Install phxagents v3.0.0

One canonical set of 51 Elixir & Phoenix skills, packaged for five AI coding runtimes. Claude Code installs the full plugin with 26 named agents and hooks; Amp, Codex, Pi and OpenCode install generated, runtime-native skill editions.

Choose your runtime

Compare capabilities →

CC Claude CodeFull canonical plugin · skills, agents, hooks and permissions /phx:* A AmpGenerated Agent Skills · project-local or global skill: invoke CX CodexNative skills plugin · optional trust-gated safety hook $elixir-phoenix:* π PiNative Git skills package /skill:* OC OpenCodeGenerated skills via sparse Git checkout skill tool

Using Amp, Codex, Pi or OpenCode? Choose it above to open its complete guide. Continue below only to install the full Claude Code plugin.

Claude Code — install the full plugin

Run these inside Claude Code, one at a time:

1. Add the marketplace

/plugin marketplace add oliver-kriska/claude-elixir-phoenix

2. Install the plugin

/plugin install elixir-phoenix

A fresh v3 install automatically pulls the small ecto and lv compatibility namespaces that preserve /ecto:* and /lv:* commands. Run /reload-plugins in an open session before using the new skills, agents and hooks.

Updating from v2.x

Install the new command-compatibility dependencies before updating the main plugin:

Terminal

claude plugin marketplace update oliver-kriska
claude plugin install ecto@oliver-kriska
claude plugin install lv@oliver-kriska
claude plugin update elixir-phoenix@oliver-kriska

Restart Claude Code afterward. Confirm that /phx:help, /ecto:n1-check, and /lv:assigns appear before continuing work.

What gets installed

Every runtime: all 51 canonical skills, complete bundled resources, and 9 fully adapted workflows.
Claude Code: 26 specialist agents, lifecycle and enforcement hooks, root instructions, permissions, and full workflow orchestration.
Amp, Codex, Pi, OpenCode: generated skill distributions with native names and invocation. Named Claude agents and its broad hook set are not silently emulated.
Tidewave MCP: supported when your project and client already expose it; registration remains external to every edition.

After install

In Claude Code, run /phx:intro for a guided walkthrough. In another runtime, load phx-plan, phx-investigate, or phx-review with the native syntax shown in its install guide.

Claude Code

/phx:plan Add a unique constraint on the email column

Security & trust

The plugin runs skills, agents, and hooks inside Claude Code with your tool permissions — so it’s independently scanned with NVIDIA SkillSpector, a static security scanner for AI-agent skills. The scan runs with zero data egress.

The baseline-aware repository scan passes after reviewed false-positive baselines. The current raw totals and every baseline decision live in SECURITY.md.
Lower-scoring CAUTION components remain documented rather than being hidden behind baselines.
The few higher-scoring components are documented false positives — a static scanner flags the safety-enforcing and security-auditing skills precisely because they name dangerous patterns in order to forbid or detect them (one flagged line literally reads NEVER bypass security checks for speed).
No data exfiltration, read-only review agents (source edits disallowed), and auditable bash hooks.

Read the full report, line-by-line triage, and a reproduce-it-yourself command in SECURITY.md.