Skip to content
/
Tier B A6 Multi-agent orchestrator

huangdijia/oh-my-claude-code-plugins

huangdijia-cc-plugins · huangdijia/oh-my-claude-code-plugins · ★ 10 · last commit 2026-03-31

Primitive shape 25 total
Commands 8 Skills 2 Subagents 7 Hooks 3 MCP tools 5
00

Summary

huangdijia/oh-my-claude-code-plugins — Summary

oh-my-claude-code-plugins by huangdijia is a Claude Code plugin marketplace hosting 5 separate plugins: subagents (7 specialist agents), git (feature/hotfix branch commands), git-flow (PR/MR creation, commit generation, git hooks), mcp (5 pre-configured MCP servers), and skills (2 workflow skills). The system implements a spec-driven development workflow: strategic-planner creates specs/<feature>/requirements.md + design.md + tasks.md, task-executor implements one task at a time with checkbox tracking, and code-reviewer validates quality. Three git automation hooks (php-cs-fixer, auto-git-add, smart-commit) fire on file edits. The MCP plugin bundles context7, sequential-thinking, memory, tavily-mcp, and chrome-devtools servers. With 10 stars and bilingual documentation, this is a well-structured multi-plugin Claude Code marketplace.

Differs from seeds: Closest to kiro (spec-driven development with requirements/design/tasks triplet) and agent-os (installation scripts + command pack), but uniquely organizes as a multi-plugin marketplace with per-domain plugin separation, includes PHP-specific git hooks, and bundles a wider MCP surface (including chrome-devtools) than any seed except claude-flow.

01

Overview

huangdijia/oh-my-claude-code-plugins — Overview

Origin

Created by huangdijia (Deeka Wong per agent file author fields) as a curated collection of Claude Code plugins. 10 stars, bilingual documentation (English + Chinese). The agent files include author: "Deeka Wong" and license: MIT in YAML frontmatter.

Philosophy

Spec-driven development with a three-phase workflow: Planning → Execution → Review. The strategic-planner enforces EARS notation for acceptance criteria.

"This plugin implements a three-phase development workflow: Planning Phase → Execution Phase → Review Phase."

The design principle is surgical execution: task-executor handles one task at a time, checking off [ ] → [x], and supports an autonomous mode for unattended execution.

Multi-Plugin Architecture

The marketplace hosts 5 independently installable plugins:

  • subagents — the 7 specialist agents
  • git — basic Git Flow branch management
  • git-flow — PR/MR workflows + commit generation + hooks
  • mcp — 5 pre-configured MCP servers
  • skills — 2 development skills

Users install only what they need.

Notable Technical Choice

The strategic-planner enforces EARS (Easy Approach to Requirements Syntax) for acceptance criteria — the same notation used by Kiro IDE. This is a cross-pollination with the spec-driven development ecosystem.

Language Targeting

The php-cs-fixer git hook reveals PHP project usage — one of the few frameworks in this batch with PHP-specific tooling.

02

Architecture

huangdijia/oh-my-claude-code-plugins — Architecture

Distribution

  • Type: Claude Code plugin marketplace (multi-plugin)
  • License: No SPDX (unlicensed/unknown)
  • Stars: 10

Install

# Add marketplace
claude plugin marketplace add huangdijia/oh-my-claude-code-plugins

# Install individual plugins
claude plugin install subagents@oh-my-claude-code-plugins
claude plugin install git@oh-my-claude-code-plugins
claude plugin install git-flow@oh-my-claude-code-plugins
claude plugin install mcp@oh-my-claude-code-plugins
claude plugin install skills@oh-my-claude-code-plugins

Directory Tree

oh-my-claude-code-plugins/
├── .claude-plugin/
│   └── marketplace.json
├── .claude/              # Global Claude config
├── plugins/
│   ├── subagents/
│   │   ├── .claude-plugin/plugin.json
│   │   └── agents/
│   │       ├── strategic-planner.md
│   │       ├── task-executor.md
│   │       ├── code-reviewer.md
│   │       ├── steering-architect.md
│   │       ├── data-scientist.md
│   │       ├── debugger.md
│   │       └── prd-writer.md
│   ├── git/
│   │   ├── .claude-plugin/
│   │   └── commands/
│   │       ├── feature.md
│   │       └── hotfix.md
│   ├── git-flow/
│   │   ├── .claude-plugin/
│   │   ├── commands/
│   │   │   ├── commit:create.md
│   │   │   ├── pr:create.md
│   │   │   ├── pr:review.md
│   │   │   ├── mr:create.md
│   │   │   ├── mr:review.md
│   │   │   └── status.md
│   │   └── hooks/
│   │       ├── php-cs-fixer.json
│   │       ├── auto-git-add.json
│   │       └── smart-commit.json
│   ├── mcp/
│   │   └── .claude-plugin/
│   └── skills/
│       └── .claude-plugin/
│           ├── feature-dev/SKILL.md
│           └── code-review/SKILL.md
└── README.md + README_CN.md

