Skip to content

kavanaghpatrick/ralph-compounder

Repository files navigation

Claude Code Plugin 29 Agents 24 Commands 18 Skills

Ralph Compounder

AI-powered development tools that compound with every use.

A Claude Code plugin marketplace that turns your AI coding assistant into a full engineering team — with specialized agents for code review, research, design, security, and workflow automation. Each cycle of work makes the next one easier.

Fork of EveryInc/compound-engineering-plugin, extended with safety hooks, spec-driven development, and multi-agent orchestration.


Quick Start

# In Claude Code
/plugin marketplace add https://github.com/kavanaghpatrick/ralph-compounder
/plugin install ralph-compounder

That's it. You now have 29 agents, 24 commands, 18 skills, and 1 MCP server.


What's Different from Compound Engineering?

Ralph Compounder builds on the original Compound Engineering plugin with several key additions:

Feature Original Ralph Compounder
Safety hooks None PreToolUse hooks block destructive ops
Secret protection None Auto-prompts before editing .env, .pem, .key files
jq injection safety Vulnerable to shell injection All hooks use --arg for safe variable passing
Spec-driven dev Manual planning Ralph Specum: research -> requirements -> design -> tasks -> execute
Multi-agent specs Single-agent Foreman Spec: 4 specialized AI agents (PM, UX, Tech, QA)
CLI tool None Convert plugins to OpenCode + Codex formats
Test coverage None 67 tests across 10 test files (hook validation, CLI, parsing)
GitHub org EveryInc kavanaghpatrick (independent fork)

Safety Hooks

Two PreToolUse hooks that act as guardrails:

validate-bash.sh — Catches destructive shell commands before they run:

  • git push --force / git reset --hard — prompts for confirmation
  • rm -rf / / rm -rf ~ / rm -rf . — hard blocked (denied)
  • rm -rf node_modules / .cache / tmp — silently allowed (safe targets)
  • All other rm -rf — prompts for confirmation

protect-env-files.sh — Prompts before editing sensitive files:

  • .env, .env.local, .env.production etc.
  • .pem certificates and .key files
  • Files with credentials in the name
  • secret.json, secret.yml, secrets.yaml etc.

The Workflow

Plan  -->  Work  -->  Review  -->  Compound  -->  Repeat
 80%         20%        ^^^           ^^^
                    multi-agent    knowledge
                     parallel      captured

Each cycle compounds: plans inform future plans, reviews catch more issues, patterns get documented. The ratio is intentional — 80% planning and review, 20% execution.

Core Commands

Command What it does
/workflows:plan Turn feature ideas into detailed implementation plans
/workflows:work Execute plans with worktrees, task tracking, and PR lifecycle
/workflows:review Multi-agent code review (15 specialized reviewers in parallel)
/workflows:compound Document solved problems so future work gets easier
/workflows:brainstorm Explore requirements and approaches before committing to a plan

Power Commands

Command What it does
/lfg Full autonomous workflow: plan -> work -> review -> ship
/slfg Same as /lfg but with swarm mode — parallel agent execution
/deepen-plan Enhance any plan with parallel research agents per section
/technical_review Multi-agent architecture and security review
/test-browser Automated browser testing on PR-affected pages

29 Specialized Agents

Agents are organized by specialty. Use them individually or let the workflow commands orchestrate them.

Review (15 agents)

Agent Focus
kieran-rails-reviewer Rails code review with strict conventions
kieran-python-reviewer Python review with strict conventions
kieran-typescript-reviewer TypeScript review with strict conventions
dhh-rails-reviewer Rails review from DHH's perspective
security-sentinel Security audits and vulnerability assessments
performance-oracle Performance analysis and optimization
architecture-strategist Architectural decisions and compliance
code-simplicity-reviewer Final pass for simplicity and minimalism
data-integrity-guardian Database migrations and data integrity
data-migration-expert Validate ID mappings match production
deployment-verification-agent Go/No-Go deployment checklists
schema-drift-detector Detect unrelated schema.rb changes
pattern-recognition-specialist Code patterns and anti-patterns
agent-native-reviewer Verify features are agent-native
julik-frontend-races-reviewer JavaScript/Stimulus race conditions

Research (5 agents)

Agent Focus
best-practices-researcher External best practices and examples
framework-docs-researcher Framework documentation deep-dives
git-history-analyzer Git history and code evolution
learnings-researcher Institutional knowledge from past solutions
repo-research-analyst Repository structure and conventions

Design (3 agents)

Agent Focus
figma-design-sync Synchronize implementations with Figma designs
design-iterator Iterative UI refinement
design-implementation-reviewer Verify UI matches designs

Workflow (5 agents)

Agent Focus
bug-reproduction-validator Systematically reproduce bug reports
pr-comment-resolver Address PR comments and implement fixes
spec-flow-analyzer Analyze user flows and identify gaps
every-style-editor Content editing to style guide
lint Ruby and ERB linting

Docs (1 agent)

Agent Focus
ankane-readme-writer READMEs following Ankane's gem template

18 Skills

Skills provide deep domain expertise that Claude can apply during any task.

