Goal-driven autonomous development for Claude Code

AUTO-BUILDER is an open-source Claude Code skill with nine specialist subagents. Give /auto-builder a goal and it plans, builds, tests, independently reviews and documents the result, pausing for you only at consequential decisions.

git clone https://github.com/bsrikumar855-dot/ClaudeCODE-Autonomous-skills.git
cd ClaudeCODE-Autonomous-skills
./install.sh --personal

Then, inside Claude Code: /auto-builder Build a REST API for a habit tracker with tests

Project status: early release (v0.1.0). Static validation and installer/hook tests pass, but the skill has not yet been validated end-to-end in live Claude Code sessions, and the Windows installer has not been executed. Read the known limitations.

What is AUTO-BUILDER?

It is a reusable Claude Code skill plus Claude Code subagents that encode an AI software development workflow: discover the real problem, define testable requirements, design the architecture, plan small tasks, implement, verify, review and deliver. It targets one common failure of autonomous coding agents: jumping straight to code and declaring success after a single green build.

Goal in, plan first

Describe an outcome in plain language. The lead agent inspects your repository, writes requirements and a task list with acceptance criteria before editing code.

Specialist subagents

Nine roles from product manager to security engineer, each with a least-privilege tool list. Reviewers, QA and security have no edit tools.

Verified, not trusted

Work proceeds in milestones that each pass a verification gate. Checks must actually run, a separate reviewer inspects important milestones, and the report labels each criterion Verified, Partially verified or Untested.

Autonomy with checkpoints

Routine decisions are made for you. It asks before deploys, spending, destructive data operations, secrets, pushing, or changes of product direction.

The lifecycle

Reconnaissance → Discovery → Product definition → Architecture → Execution planning → Implementation → Independent verification → Iterate → Delivery

Each phase has a purpose and an exit condition; see the full lifecycle and the architecture.

Works for many project types

AI/ML applications, AI agents and multi-agent systems, full-stack apps, frontends, backend APIs, developer tools, research projects, automation and data engineering. The skill adapts its discovery, architecture and verification emphasis per category.

See example goals and a first-run tutorial →