Claude Code plugin · v2.4.0

Magento 2 engineering,
as Claude Code skills.

Scaffold, review, test, fix, deploy, and audit — 36 skills built around a shared context resolver that adapts to any project and environment.

Get Started View on GitHub
Install
/plugin marketplace add muon-m2/magento2-tools
/plugin install magento2-tools@muon-m2 --scope user

Capabilities

What it does

One toolkit for the full Magento 2 lifecycle — read-only until you approve a change.

Scaffold

New modules, EAV attributes, GraphQL/REST APIs, admin forms & grids, CMS widgets, CLI commands, and message queues.

Audit & Triage

Severity-ranked findings with file:line evidence and a merged SARIF — then triage routes each one to the skill that owns the fix.

Test

Generate unit, integration, REST/GraphQL API, JavaScript, and MFTF tests with real assertions.

Fix & Remediate

Reproduce → root-cause → minimal TDD fix → regression test for one bug; or work a whole audit report batch by batch, one approval each.

Deploy & Release

Pre-flight validation, ordered deploy with rollback, and versioned releases with generated notes.

Context-aware

Hub-and-spoke around context — adapts to Docker or bare PHP, Luma, Hyvä, or Breeze, any vendor.

Token-frugal runs

Warns before a run starts in an oversized conversation, keeps test and log output out of the context, and keeps every skill intact through compaction.

The remediation loop

From findings to fixes, deterministically

An audit tells you what is wrong. The loop decides who fixes each finding, executes it through the right skill, and proves it closed.

  1. 01

    audit — find

    Every read-only dimension in parallel, consolidated into one severity-ranked report with file:line evidence and a merged SARIF for CI.

  2. 02

    triage — decide

    Routes every finding to the skill that owns it — a queue finding to message-queue, an indexer finding to indexer, a Breeze finding to breeze-adapt — then batches them in dependency order. Waived findings stay waived; low-confidence ones never drive an automated patch.

  3. 03

    remediate — fix

    One approval per batch, not per finding. Each fix lands as its own commit, still test-first. Anything needing a human — rotating a leaked credential, a compliance sign-off — is listed, never executed.

  4. 04

    audit --compare — prove

    Re-runs and diffs against the baseline: what closed, what is still open, and what the remediation itself introduced. A check that could not run is reported as skipped — never as clean.

Documentation

Read the guides

Full documentation lives in the repository, rendered on GitHub.