← Back to research
•·3 min read·company

AB Method

AB Method — a workflow system for Claude Code and Codex that grills problems into domain-grounded plans, then executes them as reviewable TDD missions or an autonomous /goal loop. 174 stars, JavaScript, MIT, revived with a v3 rewrite in April 2026.

https://github.com/ayoubben18/ab-method ↗
coding-agentsagent-skillsopen-sourcejavascriptgithubclaude-code

Key takeaways

  • Repositioned in 2026 from a Claude Code-only methodology to a dual-target workflow system: the npx installer now detects Claude Code (.claude/) and Codex (.agents/) and installs identical workflows for both
  • Went dormant from November 2025 to April 2026, then revived with a v3.0.0 rewrite — skill-driven workflows, a domain model, TDD-first missions, and a new autonomous /goal loop that runs to a verifiable stop condition
  • 174 stars, 14 forks, 0 open issues, MIT, JavaScript as of June 11, 2026; ~750 npm downloads/month — a small but actively maintained solo project, not a community movement
  • Mission-based decomposition remains the core idea — small, independently verifiable units, now with red-green-refactor TDD baked into every mission and opt-in parallel mission groups as of v3.3.0 (June 2026)

FAQ

What is the AB Method?

A workflow system for Claude Code and Codex that turns a problem into a domain-grounded plan, then executes it either as test-driven missions you review one at a time or via an autonomous /goal loop that runs to a verifiable stop condition.

Is the AB Method free?

Yes. It is MIT-licensed open source, installed via `npx ab-method`. You pay only for the underlying agent (Claude Code or Codex) you run it with.

Is the AB Method still maintained?

Yes, after a scare. It had no commits from November 2025 to April 2026, then shipped a v3.0.0 rewrite on April 25, 2026 and three more releases through v3.3.0 on June 6, 2026.

Does the AB Method work with agents other than Claude Code?

As of v3.0.0 (April 2026) it supports both Claude Code (slash commands) and OpenAI's Codex (skill-based invocation) with identical workflows; v3.1.0 made Codex support first-class.

Overview

The AB Method is an opinionated development workflow system, originally built for Claude Code's multi-agent capabilities and — since its April 2026 v3 rewrite — also targeting OpenAI's Codex. It grills a problem into a domain-grounded plan, then executes it one of two ways: test-driven missions you review one at a time, or an autonomous /goal loop that runs to a verifiable stop condition.

Key stats (as of June 11, 2026): 174 stars, 14 forks, 0 open issues, MIT license, JavaScript. Created August 2025; latest release v3.3.0 on June 6, 2026. The npm package saw roughly 750 downloads in the month ending June 2, 2026.

What Changed Since March 2026

This profile originally described a Claude Code-only methodology at 146 stars. The project has since been substantially rebuilt:

  • Dormancy, then revival. Commit history on GitHub shows no activity between November 10, 2025 and April 25, 2026 — a five-month stall that ended with the v3.0.0 breaking release: skill-driven workflows, a domain model, TDD missions, and a dual-target installer.
  • Codex support. npx ab-method now detects whether you run Claude Code (.claude/), Codex (.agents/), or both, and installs identical workflows with platform-specific triggers; v3.1.0 (May 15, 2026) made Codex support first-class.
  • Autonomous /goal loop. v3.1.x added /create-goal and /extend-goal — an autonomous execution mode with feedback-loop capture to steer long runs.
  • Handoffs and parallelism. v3.2.0 (May 23) added handoff capture so tangents discovered mid-task don't derail current work; v3.3.0 (June 6) added opt-in parallel mission groups ([pp-1], [pp-2]) and autonomous /start-task execution in subagents with auto-commits.
  • TDD-first. Every mission now runs through red-green-refactor cycles, grounded in project docs (UBIQUITOUS_LANGUAGE.md, CONTEXT.md).

Pricing

Free and open source under the MIT license; installed via npx ab-method. There is no commercial tier — costs are whatever you pay for Claude Code or Codex underneath.

Competitive Position

Strengths: Clear, opinionated methodology with TDD enforcement built in. No longer single-vendor — Claude Code and Codex from one installer. Active again, with a monthly-or-better release cadence since April 2026. Zero open issues.

Weaknesses: Small community by any measure — 174 stars, 14 forks, ~750 npm downloads/month, effectively a solo-maintainer project that has already stalled once for five months. It competes for attention with much larger methodology frameworks (Superpowers, BMAD, Spec Kit) in a crowded space. It is a methodology, not a tool — its value still depends on developer discipline.

What Developers Say

No attributed developer testimonials for the AB Method were findable in searches of Hacker News, Reddit, or developer blogs as of June 11, 2026. Its public footprint is essentially its GitHub repository and npm listing; treat the ~750 monthly npm downloads as the most honest adoption signal available.

Bottom Line

Recommended only as a lightweight, free experiment for solo developers or small teams who want an enforced plan-then-TDD-missions discipline on top of Claude Code or Codex — the v3 rewrite is genuinely more interesting than the original, and the dual-target installer plus autonomous /goal loop track where the category is heading. Not recommended as a team-wide standard: a 174-star solo project that went dark for five months once already carries real continuity risk, and larger frameworks cover the same ground with bigger communities. Outlook: revived and shipping monthly, but it needs visible adoption growth in the next two quarters to justify betting a workflow on it.

Research by Ry Walker Research

Sources

Share on XShare on LinkedIn