A production-grade multi-agent AI coding studio for OpenCode
Transform your terminal into a powerful AI engineering environment with 34 specialist agents, 71 domain skills, automated workflows, custom commands, plugins, and full multi-agent orchestration via oh-my-opencode.
| Component | Count | Description |
|---|---|---|
| Specialist Agents | 34 | Domain-expert AI assistants with scoped permissions |
| Domain Skills | 71 | Reusable instruction sets loaded on demand |
| omo Personas | 6 | Multi-agent orchestration system (Sisyphus, Prometheus, Atlas…) |
| Slash Commands | 9 | One-liner task automation |
| Custom Tools | 2 | TypeScript tools callable by agents |
| Plugins | 3 | Notifications, env injection, env protection |
| Workflows | 4 | Automated multi-step sequences |
| Validation Scripts | 7 | Python audit tools |
| MCPs | 4 | Context7, Shadcn, Netlify, Daytona |
curl -fsSL https://opencode.ai/install | bashgit clone https://github.com/skeletorflet/opencode-kit
cd opencode-kitcp .env.example .env
# Edit .env and add your Anthropic/OpenAI API keyopencodeEverything in .opencode/ is auto-loaded. No manual setup required.
This kit integrates oh-my-opencode, a community plugin that transforms OpenCode into a full multi-agent orchestration system.
| Agent | Role | Use When |
|---|---|---|
| Sisyphus | Orchestrator — plans, delegates, executes in parallel | Starting any complex task |
| Prometheus | Strategic planner — interviews you, defines scope, writes the plan | Before writing any code |
| Atlas | Plan executor — reads Prometheus plans and runs them task by task | After a plan is approved |
| Oracle | Architecture consultant (read-only) — never writes code, only advises | Architecture decisions |
| Librarian | Documentation researcher — fast lookup via lightweight model | Docs and library research |
| Explore | Codebase explorer (read-only) — maps structure fast | Understanding an unfamiliar repo |
| Hephaestus | Deep autonomous worker — explores and executes end-to-end | Long autonomous tasks |
Never use Atlas without a Prometheus plan. Prometheus plans. Atlas executes.
# 1. Start OpenCode with Sisyphus as orchestrator
opencode
# 2. Activate max performance mode
ultrawork (or: ulw)
# 3. Let Prometheus plan the feature
@prometheus "build a JWT auth system with refresh tokens"
# Prometheus interviews you, then saves the plan to .sisyphus/plans/
# 4. Atlas executes the plan
/start-work
# 5. Consult Oracle on architecture (no code touched)
@oracle "should I use event sourcing here?"
# 6. Loop until done
/ralph-loop # iterates until all TODOs are complete
/ulw-loop # ultrawork + ralph together
| Command | Description |
|---|---|
ultrawork / ulw |
Maximum parallel orchestration mode |
ultrathink |
Extended thinking (32k token budget) |
search / find |
Runs Explore + Librarian in parallel |
analyze / investigate |
Deep multi-agent analysis |
/start-work |
Atlas picks up the plan and executes |
/ralph-loop |
Iterates until all TODOs are done |
/ulw-loop |
Ultrawork + Ralph combined |
/init-deep |
Generates hierarchical AGENTS.md in subdirectories |
npx oh-my-opencode doctor |
Diagnostics |
| Feature | Config | Description |
|---|---|---|
| Hashline Edit | hashline_edit: true |
Hash-verified edits — zero stale-line corruption |
| Ralph Loop | ralph_loop.enabled: true |
Agents don't stop until every TODO is complete |
| Compaction guard | preemptive-compaction disabled |
Stable context window management |
Invoked with @agent-name in the TUI. All use OpenCode's native mode + permission frontmatter.
