AI coding agent system for learning purposes — organized and consolidated from oh-my-claudecode, oh-my-opencode, and oh-my-customcode.
Note: This project is a simplified version for learning purposes. It consolidates patterns and configurations from the above projects for educational reference. For production use, additional hardening and customization are required.
Main Conversation (orchestrator)
├─ routing skill → selects agent based on intent
├─ Task tool → spawns subagent (flat, no hierarchy)
└─ skills → provide domain knowledge and workflows
The main conversation is the sole orchestrator. All implementation work is delegated to subagents via the Task tool. Subagents cannot spawn other subagents.
plan → build → test → review → commit
See ORCHESTRATION.md for stage details, handoff contracts, and error recovery.
| Agent | Role | Access |
|---|---|---|
planner |
Requirements, design, architecture planning | Read-only |
builder |
Code implementation, language/framework patterns | Read/Write |
reviewer |
Code review, security audit, verification | Read-only |
tester |
Testing strategy, test writing, QA | Read/Write |
debugger |
Bug reproduction, root cause analysis | Read/Write |
explorer |
Codebase search, documentation | Read-only |
manager |
Git operations, releases, agent management | Read/Write |
| Skill / Command | Purpose |
|---|---|
autopilot · /autopilot |
Autonomous execution modes (standard, ralph, ultrapilot, ecomode) |
plan · /plan |
Planning, analysis, deep search, research |
review · /review |
Two-stage code review with severity ratings |
tdd · /tdd |
Test-driven development (RED → GREEN → REFACTOR) |
best-practices · /analyze |
Domain guides for languages, frameworks, infrastructure |
routing |
Intent detection, task routing, result aggregation |
memory |
Persistent knowledge management across sessions |
system · /help |
Setup, init, HUD, status, diagnostics |
git · /release |
Git workflow automation, releases, GitHub integration |
parallel · /pipeline · /team |
Parallel execution, pipeline, team, swarm modes |
/research |
Parallel research and documentation |
| Rule | Scope |
|---|---|
safety |
Git/code/file/dependency safety, hard stops |
orchestration |
Delegation, parallel execution, pipeline management |
interaction |
Communication style, output formatting, escalation |
quality |
Code quality, testing standards, verification protocol |
oh-my-codes/
├── CLAUDE.md # Project configuration
└── .claude/
├── agents/ # 7 agent definitions
│ ├── planner.md
│ ├── builder.md
│ ├── reviewer.md
│ ├── tester.md
│ ├── debugger.md
│ ├── explorer.md
│ └── manager.md
├── skills/ # 10 skill modules
│ ├── autopilot/SKILL.md
│ ├── plan/SKILL.md
│ ├── review/SKILL.md
│ ├── tdd/SKILL.md
│ ├── best-practices/
│ │ ├── SKILL.md
│ │ └── guides/
│ │ ├── languages.md # TS, Python, Go, Rust, Kotlin, Java
│ │ ├── frameworks.md # React/Next.js, FastAPI, Express, NestJS, etc.
│ │ └── infrastructure.md # Docker, AWS, PostgreSQL, Redis, Kafka, etc.
│ ├── routing/SKILL.md
│ ├── memory/SKILL.md
│ ├── system/SKILL.md
│ ├── git/SKILL.md
│ └── parallel/SKILL.md
├── commands/ # 10 slash commands
│ ├── analyze.md
│ ├── autopilot.md
│ ├── plan.md
│ ├── review.md
│ ├── tdd.md
│ ├── research.md
│ ├── release.md
│ ├── help.md
│ ├── pipeline.md
│ └── team.md
├── rules/ # 4 behavioral rule sets
│ ├── safety.md
│ ├── orchestration.md
│ ├── interaction.md
│ └── quality.md
└── hooks/
└── hooks.json
- Orchestrator-only delegation — Main conversation delegates; subagents execute
- Flat hierarchy — No nested agent spawning
- Smallest viable diff — Minimal, surgical changes
- Test-first — Failing tests before implementation
- Fail fast — 3 failures → escalate to user