Claude Code Compound Engineering

The Compound Engineering plugin extends Claude Code with specialized agents, slash commands, skills, and structured workflows for Plan → Work → Review → Compound cycles.

Installation

Step	Command
Install Claude Code	Follow claude.ai/code setup
Clone plugin	git clone https://github.com/anthropics/claude-code-plugins
Add to Claude Code	Add plugin path in Claude Code settings
Verify	Run /help in Claude Code to see available commands

Core Workflow Cycle

Phase	Slash Command	Description
Plan	/plan	Transform feature descriptions into structured implementation plans
Work	/lfg	Full autonomous engineering — execute plans with multi-agent coordination
Review	/review-pr	Comprehensive PR review using specialized parallel agents
Compound	/compound	Document solved problems to build institutional knowledge

Slash Commands

Planning & Architecture

Command	Description
/plan	Create structured implementation plans with file-level detail
/brainstorm	Explore requirements and approaches before planning
/deepen-plan	Enhance a plan with parallel research agents for depth and best practices
/plan_review	Have multiple specialized agents review a plan in parallel

Development & Execution

Command	Description
/lfg	Full autonomous engineering workflow — plan, implement, test, review
/resolve_parallel	Resolve all TODO comments using parallel processing
/resolve_pr_parallel	Resolve all PR comments using parallel processing
/resolve_todo_parallel	Resolve all pending CLI todos using parallel processing

Review & Quality

Command	Description
/review-pr	Comprehensive PR review with specialized agents (code, security, performance)
/code-review	Review a specific code change against plan and standards
/test-browser	Run browser tests on pages affected by current PR or branch
/reproduce-bug	Reproduce and investigate bugs using logs, console, and screenshots

Documentation & Knowledge

Command	Description
/compound	Document a solved problem for institutional knowledge
/changelog	Create engaging changelogs for recent merges to main
/deploy-docs	Validate and prepare documentation for GitHub Pages
/release-docs	Build and update documentation site with current components

Utilities

Command	Description
/generate_command	Create a new custom slash command
/triage	Triage and categorize findings for the CLI todo system
/feature-video	Record a feature walkthrough video and add to PR description
/agent-native-audit	Run comprehensive agent-native architecture review

Specialized Agents

Code Review Agents

Agent	Purpose
code-reviewer	Reviews for bugs, logic errors, security, and code quality
code-simplifier	Simplifies code for clarity while preserving functionality
silent-failure-hunter	Identifies silent failures and inadequate error handling
comment-analyzer	Analyzes comments for accuracy and maintainability
pr-test-analyzer	Reviews test coverage quality and completeness
type-design-analyzer	Expert analysis of type design and invariants
code-simplicity-reviewer	Final pass for YAGNI compliance and minimal complexity

Architecture & Security Agents

Agent	Purpose
architecture-strategist	Evaluates system design decisions and component boundaries
security-sentinel	Security audits, vulnerability assessment, OWASP compliance
performance-oracle	Analyzes performance, algorithms, bottlenecks, scalability
agent-native-reviewer	Ensures features are accessible to agents, not just humans
pattern-recognition-specialist	Detects design patterns, anti-patterns, and code duplication

Language-Specific Reviewers

Agent	Purpose
kieran-rails-reviewer	Strict Rails conventions and quality review
kieran-python-reviewer	Strict Python conventions and quality review
kieran-typescript-reviewer	Strict TypeScript conventions and quality review
dhh-rails-reviewer	Rails review from DHH’s perspective — anti-pattern detection
julik-frontend-races-reviewer	Frontend race condition detection in JS/Stimulus

Data & Deployment Agents

Agent	Purpose
data-integrity-guardian	Database migrations, data constraints, referential integrity
data-migration-expert	Validates ID mappings, rollback safety, data transformations
deployment-verification-agent	Pre/post-deploy checklists, SQL verification, rollback plans

Research Agents

Agent	Purpose
framework-docs-researcher	Fetches comprehensive framework docs and best practices
learnings-researcher	Searches institutional learnings in docs/solutions/
best-practices-researcher	Gathers external best practices from documentation and OSS
git-history-analyzer	Archaeological analysis of git history and code evolution
repo-research-analyst	Comprehensive repository structure and pattern analysis

Design Agents

Agent	Purpose
figma-design-sync	Synchronizes implementation with Figma designs
design-implementation-reviewer	Verifies UI matches Figma specifications
design-iterator	Iterative design refinement through N screenshot-analyze-fix cycles

Skills

Development Skills

Skill	Trigger
frontend-design	Build web components with high design quality
agent-native-architecture	Design apps where agents are first-class citizens
dspy-ruby	Ruby framework for type-safe, composable LLM apps
andrew-kane-gem-writer	Write Ruby gems following Andrew Kane patterns
dhh-rails-style	Write Ruby/Rails code in DHH’s 37signals style