| Agent | Description |
|---|---|
orchestrator |
Multi-agent coordination, task breakdown, delegation |
project-planner |
4-phase planning methodology, task-slug files |
explorer-agent |
Codebase discovery, architecture analysis |
debugger |
Systematic root-cause analysis, error tracing |
review |
Code review, PR audits, quality checks |
code-archaeologist |
Legacy code analysis and modernization paths |
| Agent | Domain |
|---|---|
frontend-specialist |
React, Next.js, TypeScript, Tailwind, Core Web Vitals |
accessibility-specialist |
WCAG 2.2, ARIA, screen readers, inclusive design |
seo-specialist |
SEO audits, E-E-A-T, GEO, structured data |
| Agent | Domain |
|---|---|
backend-specialist |
Node.js, Python, serverless, edge systems |
database-architect |
Schema design, migrations, indexing, Prisma |
data-engineer |
ETL pipelines, data warehouses, Kafka, dbt |
graphql-developer |
GraphQL APIs, Apollo, federation |
ml-engineer |
ML pipelines, model serving, RAG, fine-tuning |
cloud-architect |
AWS, Azure, GCP, serverless, containers |
devops-engineer |
CI/CD, Docker, monitoring, infrastructure |
| Agent | Language / Focus |
|---|---|
go-developer |
Go 1.22+, APIs, microservices, CLI tools |
java-developer |
Java 21+, Spring Boot, Quarkus, Micronaut |
dotnet-developer |
.NET 8, ASP.NET Core, Blazor, Minimal APIs |
kotlin-developer |
Kotlin 2.0, Ktor, Jetpack Compose |
ruby-developer |
Ruby 3.3+, Rails 7, Hanami |
php-developer |
PHP 8.3, Laravel 11, Symfony 7 |
cpp-developer |
C++23/26, systems programming, game engines |
| Agent | Domain |
|---|---|
mobile-developer |
React Native, Flutter, Expo |
mobile-native |
Swift/SwiftUI (iOS), Kotlin/Compose (Android) |
game-developer |
Unity, Godot, Unreal, Phaser, Three.js |
| Agent | Domain |
|---|---|
security-auditor |
OWASP 2025, supply chain, zero trust |
penetration-tester |
Red team, vulnerability exploitation |
test-engineer |
TDD, unit/E2E testing, coverage targets |
qa-automation-engineer |
Playwright, Cypress, CI pipelines |
performance-optimizer |
Profiling, Core Web Vitals, bundle analysis |
| Agent | Domain |
|---|---|
product-manager |
Requirements, user stories, acceptance criteria |
product-owner |
Roadmap, backlog, MVP, stakeholder management |
documentation-writer |
README, API docs, tutorials, changelogs |
Skills are SKILL.md instruction sets loaded on demand by agents. Organized by category.
clean-code · naming-conventions · code-review-checklist · systematic-debugging · tdd-workflow · testing-patterns · plan-writing · behavioral-modes · brainstorming · lint-and-validate
frontend-design · react-best-practices · nextjs-react-expert · tailwind-patterns · web-design-guidelines · ui-ux-pro-max
nodejs-best-practices · python-patterns · go-patterns · go-concurrency · java-expert · spring-boot · dotnet-expert · aspnet-core · kotlin-expert · ruby-patterns · rails-patterns · php-patterns · laravel-patterns · django-patterns · nestjs-expert
database-design · prisma-expert
docker-expert · deployment-procedures · server-management · bash-linux · powershell-windows
vulnerability-scanner · red-team-tactics · webapp-testing · performance-profiling · seo-fundamentals · geo-fundamentals
mobile-design · android-compose · swift-expert · game-development · 2d-games · 3d-games · pc-games · mobile-games · web-games · multiplayer · game-design · game-art · game-audio · vr-ar
architecture · api-patterns · mcp-builder · parallel-agents · intelligent-routing · app-builder · i18n-localization · documentation-templates · systems-programming · rust-pro · cpp-modern · realtime-collaborative · ai-agent-integration
Run inside the TUI with /command-name [args].
| Command | Agent | Description |
|---|---|---|
/plan |
project-planner |
Break down a feature into a structured task plan |
/create |
orchestrator |
Scaffold a new application or module |
/enhance |
frontend-specialist |
Improve code quality, performance, or design |
/test |
test-engineer |
Generate tests for a file or feature |
/audit-security |
security-auditor |
Full project security audit |
/deploy-check |
project-planner |
Verify the project is ready to deploy |
/fix-types |
debugger |
Find and fix TypeScript type errors |
/gen-docs |
documentation-writer |
Generate documentation for a file |
/review-pr |
review |
Review branch changes before opening a PR |
Auto-loaded from .opencode/plugins/ at startup. Written in TypeScript, running on Bun.