Required Runtime

  • Claude Code
  • PHP CS Fixer (for php-cs-fixer hook, optional)
03

Components

huangdijia/oh-my-claude-code-plugins — Components

Plugin: subagents — 7 Agents

Agent Purpose
strategic-planner Expert architect; creates EARS-format requirements.md, design.md with Mermaid diagrams, tasks.md
task-executor Surgical task implementer; one task at a time with checkbox tracking; supports autonomous mode
code-reviewer Code quality, security, maintainability review
steering-architect Project analysis and documentation setup (.ai-rules/ files)
data-scientist SQL/BigQuery/data analysis
debugger Error troubleshooting and root cause analysis
prd-writer Product requirements document generation

All agents use YAML frontmatter with name, version, description, tags, tools, permissions, categories, author, license.

Plugin: git — 2 Commands

Command Purpose
/git:feature <name> Create feature branch from develop
/git:hotfix <name> Create hotfix branch from main

Plugin: git-flow — 6 Commands + 3 Hooks

Command Purpose
/git-flow:pr:create Create GitHub Pull Request
/git-flow:pr:review 123 PR code review
/git-flow:mr:create Create GitLab Merge Request
/git-flow:mr:review 456 [scope] MR review (performance/security/style)
/git-flow:commit:create Generate semantic commit from staged changes
/git-flow:status Git Flow status display

Git Hooks:

Hook Event Purpose
php-cs-fixer.json PostToolUse (Write/Edit) Auto-format PHP files
auto-git-add.json PostToolUse (Edit) Automatically stage modified files
smart-commit.json PostToolUse Intelligent commit message generation

Plugin: mcp — 5 MCP Servers

Server Purpose
context7 Up-to-date documentation retrieval
sequential-thinking Structured problem-solving (chain of thought)
memory Persistent knowledge graph
tavily-mcp Web search + content extraction
chrome-devtools Browser automation and debugging

Plugin: skills — 2 Skills

Skill Purpose
feature-dev Guided feature development (codebase understanding, architecture)
code-review Advanced code review (quality, security, maintainability)
05

Prompts

huangdijia/oh-my-claude-code-plugins — Prompt Excerpts

Excerpt 1: strategic-planner Agent (Chinese/English)

Technique: Interactive multi-phase workflow with explicit phase gating

---
name: strategic-planner
version: "1.0.0"
description: 专家级软件架构师和协作规划师。负责功能需求分析、技术设计和任务规划。
tags: [planning, architecture, design, requirements]
author: "Deeka Wong"
license: MIT
---

# 角色：专家级AI软件架构师与协作规划师

## 规则
- 规划模式：仅限问答 — 绝对不允许编写代码，不做文件更改。
- 例外情况：你可以创建或修改 requirements.md、design.md 和 tasks.md 文件

## 工作流程

### 第1阶段：需求定义（互动循环）
1. 生成草稿 → requirements.md（EARS格式）
2. 审查和完善 → 提出具体澄清问题
3. 最终确定 → 保存并征求确认进入设计阶段

### 第2阶段：技术设计（互动循环）
1. 生成草稿 → design.md（完整技术蓝图 + Mermaid图表）
2. 识别关键架构决策 → 提出替代方案比较
3. 最终确定 → 保存并确认进入任务生成阶段

### 第3阶段：任务生成
(...)

Analysis: Strict mode-switching ("PLAN ONLY, no code"), explicit phase gating ("do NOT proceed until user explicitly approves"), EARS requirement notation, and Mermaid diagrams in design phase. The interactive clarification loop (ask→answer→update draft) is unusual — most frameworks generate specs all at once.

Excerpt 2: task-executor Agent

Technique: One-task-at-a-time constraint + autonomous mode flag

---
name: task-executor
description: AI软件工程师，专注于执行单个具体任务。具有外科手术般的精确度，严格按照任务清单逐项实现。
tags: [execution, implementation, coding, testing]
---

# 前言：执行模式 — 一次一个任务

你的重点是手术般的精确度。你将每次执行一个且仅一个任务。

## 自主模式
如果用户明确表示希望你自主继续任务：
- 跳过用户审核要求：实现后立即将任务标记为完成
- 继续下一个任务：完成一个任务后，自动进行下一个未勾选的任务
- 仅在出错时停止

## 指令
1. 识别任务：打开specs//tasks.md并找到第一个未勾选（[ ]）的任务
2. 理解任务：参考design.md和requirements.md
3. 实施更改：应用原子代码更改
4. 验证：运行相关测试
5. 更新任务状态：从[ ] → [x]
6. 停止：等待用户审核（除非自主模式）

Analysis: The task-executor is deliberately constrained to ONE task per invocation (atomic execution). The autonomous mode flag is an explicit override that skips user review gates — a "trust the agent" escape hatch. The specs/<feature-name>/ path injection makes it context-aware per feature.

