convez.ai
Agente
aegntic

compound-engineering

aegntic/compound-engineering

Ver no GitHub
InicialPluginScore · de 0 a 100

Portable AI-powered development tools for any coding agent. Workflow orchestration, TDD pipelines, code review, and knowledge compounding for Claude Code, Copilot, and OpenCode.

Instalação

$npx convez-ai aegntic/compound-engineering

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
57%
Último commit
há 2 semanas
Versão
v5.0.0
Issues abertas
0
Contribuidores
2

README

A tradução para o português deste README ainda está na fila.

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?

ProblemSolution
AI agents solve the same problem twice/workflows:compound documents every solved problem as searchable knowledge
Code review is shallow and generic17 specialized reviewers per language and domain (Rust ownership, TypeScript types, OWASP security)
Planning disconnected from executionPipeline 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 developerdocs/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 ─────────────┘
PhaseCommandWhat it does
Define/workflows:constitutionEstablish durable project principles in docs/constitution.md
Discover/workflows:ideateScan the repo, generate improvement ideas, rank them
Explore/workflows:brainstormEstablish WHY and WHAT — problem narrative, user story, architectural context
Plan/workflows:planStructured implementation plan with tasks, dependencies, and evidence contracts
Deepen/deepen-planExpand thin spots in the plan before execution begins
Execute/workflows:workTDD-driven execution with red-green-refactor evidence capture
Review/workflows:reviewMulti-agent code review using your configured reviewers
Compound/workflows:compoundDocument 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.

AgentPurpose
design-implementation-reviewerCompare live UI against Figma designs
design-iteratorIterative UI refinement through screenshot-analyze-improve cycles
figma-design-syncDetect and fix visual differences between implementation and Figma

Research (8)

Gather knowledge before and during implementation.

AgentPurpose
best-practices-researcherResearch external best practices and documentation
framework-docs-researcherGather framework/library documentation and examples
git-history-analyzerArchaeological analysis of git history
issue-intelligence-analystAnalyze GitHub issues for recurring patterns
learnings-researcherSearch past solutions in docs/solutions/
repo-research-analystResearch repository structure and conventions

Review (17)

Language-specific and cross-cutting code review.

AgentPurpose
agent-native-reviewerEnsure agent-native action parity
architecture-strategistSOLID compliance and clean architecture review
code-simplicity-reviewerComplexity elimination and readability
data-integrity-guardianDatabase migration and data model safety
data-migration-expertValidate data migrations against reality
deployment-verification-agentGo/No-Go deployment checklists
frontend-race-conditions-reviewerAsync UI race conditions and stale state
laravel-reviewerLaravel 11+ / PHP 8.3+ best practices
nestjs-reviewerNestJS simplicity, performance, security
pattern-recognition-specialistDesign patterns and anti-patterns
performance-oraclePerformance bottlenecks and algorithmic complexity
python-reviewerPythonic patterns, type safety, security
rust-reviewerOwnership correctness, unsafe discipline, zero-cost abstractions
schema-drift-detectorSchema snapshot and artifact drift detection
security-sentinelOWASP Top 10 security audit
typescript-reviewerType safety, modern patterns, maintainability
vue-reviewerVue.js/Nuxt best practices and performance

Workflow (3)

Automate development operations.

AgentPurpose
bug-reproduction-validatorSystematically reproduce and validate bug reports
pr-comment-resolverAddress PR review comments with requested changes
spec-flow-analyzerAnalyze specs for user flow completeness

Commands

Workflows

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

CommandDescription
/workflows:constitutionDefine project engineering principles
/workflows:ideateGenerate and rank improvement ideas
/workflows:brainstormExplore WHY and WHAT before planning
/workflows:planStructured implementation planning
/workflows:architectureArchitecture review and improvement
/workflows:workTDD-driven execution with evidence capture
/workflows:reviewMulti-agent code review
/workflows:compoundDocument solved problems as searchable knowledge
/workflows:compound-refreshRefresh stale learnings against current codebase

Operations

CommandDescription
/lfgFull autonomous pipeline: plan → work → review → ship
/slfgSame as /lfg but with parallel swarm execution
/ralph-loopRed-green-refactor loop with completion promise
/cancel-ralphCancel an active ralph loop
/triageTriage GitHub issues
/reproduce-bugSystematic bug reproduction
/resolve_parallelResolve PR comments in parallel
/resolve_todo_parallelResolve TODO items in parallel
/test-browserBrowser-based testing
/changelogGenerate changelog
/deepen-planExpand thin spots in a plan
/deploy-docsDeploy documentation site
/feature-videoRecord feature walkthrough video
/heal-skillRepair a broken skill
/create-agent-skillCreate a new agent or skill
/report-bugReport a bug with structured context
/generate_commandGenerate a new command

Skills

SkillPurpose
agent-browserBrowser automation via Vercel agent-browser CLI
agent-native-architectureBuild applications where agents are first-class citizens
agent-native-auditScored audit of agent-native architecture principles
brainstormingExplore user intent before implementing
cavemanUltra-compressed communication (~75% token reduction)
compound-docsCapture solved problems as searchable documentation
compound-refreshRefresh stale learnings against current codebase
create-agent-skillsExpert guidance for authoring coding agent skills
document-reviewRefine brainstorm/plan documents
file-todosFile-based todo tracking system
finishing-branchVerify quality and clean up feature branches
frontend-designProduction-grade frontend interfaces
gemini-imagegenImage generation via Gemini API
git-worktreeGit worktree management for parallel development
grill-meRelentlessly interview user about a plan or design
ideateGenerate and evaluate grounded improvement ideas
laravel-conventionsModern Laravel 11+ / PHP 8.3+ coding standards
orchestrating-swarmsMulti-agent swarm orchestration
rcloneCloud storage file management via rclone
resolve-pr-parallelResolve PR comments using parallel processing
setupConfigure review agents for your stack
skill-creatorGuide for creating effective skills
systematic-debuggingStructured 4-phase debugging methodology
ubiquitous-languageDDD-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

PlatformStatusInstall
Claude CodeSupportedclaude install-plugin
GitHub CopilotSupportedCopy agents + skills
OpenCodeSupportedCopy agents + skills
Any MCP-compatible agentSkills are plain markdownCopy 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.


Ver o original no GitHub

Bom para

Relacionados