toolkit

A Claude Code plugin with multi-agent orchestration, TDD workflows, and advanced productivity commands.

---

Installation

# One-line install (recommended)
curl -fsSL https://raw.githubusercontent.com/ruizrica/toolkit/main/install.sh | bash

This clones to ~/.toolkit, registers the plugin, installs scripts/skills, creates agent memory directories (~/.claude/agent-memory/), and installs optional tools (agent-browser, just-bash, agent-memory).

Manual Install

git clone https://github.com/ruizrica/agent-toolkit.git
claude plugins add ./agent-toolkit/plugins/toolkit
mkdir -p ~/.claude/slash_commands
cp plugins/toolkit/scripts/handbook.py ~/.claude/slash_commands/
cp plugins/toolkit/scripts/coderabbit_workflow.py ~/.claude/slash_commands/
mkdir -p ~/.claude/agent-memory/{daily-logs,sessions,procedures}

---

Quick Start

Multi-Agent Development

/team Implement user authentication with OAuth support

Spawns multiple specialized agents to work on your feature in parallel.

Code Review & Fixes

/review --base main

Runs CodeRabbit review, creates parallel fix tasks, and verifies completion.

Session Continuity with Agent Memory

/compact      # Save state + write daily log
/clear        # Clear context
/restore      # Resume with daily log context

Uses a local memory system (~/.claude/agent-memory/) with daily logs, session snapshots, and semantic memory. The PreCompact hook automatically captures session snapshots; /compact writes a quality daily log entry; /restore bootstraps from today's and yesterday's logs.

For a faster, minimal compact without memory writes: /compact-min

Spec-Driven Development

/kiro Add user authentication with OAuth

Generates requirements, design, and tasks using Kiro methodology, then executes with parallel agents.

Sandboxed Bash (just-bash)

A sandboxed bash interpreter from Vercel Labs that lets Claude run commands safely with a read-only filesystem, no network access, and in-memory-only writes. Includes 75+ built-in commands (jq, yq, xan, sqlite3, rg, awk, sed, etc.).

# Explore a codebase safely (read-only, can't break anything)
just-bash -c 'find . -name "*.ts" | wc -l'

# Process JSON/CSV/YAML data
just-bash -c 'cat data.json | jq ".items[] | .name"'

# Test destructive scripts safely (writes stay in memory)
just-bash --allow-write -c 'rm -rf src && echo "nothing happened on disk"'

The toolkit installs a skill file (~/.claude/skills/just-bash.md) that teaches Claude when to use the sandbox vs regular Bash. See docs for the full command list and usage patterns.

Git Worktree Workflow

/setup      # Initialize project context
/worktree   # Create isolated worktree (auto-generates branch)
# ... develop ...
/save       # Merge back to main

---

Documentation

Section	Description
Commands	All 16 commands with usage and examples
Agents	All 9 specialized agents with invocation patterns
Skills	Skill reference files for CLI tools

---

Commands at a Glance

Core Commands

Command	Description	Docs
/team	Multi-agent parallel implementation	→
/haiku	10 Haiku agents managed by Opus	→
/review	CodeRabbit review + parallel fixes	→
/handbook	Generate AI-optimized project docs	→
/@implement	Process @implement comments	→
/rlm	Large document processing	→
/gherkin	Extract business rules to Gherkin	→
/kiro	Spec-driven development with Kiro methodology	→
/agent-memory	Hybrid search over agent memory files	→

Session & Git

Command	Description	Docs
/compact	Memory-aware session compact (daily log + state)	→
/compact-min	Ultra-minimal session compact (fast, no memory)	→
/restore	Restore after /clear (loads daily logs)	→
/setup	Initialize context and run agent-memory index	→
/worktree	Create isolated worktree (auto-generates branch)	→
/save	Commit, merge, cleanup	→
/stable	Create stable checkpoint	→

---

Agents at a Glance