09

Uniqueness

huangdijia/oh-my-claude-code-plugins — Uniqueness

Differs from Seeds

Closest to kiro (spec-driven development with requirements/design/tasks triplet + EARS notation) and agent-os (install scripts + command pack). However, huangdijia uniquely: (1) uses EARS requirement notation (borrowed from Kiro) but implements it in a Claude Code plugin rather than a proprietary IDE; (2) the auto-git-add hook automatically stages every file edit — the most aggressive git automation hook in the batch, absent from all seeds; (3) bundles chrome-devtools MCP server (browser automation) alongside documentation/search MCPs — a broader MCP surface than any seed; (4) organizes as a multi-plugin marketplace where each domain (agents, git, git-flow, mcp, skills) is separately installable.

The strategic-planner's interactive multi-phase process (ask question → update draft → user approves → next phase) is also distinct: most frameworks generate specs in one pass, but huangdijia's planner iterates with the user at each phase gate.

Positioning

  • Strongest spec-driven development workflow in this "oh-my-*" batch (alongside the parent Kiro-inspired pattern)
  • PHP-specific tooling (php-cs-fixer hook) reveals backend/PHP developer target audience
  • Multi-plugin marketplace structure is unique in this batch — all others ship as a single monolithic plugin

Observable Failure Modes

  1. auto-git-add danger: Auto-staging every edit can accidentally include unintended files in commits; no exclusion rules visible
  2. EARS adoption friction: EARS notation requires users to understand the format; the planner doesn't explain it clearly in README
  3. Phase-gate fatigue: 4 explicit approval gates (requirements → design → tasks → implementation) may feel bureaucratic for small features
  4. PHP-specific hooks: php-cs-fixer hook requires php-cs-fixer CLI installed; non-PHP projects will encounter hook failures
  5. MCP server dependencies: 5 MCP servers each require separate API keys (tavily), browser setup (chrome-devtools), or server setup (memory graph) — significant setup overhead
04

Workflow

huangdijia/oh-my-claude-code-plugins — Workflow

Spec-Driven Development (3-Phase)

Phase 1: Strategic Planning
  → Use strategic-planner agent
  → Interactive 3-phase interview:
      Phase 1: Requirements Definition (EARS format) → requirements.md
      Phase 2: Technical Design (Mermaid diagrams) → design.md
      Phase 3: Task Generation → tasks.md
  → User must approve each phase before proceeding

Phase 2: Task Execution
  → Use task-executor agent
  → Reads tasks.md
  → Implements ONE task at a time
  → Marks [ ] → [x] on completion
  → Runs tests before marking complete
  → Autonomous mode: skips user review gates

Phase 3: Quality Assurance
  → Use code-reviewer agent
  → Reviews code quality + security + maintainability

Strategic Planner Phase Flow

Initial Step: Check if new feature or existing
  → New: ask for kebab-case name, create specs/<name>/
  → Existing: load specs/<name>/*.md, ask which phase to refine

Phase 1 Requirements:
  → Generate EARS-format requirements.md draft
  → Interactive clarification questions
  → "Do NOT proceed to Phase 2 until user explicitly approves"

Phase 2 Design:
  → Generate technical blueprint (data models, APIs, components, Mermaid diagrams)
  → Present alternatives with tradeoffs
  → User approves before Phase 3

Phase 3 Tasks:
  → Generate granular task checklist
  → User approves

Phase-to-Artifact Map

Phase Agent Artifact
Requirements strategic-planner specs/<name>/requirements.md
Design strategic-planner specs/<name>/design.md
Tasks strategic-planner specs/<name>/tasks.md
Implementation task-executor Code changes + updated task checkboxes
Review code-reviewer Review report

Approval Gates (4)

  1. User approves requirements before design phase
  2. User approves design before task generation
  3. User approves tasks before execution
  4. Tests must pass before task marked complete
06

Memory Context

huangdijia/oh-my-claude-code-plugins — Memory & Context

State Storage

  • Type: File-based, per-feature
  • Spec files: specs/<feature-name>/requirements.md, design.md, tasks.md
  • Project rules: .ai-rules/product.md, .ai-rules/tech.md, .ai-rules/structure.md

Project Context Files (.ai-rules/)

The steering-architect agent creates .ai-rules/ files that all agents load:

  • .ai-rules/product.md — product vision
  • .ai-rules/tech.md — technology stack
  • .ai-rules/structure.md — project conventions
  • .ai-rules/*.md — any custom rules

Every agent reads these files at startup for global project context.

