Introduction

GET STARTED

Maverick AI is a vendor-neutral workflow for AI-assisted changes in existing repositories. It is not another coding agent: it is the structure around the agent — a PRD, a technical spec grounded in your code, a graph of tasks, recorded evidence and a human review decision — so a diff produced in minutes can still be defended in review.

demo — zsh

▍

Why it exists

An agent can produce a diff quickly. Engineering confidence comes from being able to answer, later, what was intended, what the agent was allowed to touch, how it was verified and who decided it was ready. Maverick keeps those answers as plain files next to your code.

Without: prompt → agent → diff → “looks good”
With:    idea → PRD → techspec → tasks → build → verify → review → PR

The CLI only starts the flow. Every stage then runs inside your AI coding agent as a skill (maverick-new, maverick-build, maverick-review, maverick-ship and six more), written in your project's docs language while the code keeps its own. See Everything happens in the agent.

When to use it

Use it for features, bug fixes, refactors, migrations and reviews where a human remains accountable. Every task follows one pipeline. Risky work gets the strict workflow, which adds a security gate before review: a one-line fix and a payment migration should not carry the same ceremony.

Your idea

What does it touch?

How risky is it?

WORKFLOW

Standard

PRD, technical spec and task graph, then build, verify and human review. Best for: features, fixes and refactors.

PRDTECHSPECTASKSBUILDVERIFYREVIEWPRDONE

Where to go next

Installation →Node 20+, run with npx or install globally.Your first task →From an idea to review-ready evidence.Maverick Lens →Run tasks, agents and gates, and explore the architecture, locally.CLI reference →Search every command, alias and usage.

Common mistake

Asking for implementation before the acceptance criteria exist. Write the PRD first — it is also what the reviewer reads.