Claude-Code

What Changed

Released v1.4.0 of my Elixir Phoenix Guide plugin. Two new skills, three new hooks, and a fix for something that’s been bugging me: subagents ignoring the rules.

The Problem with v1.3.x

The plugin had five skills covering Elixir core, LiveView, Ecto, uploads, and testing. That’s solid for web layer code. But two big gaps remained.

No OTP guidance. GenServer, Supervisor, Task, Agent, ETS. These are core to any real Elixir application, and the plugin had nothing to say about them. Claude Code would write GenServers with blocking handle_call callbacks, skip supervision trees, or reach for Agent when a simple GenServer would do.

What Changed

Released v1.3.0 of my Elixir Phoenix Guide plugin. This one is all about testing.

v1.2.0 built out four skills that cover Elixir patterns, LiveView, Ecto, and file uploads. What I didn’t have was anything for tests. You could open a _test.exs file and the plugin had nothing to say. That felt like a hole worth fixing.

The Problem with v1.2.0

The four skills worked well for production code but testing was just not covered. Claude Code would write a context module following strict patterns, then write the tests with no real guidance. The result was inconsistent test setup, missing edge cases, and assertion styles that made failures hard to read.

What Changed

Released v1.2.0 of my Elixir Phoenix Guide plugin. This is a major restructuring: eight separate skills consolidated into four essential modules, each with enforced rules and real-time validation.

This isn’t a minor update. It’s a complete rethinking of how Claude Code should guide Elixir/Phoenix development.

The Problem with v1.1.x

In v1.1.x, I had eight skills covering different aspects of Elixir and Phoenix development. The structure worked, but created problems:

What’s New

Released v1.1.2 of my Elixir Claude optimization plugin. The focus: making skills easier to discover and use.

The Problem

In v1.0.0, Claude Code would often miss applicable skills. The “Use when” language wasn’t directive enough, and there was no systematic way to identify which skills applied to a task.

Result: Skills existed but weren’t being invoked consistently.

What Changed

New Skill Discovery System

Added a meta-skill called skill-discovery that provides a systematic checklist based on file types and task requirements. Think of it as a flowchart for skill selection.

The Experiment

I’ve been using Claude Code for Elixir development and wondered: Does customizing the AI with skills, hooks, and agent documentation actually make a difference?

To find out, I built the same Phoenix LiveView image gallery application twice:

• Build 1 (Baseline): Minimal configuration - just project guidelines and specs
• Build 2 (Plugin): Custom skills, hooks, agent docs, and comprehensive project documentation

Both builds used identical project specifications and the same base agent guidelines. The only difference was the additional plugin configuration in Build 2.