| Plugin | Description |
|---|---|
notification.ts |
Windows toast notification when session goes idle (native PowerShell, TUI-safe) |
inject-env.ts |
Injects PROJECT_ROOT and OPENCODE_KIT_ACTIVE into every shell execution |
env-protection.ts |
Blocks agents from reading .env files directly |
TypeScript tools callable by agents via the OpenCode tool API.
| Tool | Description |
|---|---|
read-env |
Lists environment variable names from .env files — values are never exposed |
run-script |
Runs a package.json script and returns output with proper error handling |
Automated multi-step sequences. Run from the TUI or referenced by agents.
| Workflow | Description |
|---|---|
01-project-kickoff |
From idea → planned → scaffolded → first commit |
02-tdd-feature |
Red → Green → Refactor TDD development loop |
03-pre-flight-audit |
Security · lint · build · test · commit before any deploy |
04-omo-main-flow |
Full omo orchestration: Sisyphus → Prometheus → Atlas → Oracle |
Model Context Protocol servers configured in opencode.json.
| MCP | Description |
|---|---|
| Context7 | Live documentation for any library, injected directly into context |
| Shadcn | Shadcn/ui component registry and management |
| Netlify | Deploy and manage Netlify projects from the TUI |
| Daytona | Isolated sandbox environments (requires DAYTONA_API_KEY) |
Python scripts for quality assurance. Run via python .opencode/scripts/<script>.py.
| Script | When to Use |
|---|---|
verify_all.py |
Comprehensive full-stack verification suite |
checklist.py |
Priority-based project audit (Security → Lint → Schema → Test → UX → SEO) |
lint_runner.py |
Code linting automation |
test_runner_multi.py |
Multi-framework test execution |
security_scan.py |
OWASP vulnerability scanning |
ux_audit.py |
UI/UX quality audit |
auto_preview.py |
Preview management |
Copy .env.example to .env and fill your keys. The inject-env.ts plugin makes them available to all agent shell executions automatically.
cp .env.example .envKey variables:
| Variable | Description |
|---|---|
ANTHROPIC_API_KEY |
Anthropic Claude API key |
OPENAI_API_KEY |
OpenAI API key |
CONTEXT7_API_KEY |
Context7 MCP (Upstash) |
DAYTONA_API_KEY |
Daytona sandbox MCP |
DAYTONA_SERVER_URL |
Daytona server endpoint |
opencode-kit/
├── .env.example ← Environment variable template
├── opencode.json ← Main OpenCode config (MCPs, agents, permissions)
├── AGENTS.md ← Global agent routing rules
├── DOC/ ← OpenCode reference documentation (41 files)
└── .opencode/
├── agents/ ← 34 specialist agents (.md)
├── skills/ ← 71 domain skills (SKILL.md)
├── commands/ ← 9 slash commands (.md)
├── tools/ ← 2 custom tools (.ts)
├── plugins/ ← 3 plugins (.ts)
├── workflows/ ← 4 automated workflows (.md)
├── scripts/ ← 7 Python validation scripts
├── prompts/ ← System prompt templates
├── rules/RULES.md ← Global behavior rules (P0 priority)
├── oh-my-opencode.jsonc ← omo multi-agent config
├── ARCHITECTURE.md ← Technical architecture overview
└── package.json ← Plugin dependencies (Bun)
The kit enforces automatic agent selection via RULES.md. When you write a request, the AI:
- Classifies the request type (question / code / design / security…)
- Selects the best specialist agent(s) automatically
- Announces
🤖 Applying knowledge of @[agent-name]... - Loads required skills from the agent's frontmatter
- Applies the agent's specific rules and coding standards
Priority: RULES.md (P0) > Agent .md (P1) > SKILL.md (P2)
| Resource | Link |
|---|---|
| OpenCode | opencode.ai |
| OpenCode GitHub | github.com/anomalyco/opencode |
| oh-my-openagent | github.com/code-yeongyu/oh-my-openagent |
| OpenCode Discord | opencode.ai/discord |
| Context7 | context7.com |
Read ARCHITECTURE.md for the full technical overview.