Workflow Skills

Skill	Trigger
brainstorming	Explore intent and approaches before implementation
git-worktree	Manage Git worktrees for isolated parallel development
file-todos	Manage file-based todo tracking system
every-style-editor	Review and edit text for Every’s style guide
compound-docs	Capture solved problems as categorized documentation

Tool Integration Skills

Skill	Trigger
agent-browser	Browser automation using Vercel’s agent-browser CLI
rclone	Upload and sync files across cloud storage providers
gemini-imagegen	Generate and edit images using Gemini API
xcode-test	Build and test iOS apps on simulator
skill-creator	Guide for creating effective new skills

Hooks

Hook	Event	Purpose
SessionStart	Session initialization	Set up context, load memories
PreToolUse	Before tool execution	Validate, block dangerous commands
PostToolUse	After tool execution	Process results, trigger follow-ups
Stop	Agent completion	Final checks, cleanup
UserPromptSubmit	User sends message	Pre-process user input

Multi-Agent Review Workflow

The /review-pr command launches multiple specialized agents in parallel:

┌─────────────────────────────────────────────────┐
│                  /review-pr                      │
├─────────────────────────────────────────────────┤
│  ┌──────────┐  ┌──────────┐  ┌──────────┐      │
│  │ Code     │  │ Security │  │ Perf     │      │
│  │ Reviewer │  │ Sentinel │  │ Oracle   │      │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘      │
│       │              │              │            │
│  ┌────┴─────┐  ┌────┴─────┐  ┌────┴─────┐      │
│  │ Silent   │  │ Type     │  │ Pattern  │      │
│  │ Failure  │  │ Design   │  │ Recog.   │      │
│  └────┬─────┘  └────┬─────┘  └────┴─────┘      │
│       └──────────┬───┘──────────────┘            │
│            Consolidated Report                   │
└─────────────────────────────────────────────────┘

Common Use Cases

Feature Development

# 1. Brainstorm the approach
/brainstorm

# 2. Create a structured plan
/plan

# 3. Deepen with research agents
/deepen-plan

# 4. Execute autonomously
/lfg

# 5. Review the PR
/review-pr

# 6. Document what you learned
/compound

Resolving PR Comments

# Resolve all PR comments in parallel
/resolve_pr_parallel

# Or resolve TODOs across the codebase
/resolve_parallel

Creating Custom Commands

# Generate a new slash command
/generate_command

# Create a new skill
Use the skill-creator skill for guidance

Configuration

Plugin Structure

compound-engineering/
├── plugin.json          # Plugin manifest
├── commands/            # Slash command definitions
│   ├── lfg.md
│   ├── plan.md
│   ├── review-pr.md
│   └── ...
├── agents/              # Agent definitions
│   ├── code-reviewer.md
│   ├── security-sentinel.md
│   └── ...
├── skills/              # Skill definitions
│   ├── frontend-design/
│   ├── agent-native-architecture/
│   └── ...
└── hooks/               # Event hooks
    ├── session-start.md
    └── ...

Agent Configuration

Agents are defined in Markdown files with YAML frontmatter:

---
name: "code-reviewer"
description: "Reviews code for bugs, logic errors, and quality"
tools:
  - Glob
  - Grep
  - Read
  - Task
model: sonnet  # or opus, haiku
---

Best Practices

• Start with /brainstorm before jumping into implementation — it surfaces edge cases early
• Use /plan for anything non-trivial — even “simple” features benefit from structured thinking
• Run /review-pr before merging — the multi-agent review catches issues single-pass reviews miss
• Document with /compound — institutional knowledge prevents repeating mistakes
• Use /deepen-plan for complex features — research agents add depth from docs and best practices
• Leverage parallel resolution — /resolve_parallel handles TODOs faster than sequential fixes
• Create custom skills for repeated patterns — encapsulate domain knowledge for reuse
• Use haiku model for quick agents — minimize cost and latency for straightforward tasks

Troubleshooting

Issue	Solution
Agent not found	Verify plugin path in Claude Code settings and restart
Command not recognized	Run /help to list available commands, check plugin.json
Agent returns empty	Check model availability, may need API key for opus/sonnet
Slow multi-agent review	Use haiku model for quick agents, sonnet for critical ones
Skill not triggering	Check skill description matches your prompt phrasing
Hook not firing	Verify hook event type matches (PreToolUse vs PostToolUse)
Plan too vague	Use /deepen-plan with research agents for more detail
Review misses context	Provide the PR URL or branch name explicitly to /review-pr