Skip to content

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AI Agent Harness Framework

A documentation-first development framework designed as a Claude Code plugin that implements a spec-feature-task approach for systematic AI-assisted software development.

Overview

This framework provides a structured methodology for planning, implementing, and tracking software projects with AI assistance. It emphasizes documentation-driven development where specifications guide implementation through clearly defined features and tasks.

Key Features

  • Spec-Feature-Task Architecture: Three-tier hierarchy for organizing project requirements
  • Documentation-First Workflow: All development starts with comprehensive documentation
  • Claude Code Integration: Works seamlessly as a Claude Code plugin (/df:* commands)
  • Automated Task Management: Pick and implement tasks systematically
  • Built-in Project Scaffolding: Full Next.js 15 + TypeScript + Prisma + SQLite starter

Architecture

PROJECT (docs/project-spec.md)
└── FEATURES (docs/features/<name>/)
    ├── CLAUDE.md (AI context)
    ├── spec.md (detailed specification)
    └── tasks.md (task tracking)
        └── TASKS (individual work items)
            ├── Status: not started | in progress | completed | blocked | wont do
            ├── Acceptance criteria
            └── Result documentation

Available Skills

Core Skills (Implemented)

  • /df:init - Initialize new project with complete documentation structure
  • /df:init-nextjs - Scaffold Next.js 15 application with TypeScript, Prisma, SQLite, and shadcn/ui
  • /df:work - Pick and implement the next available task

Planned Skills

  • /df:audit - Security analysis and vulnerability scanning
  • /df:review - Code quality review and recommendations
  • /df:commit - Automated commit with conventional commit messages

Workflow

  1. Initialize Project: /df:init conducts discovery and creates documentation
  2. Scaffold Code: /df:init-nextjs creates the application structure
  3. Implement Features: /df:work systematically works through tasks
  4. Track Progress: Tasks automatically update with status and results

Project Structure

/
├── .claude-plugin/          # Plugin configuration
├── rules/                   # Automated documentation rules
│   ├── documentation.md     # Enforce documentation standards
│   └── post-implementation.md # Update docs after code changes
├── skills/                  # Framework capabilities
│   ├── init/               # Project initialization
│   ├── init-nextjs/        # Next.js scaffolding
│   ├── work/               # Task implementation
│   ├── audit/              # Security audit (planned)
│   ├── review/             # Code review (planned)
│   └── commit/             # Commit workflow (planned)
└── templates/              # Reusable documentation templates
    ├── project-spec.md     # Project specification
    ├── feature-spec.md     # Feature specification
    ├── feature-claude-md.md # AI context template
    └── tasks.md            # Task tracking template

Documentation Standards

The framework enforces strict documentation standards through automated rules:

  • Task Statuses: Must be one of: not started, in progress, completed, blocked, wont do
  • Acceptance Criteria: Required for all tasks
  • Result Documentation: Updated when tasks are completed
  • Context Files: CLAUDE.md files provide AI-optimized project context

Technology Stack (Generated Projects)

  • Frontend: Next.js 15 (App Router), React, TypeScript
  • Styling: Tailwind CSS, shadcn/ui components
  • Database: Prisma ORM with SQLite
  • Testing: Script-first testing approach
  • Package Manager: Yarn

Getting Started

  1. Install as a Claude Code plugin
  2. Run /df:init to start a new project
  3. Answer discovery questions about your project
  4. Run /df:init-nextjs to scaffold the code
  5. Run /df:work repeatedly to implement features

Design Principles

  • Documentation-First: Specs drive implementation
  • Transparency: All work tracked in plain text Markdown
  • AI-Optimized: Context files and criteria designed for autonomous work
  • Idempotent: Skills can be re-run safely
  • Single Responsibility: Each skill handles one workflow

License

[License information to be added]

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors