Skip to content

aegntic/compound-engineering

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Compound Engineering

Portable AI-powered development tools for any coding agent.

Claude Code · GitHub Copilot · OpenCode

29 specialized agents · 32 commands · 24 skills · MIT License

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


Why Compound Engineering?

Problem Solution
AI agents solve the same problem twice /workflows:compound documents every solved problem as searchable knowledge
Code review is shallow and generic 17 specialized reviewers per language and domain (Rust ownership, TypeScript types, OWASP security)
Planning disconnected from execution Pipeline produces artifacts that feed the next phase — nothing is lost
One-shot features need manual handoffs /lfg chains plan → deepen → work → review → ship autonomously
Knowledge leaves with the developer docs/solutions/ compounds institutional memory in git

Install

For Claude Code, Copilot CLI, OpenCode, or any agent that supports skills/commands.

# Claude Code — install as plugin
claude install-plugin https://github.com/aegntic/compound-engineering

# Or clone directly into your skills directory
git clone https://github.com/aegntic/compound-engineering.git \
  ~/.claude/skills/compound-engineering

# Configure for your stack (detects languages, frameworks, picks reviewers)
/setup

The Workflow

Compound Engineering provides a progression of workflows that build on each other. Each phase produces artifacts that feed the next, so nothing is lost and context compounds.

/constitution → /ideate → /brainstorm → /plan → /deepen-plan → /work → /review → /compound
                                                                                    ↑
                                                    knowledge feeds back ─────────────┘
Phase Command What it does
Define /workflows:constitution Establish durable project principles in docs/constitution.md
Discover /workflows:ideate Scan the repo, generate improvement ideas, rank them
Explore /workflows:brainstorm Establish WHY and WHAT — problem narrative, user story, architectural context
Plan /workflows:plan Structured implementation plan with tasks, dependencies, and evidence contracts
Deepen /deepen-plan Expand thin spots in the plan before execution begins
Execute /workflows:work TDD-driven execution with red-green-refactor evidence capture
Review /workflows:review Multi-agent code review using your configured reviewers
Compound /workflows:compound Document the solved problem as searchable knowledge in docs/solutions/

One-Shot Autonomous Mode

For features that don't need exploration — go straight to shipping:

/lfg add user authentication with JWT

This chains the entire pipeline automatically: plan → deepen → work → review → resolve → test → video → done.

For parallel swarm execution:

/slfg add user authentication with JWT

Agents

Design (3)

Review and iterate on visual implementation.

Agent Purpose
design-implementation-reviewer Compare live UI against Figma designs
design-iterator Iterative UI refinement through screenshot-analyze-improve cycles
figma-design-sync Detect and fix visual differences between implementation and Figma

Research (8)

Gather knowledge before and during implementation.

Agent Purpose
best-practices-researcher Research external best practices and documentation
framework-docs-researcher Gather framework/library documentation and examples
git-history-analyzer Archaeological analysis of git history
issue-intelligence-analyst Analyze GitHub issues for recurring patterns
learnings-researcher Search past solutions in docs/solutions/
repo-research-analyst Research repository structure and conventions

Review (17)

Language-specific and cross-cutting code review.

Agent Purpose
agent-native-reviewer Ensure agent-native action parity
architecture-strategist SOLID compliance and clean architecture review
code-simplicity-reviewer Complexity elimination and readability
data-integrity-guardian Database migration and data model safety
data-migration-expert Validate data migrations against reality
deployment-verification-agent Go/No-Go deployment checklists
frontend-race-conditions-reviewer Async UI race conditions and stale state
laravel-reviewer Laravel 11+ / PHP 8.3+ best practices
nestjs-reviewer NestJS simplicity, performance, security
pattern-recognition-specialist Design patterns and anti-patterns
performance-oracle Performance bottlenecks and algorithmic complexity
python-reviewer Pythonic patterns, type safety, security
rust-reviewer Ownership correctness, unsafe discipline, zero-cost abstractions
schema-drift-detector Schema snapshot and artifact drift detection
security-sentinel OWASP Top 10 security audit
typescript-reviewer Type safety, modern patterns, maintainability
vue-reviewer Vue.js/Nuxt best practices and performance

Workflow (3)

Automate development operations.

Agent Purpose
bug-reproduction-validator Systematically reproduce and validate bug reports
pr-comment-resolver Address PR review comments with requested changes
spec-flow-analyzer Analyze specs for user flow completeness

Commands

Workflows

The core progression — each produces artifacts that feed the next.

Command Description
/workflows:constitution Define project engineering principles
/workflows:ideate Generate and rank improvement ideas
/workflows:brainstorm Explore WHY and WHAT before planning
/workflows:plan Structured implementation planning
/workflows:architecture Architecture review and improvement
/workflows:work TDD-driven execution with evidence capture
/workflows:review Multi-agent code review
/workflows:compound Document solved problems as searchable knowledge
/workflows:compound-refresh Refresh stale learnings against current codebase

Operations

Command Description
/lfg Full autonomous pipeline: plan → work → review → ship
/slfg Same as /lfg but with parallel swarm execution
/ralph-loop Red-green-refactor loop with completion promise
/cancel-ralph Cancel an active ralph loop
/triage Triage GitHub issues
/reproduce-bug Systematic bug reproduction
/resolve_parallel Resolve PR comments in parallel
/resolve_todo_parallel Resolve TODO items in parallel
/test-browser Browser-based testing
/changelog Generate changelog
/deepen-plan Expand thin spots in a plan
/deploy-docs Deploy documentation site
/feature-video Record feature walkthrough video
/heal-skill Repair a broken skill
/create-agent-skill Create a new agent or skill
/report-bug Report a bug with structured context
/generate_command Generate a new command

Skills

Skill Purpose
agent-browser Browser automation via Vercel agent-browser CLI
agent-native-architecture Build applications where agents are first-class citizens
agent-native-audit Scored audit of agent-native architecture principles
brainstorming Explore user intent before implementing
caveman Ultra-compressed communication (~75% token reduction)
compound-docs Capture solved problems as searchable documentation
compound-refresh Refresh stale learnings against current codebase
create-agent-skills Expert guidance for authoring coding agent skills
document-review Refine brainstorm/plan documents
file-todos File-based todo tracking system
finishing-branch Verify quality and clean up feature branches
frontend-design Production-grade frontend interfaces
gemini-imagegen Image generation via Gemini API
git-worktree Git worktree management for parallel development
grill-me Relentlessly interview user about a plan or design
ideate Generate and evaluate grounded improvement ideas
laravel-conventions Modern Laravel 11+ / PHP 8.3+ coding standards
orchestrating-swarms Multi-agent swarm orchestration
rclone Cloud storage file management via rclone
resolve-pr-parallel Resolve PR comments using parallel processing
setup Configure review agents for your stack
skill-creator Guide for creating effective skills
systematic-debugging Structured 4-phase debugging methodology
ubiquitous-language DDD-style ubiquitous language glossary extraction

Project Structure

compound-engineering/
├── agents/                # 29 specialized AI agents
│   ├── design/            # Visual design review
│   ├── research/          # Knowledge gathering
│   ├── review/            # Code review (language + cross-cutting)
│   └── workflow/          # Development operations
├── commands/              # 32 slash commands
│   └── workflows/         # Core workflow orchestrations
├── skills/                # 24 skills (references, templates, scripts)
├── platforms/             # Platform-specific hooks
└── plugin.yaml            # Plugin metadata

Docs Convention

Workflows produce structured output in your project's docs/ directory:

docs/
├── constitution.md        # Project principles (keep in git)
├── solutions/             # Compounded learnings (keep in git)
├── ideation/              # Ideation artifacts (gitignore)
├── plans/                 # Implementation plans (gitignore)
├── brainstorms/           # Brainstorm documents (gitignore)
└── execution-sessions/    # Work session logs (gitignore)

Solutions are institutional memory — the compounding payoff. Everything else is ephemeral scaffolding.

Configuration

Run /setup to create compound-engineering.local.md in your project root. It auto-detects your stack and selects the right language reviewers for /workflows:review.

MCP Integration

Includes Context7 MCP server for enhanced documentation lookup during research and review.

Adding Components

New Agent

Create agents/{category}/{name}.md:

---
name: my-reviewer
description: Reviews code for X, Y, Z
model: claude-sonnet-4.6
---

## Mission
...

## Focus Areas
...

## Report
...

New Command

Create commands/{name}.md:

---
name: my-command
description: What this command does
argument-hint: '[optional arguments]'
---

# Command Title

Instructions...

New Skill

Create skills/{name}/SKILL.md plus optional references/, scripts/, or templates/ subdirectories.

Compatibility

Platform Status Install
Claude Code Supported claude install-plugin
GitHub Copilot Supported Copy agents + skills
OpenCode Supported Copy agents + skills
Any MCP-compatible agent Skills are plain markdown Copy what you need

Languages & Frameworks

Built-in reviewers for: Rust · TypeScript · Python · Laravel/PHP · Vue/Nuxt · NestJS · Frontend (React, vanilla) · SQL/Database migrations

Contributing

  1. Fork the repo
  2. Create your branch (git checkout -b feat/my-agent)
  3. Add agent/command/skill with frontmatter
  4. Update README counts
  5. Open a PR

See .github/ISSUE_TEMPLATE/ for agent and feature request templates.

License

MIT — use freely in personal and commercial projects.