oh-my-agent: Portable Multi-Agent Harness

한국어 | 中文 | Português | 日本語 | Français | Español | Nederlands | Polski | Русский | Deutsch

The portable, role-based agent harness for serious AI-assisted engineering.

Orchestrate 10 specialized domain agents (PM, Frontend, Backend, DB, Mobile, QA, Debug, Brainstorm, DevWorkflow, Terraform) via Serena Memory. oh-my-agent uses .agents/ as the source of truth for portable skills and workflows, then projects compatibility to other AI IDEs and CLIs. It combines role-based agents, explicit workflows, real-time observability, and standards-aware guidance for teams that want less AI slop and more disciplined execution.

Table of Contents

• Architecture
• Why Different
• Compatibility
• The .agents Spec
• What Is This?
• Quick Start
• Sponsors
• License

Architecture

flowchart TD
    subgraph Workflows["Workflows"]
        direction TB
        W0["/brainstorm"]
        W1["/coordinate"]
        W1b["/ultrawork"]
        W2["/orchestrate"]
        W3["/plan"]
        W4["/review"]
        W5["/debug"]
    end

    subgraph Orchestration["Orchestration"]
        direction TB
        PM[pm-agent]
        WF[workflow-guide]
        ORC[orchestrator]
    end

    subgraph Domain["Domain Agents"]
        direction TB
        FE[frontend-agent]
        BE[backend-agent]
        DB[db-agent]
        MB[mobile-agent]
        TF[tf-infra-agent]
    end

    subgraph Quality["Quality"]
        direction TB
        QA[qa-agent]
        DBG[debug-agent]
    end

    Workflows --> Orchestration
    Orchestration --> Domain
    Domain --> Quality
    Quality --> CMT([commit])

Loading

Why Different

• .agents/ is the source of truth: skills, workflows, shared resources, and config live in one portable project structure instead of being trapped inside one IDE plugin.
• Role-shaped agent teams: PM, QA, DB, Infra, Frontend, Backend, Mobile, Debug, and Workflow agents are modeled like an engineering org, not just a pile of prompts.
• Workflow-first orchestration: planning, review, debug, and coordinated execution are first-class workflows, not afterthoughts.
• Standards-aware by design: agents now carry focused guidance for ISO-driven planning, QA, database continuity/security, and infrastructure governance.
• Built for verification: dashboards, manifest generation, shared execution protocols, and structured outputs favor traceability over vibe-only generation.

Compatibility

oh-my-agent is designed around .agents/ and then bridges to other tool-specific skill folders when needed.

Tool / IDE	Skills Source	Interop Mode	Notes
Antigravity	.agents/skills/	Native	Primary source-of-truth layout
Claude Code	.claude/skills/ + .claude/agents/	Native + Adapter	Symlinks for domain skills + native workflow skills, subagents, and CLAUDE.md
OpenCode	.agents/skills/	Native-compatible	Uses the same project-level skill source
Amp	.agents/skills/	Native-compatible	Shares the same project-level source
Codex CLI	.agents/skills/	Native-compatible	Works from the same project skill source
Cursor	.agents/skills/	Native-compatible	Can consume the same project-level skill source
GitHub Copilot	.github/skills/	Optional symlink	Installed when selected during setup

See SUPPORTED_AGENTS.md for the current support matrix and interoperability notes.

Claude Code Native Integration

Claude Code has first-class native integration beyond symlinks:

• CLAUDE.md — project identity, architecture, and rules (auto-loaded by Claude Code)
• .claude/skills/ — 12 workflow skills mapped from .agents/workflows/ (e.g., /orchestrate, /coordinate, /ultrawork)
• .claude/agents/ — 7 subagent definitions spawned via Task tool (backend-impl, frontend-impl, mobile-impl, db-impl, qa-reviewer, debug-investigator, pm-planner)
• Native loop patterns — Review Loop, Issue Remediation Loop, and Phase Gate Loop using synchronous Task tool results instead of CLI polling

Domain skills (backend-agent, frontend-agent, etc.) remain as symlinks from .agents/skills/. Workflow skills are native SKILL.md files that reference the corresponding .agents/workflows/*.md source of truth.

The .agents Spec

oh-my-agent treats .agents/ as a portable project convention for agent skills, workflows, and shared context.