Architecture & Development

Skill Description
agent-native-architecture Build AI agents using prompt-native patterns
dhh-rails-style Ruby/Rails in DHH's 37signals style
andrew-kane-gem-writer Ruby gems following Andrew Kane's patterns
dspy-ruby Type-safe LLM applications with DSPy.rb
frontend-design Production-grade frontend interfaces
create-agent-skills Expert guidance for creating Claude Code skills
skill-creator Skill creation workflow with validation

Content & Workflow

Skill Description
compound-docs Capture solved problems as categorized docs
brainstorming Collaborative requirement exploration
document-review Structured self-review for documents
every-style-editor Style guide compliance checking
file-todos File-based todo tracking
git-worktree Parallel development with worktrees

Automation & Media

Skill Description
orchestrating-swarms Multi-agent swarm coordination
agent-browser CLI browser automation (Vercel agent-browser)
gemini-imagegen Image generation/editing via Gemini API
rclone Upload to S3, Cloudflare R2, B2
resolve-pr-parallel Parallel PR comment resolution

MCP Server

Server Description
context7 Documentation lookup for 100+ frameworks (Rails, React, Next.js, Vue, Django, Laravel, etc.)

Starts automatically when the plugin is enabled. Provides resolve-library-id and get-library-docs tools.


CLI Tool (OpenCode + Codex)

Ralph Compounder includes a Bun/TypeScript CLI that converts Claude Code plugins into other agent formats.

# Convert to OpenCode format
bunx @kavanaghpatrick/ralph-compounder install ralph-compounder --to opencode

# Convert to Codex format
bunx @kavanaghpatrick/ralph-compounder install ralph-compounder --to codex

# Both at once
bunx @kavanaghpatrick/ralph-compounder install ralph-compounder --to opencode --also codex

Sync Personal Config

# Sync your ~/.claude/ skills and MCP servers to OpenCode
bunx @kavanaghpatrick/ralph-compounder sync --target opencode

# Sync to Codex
bunx @kavanaghpatrick/ralph-compounder sync --target codex

Skills are symlinked (not copied) so changes in Claude Code reflect immediately.


Philosophy

Each unit of engineering work should make subsequent units easier — not harder.

Traditional development accumulates technical debt. Every feature adds complexity. The codebase gets harder to work with over time.

Compound engineering inverts this:

  1. Plan thoroughly before writing code
  2. Review with multiple specialists to catch issues early
  3. Document learnings so they're reusable (/workflows:compound)
  4. Keep quality high so future changes are easy

The name "Compounder" comes from this compounding effect — each cycle of Plan -> Work -> Review -> Compound makes the next cycle faster and higher quality.


Project Structure

ralph-compounder/
├── .claude-plugin/
│   └── marketplace.json        # Marketplace catalog
├── plugins/
│   ├── ralph-compounder/       # Main plugin (29 agents, 24 commands, 18 skills)
│   │   ├── agents/             # review/, research/, design/, workflow/, docs/
│   │   ├── commands/           # Slash commands + workflows/
│   │   ├── skills/             # 18 skill directories
│   │   ├── hooks/              # Safety hooks (validate-bash, protect-env)
│   │   └── .claude-plugin/     # Plugin metadata
│   └── coding-tutor/           # Coding tutorial plugin
├── src/                        # CLI tool (Bun/TypeScript)
│   ├── commands/               # install, convert, sync, list
│   ├── targets/                # opencode, codex writers
│   └── utils/                  # Parsing, frontmatter, permissions
├── tests/                      # 67 tests across 10 files
└── docs/                       # Documentation site (GitHub Pages)

Development

# Run tests
bun test

# Local dev
bun run src/index.ts install ./plugins/ralph-compounder --to opencode

# Validate configs
jq . .claude-plugin/marketplace.json
jq . plugins/ralph-compounder/.claude-plugin/plugin.json

Learn More


License

MIT

About

AI-powered development tools that compound with every use

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors