xp-agents

██╗  ██╗██████╗        █████╗  ██████╗ ███████╗███╗   ██╗████████╗███████╗
╚██╗██╔╝██╔══██╗      ██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝██╔════╝
 ╚███╔╝ ██████╔╝══════███████║██║  ███╗█████╗  ██╔██╗ ██║   ██║   ███████╗
 ██╔██╗ ██╔═══╝       ██╔══██║██║   ██║██╔══╝  ██║╚██╗██║   ██║   ╚════██║
██╔╝ ██╗██║           ██║  ██║╚██████╔╝███████╗██║ ╚████║   ██║   ███████║
╚═╝  ╚═╝╚═╝           ╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝   ╚═╝   ╚══════╝

Extreme Programming in a Box

XP-Agents is a Claude Code plugin that makes your agents — solo or in teams — write better software through XP practices. Command hooks enforce deterministic rules. Plugin subagents provide judgment-based guidance. A broadcast event log keeps every agent aligned. Zero config. Install and go.

Forward (from the Human)

Agents are non-deterministic, dishonest, lazy and they constantly cheat, all in their unshakable desire to achieve their goal. They ignore what you tell them, they think instructions are for the weak. They do things their own way.

Kind of like humans.

Over 25 years I've been running, hiring and working with engineers. Messy, spectrumy, brilliant, strange. They run the gamut. But one thing has constantly brought them together into a powerful team, and those are the values and discipline of Extreme Programming.

So, as Agents become more and more human like, using some of the same practices with them just made sense. This plugin is my attempt to coral them into a working team, first as a teammate of me (solo agent) and then as teammates of each other (CLI teammates running in parallel worktrees).

So, give it a shot and drop me an issue. I'd love to hear what you think...

Paul, the human theoretically in charge..;)

TL;DR

What it does: Command hooks fire automatically on tool calls — blocking conflicts, enforcing TDD, running linters, and tracking status. Plugin subagents provide strategic guidance: a plan reviewer validates plans and a retrospective analyst surfaces cross-session learning. Inline skills handle work selection, quality review, sprint planning, and SMM curation. Everything is broadcast through a shared event log visible to every agent.

How it works: Deterministic enforcement (tests, lint, conflicts, security) lives in command hooks — they fire every time. Judgment work (plan analysis, retrospectives, quality review) lives in plugin subagents and inline skills, triggered by hook nudges or stop gates.

Who it's for: Anyone using Claude Code — solo agents benefit from the quality review, retrospectives, and TDD enforcement. Teams benefit from the broadcast coordination layer that keeps every agent aligned.

---

Install

From within a Claude Code session:

# Add the marketplace
/plugin marketplace add paulingalls/xp-agents

# Install
/plugin install xp-agents@xp-agents

Or from your terminal (skips the interactive scope picker):

# Add the marketplace
claude plugin marketplace add paulingalls/xp-agents

# Install at user scope (available across all projects)
claude plugin install xp-agents@xp-agents --scope user

# Or install at project scope (shared with team via .claude/settings.json)
claude plugin install xp-agents@xp-agents --scope project

For local development, use --plugin-dir (session-only, not persisted):

claude --plugin-dir /path/to/xp-agents/plugins/xp-agents

Requirements: Python 3.11+ on PATH. macOS or Linux. Zero external packages.

Scopes: User scope makes xp-agents available on all your projects. Project scope shares it with your team via version control. Both work with CLI teammates — the SMM is stored in CLAUDE_PLUGIN_DATA (shared across worktrees).

For teams: Add this to your project's .claude/settings.json so teammates can discover the plugin:

{
  "extraKnownMarketplaces": {
    "xp-agents": {
      "source": { "source": "github", "repo": "paulingalls/xp-agents" }
    }
  }
}

Each person installs individually. The marketplace entry is just discovery.

---

What Happens When You Use It

First session:

$ claude
> Starting session...

[xp-agents] Initializing SMM for project... ✓
[xp-agents] Kickoff required — run /xp-kickoff

> /xp-kickoff

[xp-agents] First session — no retrospective data.
[xp-agents] What are your goals for this session?

> "Build a REST API for user management with auth and role-based access."

[xp-agents] Goal recorded.
[xp-agents] Running housekeeping — curating Shared Mental Model...
[xp-agents] SMM curated. Process guide loaded. Ready to work.