actions, guides, tools, and task templates into your project.

The rules work as part of a complete agentic coding system:

• Rules (installed by this CLI) teach the AI how to build software
• Tasks (YAML files) tell the AI what to build next
• Code (your project) gets created and modified through this workflow

This creates a virtuous loop: Ideas become Tasks → Tasks invoke Rules → Rules change Code → passing code closes Tasks.

Installation

npm install -g agentic-coding

Or clone and link locally:

git clone <your-repo>
cd agentic-coding
npm link

Usage

Navigate to any project directory and run:

agentic-coding setup

This will copy all rules and task templates to your current directory, creating:

your-project/
├── .cursor/
│   └── rules/
│       ├── actions/     # Step-by-step playbooks
│       ├── behaviours/  # Always-on guardrails  
│       ├── guides/      # Deep-dive reference docs
│       └── tools/       # External CLI/API cheat-sheets
└── .tasks/
    ├── 0-draft/         # Ideas being refined
    ├── now/             # Current sprint work  
    ├── next/            # Prioritized backlog
    ├── later/           # Future considerations
    ├── done/            # Completed tasks
    └── README.md        # Task system documentation

What's Included

Actions

Step-by-step playbooks that turn chat commands into deterministic procedures:

• Task execution workflows
• Debugging procedures
• Code review checklists

Behaviours

Always-on guardrails that keep AI output consistent:

• Code style and naming conventions
• Tone and communication guidelines
• Security best practices

Guides

Deep-dive reference documentation:

• Library usage policies
• Architecture decisions
• Security guidelines

Tools

Cheat-sheets for external CLIs and APIs:

• Common command patterns
• Configuration templates
• Integration guides

Tasks

Structured YAML templates for work organization:

• Folder structure for task lifecycle (draft → now → next → later → done)
• YAML schema for executable specifications
• Context-aware task definitions with acceptance criteria

Rule System Architecture

The rules are designed to work with Cursor's context system:

• Agent-Requested: AI decides when to include based on relevance
• Always: Universal guidance that applies to every interaction
• Auto-Attached: Path-specific rules that load automatically

YAML Task Board – Work Items That Explain Themselves

The rules work best when paired with a structured task system. A task is both a ticket and an executable spec:

id: FEAT-102           # Immutable slug
context:               # ← Cursor loads these first (critical!)
  code:
    - src/features/workflowDesigner/…
  docs:
    - .cursor/guides/ui/confirmations.mdc
work_steps:            # Ordered, file-level instructions
  - step_id: 1
    targets:
      - src/…
    instructions: Create DeleteConfirmationDialog…
acceptance:            # Proof that we're done
  automated:
    - command: "npm run test --silent"
      description: All unit tests green
  manual:
    - description: Dialog matches design in Figma link
self_checklist:        # "Did I …?" reminders
  - "No direct DOM manipulation"

Because the schema is explicit, Cursor can:

1. Load context before editing (no blind changes)
2. Execute work_steps sequentially via edit_file, run_terminal_cmd, etc.
3. Prove success by running the same commands you would script in CI

Task Lifecycle

Organize tasks in your project using a simple folder structure:

.tasks/
├── 0-draft/     # Ideas being refined
├── now/         # Current sprint work
├── next/        # Prioritized backlog
├── later/       # Future considerations
└── done/        # Completed tasks

Or simply update the status: field in each YAML file.

A Day in the Life

1. Idea appears in chat: "We should support dark mode"
2. Dev runs /Generate Task → CREATE_TASK interviews them and emits THEM-201-dark-mode.yml into 0-draft/
3. Team refines acceptance criteria; file moves to next/
4. Sprint starts. Agent (or dev) opens task and says "execute" → EXECUTE_TASK loads context, edits code, runs tests, ticks checklist
5. All checks pass, PR auto-opens with the diff; task file lands in done/
6. Any new knowledge becomes a Guide so the next dev never repeats the research

Ideas become Tasks → Tasks invoke Rules → Rules change Code → passing code closes Tasks – a virtuous loop.

Development

To modify or extend the rules:

1. Edit files in the rules/ directory
2. Test with agentic-coding setup in a test project
3. Commit changes to version control

Contributing

1. Fork the repository
2. Create a feature branch
3. Add or modify rules in the appropriate category
4. Test the CLI tool
5. Submit a pull request

---

Philosophy: Process should accelerate engineers, not handcuff them. These rules turn tacit knowledge into executable guidance, letting humans and AI collaborate more effectively.