Ultra Claude

A Claude Code plugin that implements spec-driven development. Documentation governs code growth. Agent teams coordinate execution.

Get Started Documentation →
Not a framework, library, or runtime

What This Is

Ultra Claude is a portable, reusable Claude Code plugin that turns any project into a structured, specification-driven development environment. You install it, it establishes a documentation layer, and from that point forward — documentation controls how the codebase grows.

It is a collection of skills and agents packaged as a Claude Code plugin.

Capabilities

Planning Modes

Three specialized modes — Feature, Debug, and Verification — each produce structured plans with end-to-end testable tasks and approval workflows. Optional Discovery mode for research-only work.

👥

Agent Teams

Executor, Code Reviewer, and Tester coordinate in parallel per task, with a shared Tech Knowledge agent for external library docs and Explore agents for codebase investigation.

Doc-Code Verification

Detects drift between documentation and code. Surveyor and checker agents compare claims against reality.

💾

Checkpoint Recovery

Plan management with state snapshots. Sessions can die and resume — task lists, shared memory, and research files survive.

📚

Spec-Driven Development

Code is a derived artifact. Architecture docs exist before code. When code diverges from specs, you fix the spec first.

Core Philosophy

Code is a derived artifact

The specification is the source of truth. Architecture docs exist before code is written, and when code diverges from specs, you fix the spec first.

Governance, not bureaucracy

Documentation acts like zoning laws — you build freely within the constraints, but the constraints control direction. Three enforcement levels: soft, medium, hard.

Proportional scrutiny

Additive changes flow freely. Compatible changes get lightweight review. Breaking changes require updating the architecture doc first.

Documentation
System Overview
Interactive diagram of the full system
Architecture
Philosophy, layers, modes, governance
Execution
Agent teams, task lists, shared memory
Components
Skills, agents, templates, config
Workflows
Step-by-step flows for each mode
Decisions
Architectural decisions + open questions