• Skills live in .agents/skills/<skill-name>/SKILL.md
• Shared resources live in .agents/skills/_shared/
• Workflows live in .agents/workflows/*.md
• Project config lives in .agents/config/
• CLI metadata and packaging stay aligned through generated manifests

See AGENTS_SPEC.md for the project layout, required files, interoperability rules, and source-of-truth model.

What Is This?

A collection of Agent Skills enabling collaborative multi-agent development. Work is distributed across expert agents with explicit roles, workflows, and verification boundaries:

Agent	Specialization	Triggers
Brainstorm	Design-first ideation before planning	"brainstorm", "ideate", "explore idea"
Workflow Guide	Coordinates complex multi-agent projects	"multi-domain", "complex project"
PM Agent	Requirements analysis, task decomposition, architecture	"plan", "break down", "what should we build"
Frontend Agent	React/Next.js, TypeScript, Tailwind CSS	"UI", "component", "styling"
Backend Agent	FastAPI, PostgreSQL, JWT authentication	"API", "database", "authentication"
DB Agent	SQL/NoSQL modeling, normalization, integrity, backup, capacity	"ERD", "schema", "database design", "index tuning"
Mobile Agent	Flutter cross-platform development	"mobile app", "iOS/Android"
QA Agent	OWASP Top 10 security, performance, accessibility	"review security", "audit", "check performance"
Debug Agent	Bug diagnosis, root cause analysis, regression tests	"bug", "error", "crash"
Developer Workflow	Monorepo task automation, mise tasks, CI/CD, migrations, release	"dev workflow", "mise tasks", "CI/CD pipeline"
TF Infra Agent	Multi-cloud IaC provisioning (AWS, GCP, Azure, OCI)	"infrastructure", "terraform", "cloud setup"
Orchestrator	CLI-based parallel agent execution with Serena Memory	"spawn agent", "parallel execution"
Commit	Conventional Commits with project-specific rules	"commit", "save changes"

Quick Start

Prerequisites

• AI IDE (Antigravity, Claude Code, Codex, Gemini, etc.)

Option 1: One-Line Install (Recommended)

curl -fsSL https://raw.githubusercontent.com/first-fluke/oh-my-agent/main/cli/install.sh | bash

Automatically detects and installs missing dependencies (bun, uv), then launches the interactive setup.

Option 2: Manual Install

# Install bun if you don't have it:
# curl -fsSL https://bun.sh/install | bash

# Install uv if you don't have it:
# curl -LsSf https://astral.sh/uv/install.sh | sh

bunx oh-my-agent

Select your project type and skills will be installed to .agents/skills/, with compatibility symlinks created under .agents/skills/ and .claude/skills/.

Preset	Skills
✨ All	Everything
🌐 Fullstack	brainstorm, frontend, backend, db, pm, qa, debug, commit
🎨 Frontend	brainstorm, frontend, pm, qa, debug, commit
⚙️ Backend	brainstorm, backend, db, pm, qa, debug, commit
📱 Mobile	brainstorm, mobile, pm, qa, debug, commit
🚀 DevOps	brainstorm, tf-infra, dev-workflow, pm, qa, debug, commit

Option 3: Global Installation (For Orchestrator)

To use the core tools globally or run the SubAgent Orchestrator:

bun install --global oh-my-agent

You'll also need at least one CLI tool:

CLI	Install	Auth
Gemini	bun install --global @google/gemini-cli	gemini auth
Claude	curl -fsSL https://claude.ai/install.sh | bash	claude auth
Codex	bun install --global @openai/codex	codex auth
Qwen	bun install --global @qwen-code/qwen	qwen auth

2. Chat

Explicit coordination (user-triggered workflow):

/coordinate
→ Step-by-step: PM planning → agent spawning → QA review

Complex project (workflow-guide coordinates):

"Build a TODO app with user authentication"
→ workflow-guide → PM Agent plans → agents spawned in Agent Manager

Simple task (single agent auto-activates):

"Create a login form with Tailwind CSS and form validation"
→ frontend-agent activates

Commit changes (conventional commits):

/commit
→ Analyze changes, suggest commit type/scope, create commit with Co-Author

3. Monitor with Dashboards

For dashboard setup and usage details, see web/content/en/guide/usage.md.

Sponsors

This project is maintained thanks to our generous sponsors.

Like this project? Give it a star!

gh api --method PUT /user/starred/first-fluke/oh-my-agent

Try our optimized starter template: fullstack-starter

🚀 Champion

🛸 Booster

☕ Contributor

Become a sponsor →

See SPONSORS.md for a full list of supporters.

Star History

License

MIT