MCP Memory Server

The memory MCP server (from the mcp plugin) provides a persistent knowledge graph for conversation context across sessions.

Task Progress

  • tasks.md checkbox state ([ ] → [x]) provides visible task progress
  • Persists across sessions in the project directory

Cross-Session Handoff

  • Yes: spec files + task checklist survive session restarts
  • .ai-rules/ files inject project context on every session start

Context Injection Pattern

Strategic-planner and task-executor reference @.ai-rules/product.md, @.ai-rules/tech.md, @.ai-rules/structure.md — these are file reference annotations that Claude Code resolves when loading the agent.

07

Orchestration

huangdijia/oh-my-claude-code-plugins — Orchestration

Multi-Agent Pattern

Pattern: Sequential (phase-gated pipeline, not parallel)

strategic-planner → (user approval gates) → task-executor → code-reviewer

Agents do not run in parallel — each phase must complete and be approved before the next begins.

Spec-Driven Architecture

The specs/<feature-name>/ directory acts as the shared state:

  • strategic-planner writes the specs
  • task-executor reads and executes from specs
  • code-reviewer reads final code + specs to verify

Git Automation Hooks (3)

Hook Trigger Action
php-cs-fixer PostToolUse (Write/Edit) Auto-format PHP files via php-cs-fixer CLI
auto-git-add PostToolUse (Edit) Automatically git add modified files
smart-commit PostToolUse Generate semantic commit from staged changes

These are the most aggressive git automation hooks in this batch — auto-staging files on every edit is a significant behavioral change.

Isolation Mechanism

  • None: All agents edit in the working directory
  • No git worktrees

Execution Mode

  • Interactive loop with explicit approval gates between phases
  • Autonomous mode removes some gates for unattended execution

Multi-Model

  • No: All agents inherit the Claude model from the main session
  • No cross-provider routing

MCP Integration

5 MCP servers bundled in the mcp plugin:

  • chrome-devtools (browser automation) — most sophisticated MCP in this batch
  • sequential-thinking (chain of thought)
  • memory (persistent knowledge graph)
  • context7 (documentation)
  • tavily-mcp (web search)
08

Ui Cli Surface

huangdijia/oh-my-claude-code-plugins — UI/CLI Surface

CLI Binary

  • Exists: No dedicated binary
  • Installation via Claude Code plugin marketplace commands

Slash Commands

Command Plugin Purpose
/git:feature <name> git Create feature branch
/git:hotfix <name> git Create hotfix branch
/git-flow:pr:create git-flow Create GitHub PR
/git-flow:pr:review 123 git-flow Review PR
/git-flow:mr:create git-flow Create GitLab MR
/git-flow:mr:review 456 [scope] git-flow Review MR
/git-flow:commit:create git-flow Generate semantic commit
/git-flow:status git-flow Git Flow status

Marketplace Structure

# Each plugin installable independently
claude plugin install subagents@oh-my-claude-code-plugins
claude plugin install git@oh-my-claude-code-plugins
claude plugin install git-flow@oh-my-claude-code-plugins
claude plugin install mcp@oh-my-claude-code-plugins
claude plugin install skills@oh-my-claude-code-plugins

Local UI

  • Exists: No
  • Terminal only via Claude Code

Documentation

  • README.md (English) + README_CN.md (Chinese)
  • Well-documented workflow diagrams in README

MCP Dashboard (via chrome-devtools)

The chrome-devtools MCP server provides browser automation — users can have Claude debug web UIs through this MCP, though it's not a local dashboard for the framework itself.

Related frameworks

same archetype · same primary tool · same memory type

Claude-Flow / Ruflo ★ 55k
A6 Multi-agent orchestrator

Eliminates single-agent context limits and sequential bottlenecks by orchestrating fault-tolerant swarms of specialized AI agents…

Hermes Agent (NousResearch) ★ 168k
A6 Multi-agent orchestrator

Self-improving personal AI agent with closed learning loop, 7 terminal backends, and messaging gateway — not tied to any AI…

OpenCode ★ 165k
A6 Multi-agent orchestrator

Terminal-first AI coding agent with multi-model routing, native desktop app, and a typed .opencode/ configuration system for…

OpenHands ★ 75k
A6 Multi-agent orchestrator

Open-source AI software development platform (open-source Devin alternative) with Docker sandbox isolation, 77.6% SWE-bench…

DeerFlow ★ 70k
A6 Multi-agent orchestrator

Long-horizon superagent that researches, codes, and creates by orchestrating parallel sub-agents with isolated contexts in Docker…

oh-my-openagent (omo) ★ 60k
A6 Multi-agent orchestrator

Multi-provider AI agent orchestration for OpenCode: escape vendor lock-in by routing Sisyphus (Claude/Kimi/GLM) and Hephaestus…