Summary

This Claude Code plugin addresses common, hard-to-catch bugs and production issues in Elixir/Phoenix applications by leveraging a suite of 20 specialist AI agents. It provides developers using Claude Code with an automated workflow for planning, implementing, reviewing, and verifying code changes, enforcing domain-specific best practices through "Iron Laws." The primary benefit is significantly reducing common Elixir/Phoenix mistakes and improving code quality through AI-driven, context-aware assistance.

How It Works

The plugin orchestrates a multi-stage lifecycle: Plan, Work, Verify, Review, and Compound. It utilizes a hierarchical agent structure (Orchestrators, Specialists, Lightweight) that coordinate tasks. Key to its operation are "Iron Laws"—non-negotiable rules derived from common production pitfalls (e.g., avoiding :float for monetary values, ensuring Oban job idempotency). Development plans are managed within dedicated namespaces (.claude/plans/{slug}/), acting as a filesystem-based state machine. The Context Supervisor pattern compresses and deduplicates outputs from parallel research and review agents to manage Claude's context window effectively. Skills are auto-loaded based on the files being edited, providing relevant domain knowledge (e.g., LiveView patterns, Ecto schemas) without explicit commands.

Quick Start & Requirements

• Installation: Recommended installation is via the Claude Code marketplace: /plugin marketplace add oliver-kriska/claude-elixir-phoenix followed by /plugin install elixir-phoenix. Local installation for development is also supported via git clone.
• Prerequisites: Requires the Claude Code CLI and an Elixir/Phoenix project. Optional integration with Tidewave is available for runtime debugging.
• Getting Started: An interactive tutorial is available via /phx:intro.
• Documentation: The primary documentation is the README itself, with commands and concepts detailed therein.

Highlighted Details

• Features 20 specialist agents and 41 distinct skills.
• Enforces 22 "Iron Laws" covering LiveView, Ecto, Oban, Security, and Elixir best practices.
• Integrates with Tidewave MCP for runtime introspection and debugging.
• Supports a full development lifecycle: /phx:plan (research & planning), /phx:work (implementation), /phx:verify (automated checks), /phx:review (multi-agent audit), and /phx:compound (knowledge capture).
• Utilizes parallel agent execution for research and code review phases.
• Provides commands for quick tasks (/phx:quick), deep audits (/phx:audit), and autonomous development (/phx:full).

Maintenance & Community

The README mentions an ongoing elixir-inspector plugin development. No specific details regarding active maintainers, sponsorships, or community channels (like Discord or Slack) are provided. Contributions are welcomed via Pull Requests.

Licensing & Compatibility

The project is released under the MIT license, which is permissive and generally suitable for commercial use and integration into closed-source projects. It is designed for Elixir/Phoenix projects.

Limitations & Caveats

This plugin is dependent on the Claude Code CLI and its underlying LLM capabilities. While it aims to catch errors missed by traditional testing, it does not replace the need for comprehensive unit, integration, and end-to-end testing. The README is highly technical and dense, requiring a strong understanding of Elixir, Phoenix, and AI agent concepts for effective utilization.