AI Agent Workflow System for Product Managers — Powered by Claude Code Skills
3-Layer Core + Validation Loop · 6 Subagents · 12 Professional Skills · Complete Product Lifecycle
Quick Start • Skills • Subagents • 中文文档
Oh-My-PM is a comprehensive AI Agent workflow system for Product Managers, built as a Claude Code Plugin. Instead of building a standalone SaaS platform, it leverages AI Agent plugins and workflow orchestration to automate and enhance product management tasks.
| Feature | Description |
|---|---|
| 3-Layer Core + Validation | Main line Perception → Strategy → Design, with validation skills closing the feedback loop |
| Scenario-Driven PRD | Iteration updates / New features / 0-1 products — no random assumptions |
| Industry Benchmarking | Auto-benchmark against best practices for professional output |
| HTML Prototype Generation | Generate interactive HTML prototypes from PRDs |
| Human-AI Collaboration | Autopilot / Copilot / Manual modes |
- Claude Code or OpenAI Codex CLI tool
- Clone this repository to your local project directory
# Install all skills npx skills add kelegele/oh-my-pm -a claude-code # Install a single skill npx skills add kelegele/oh-my-pm --skill prd-gen -a claude-code # Install multiple skills npx skills add kelegele/oh-my-pm --skill prd-gen --skill competitive-analysis -a claude-code # List available skills npx skills add kelegele/oh-my-pm --list
# Install all skills npx skills add kelegele/oh-my-pm -a codex # Install a single skill npx skills add kelegele/oh-my-pm --skill prd-gen -a codex # List available skills npx skills add kelegele/oh-my-pm --list
| Channel | What you get | Notes |
|---|---|---|
npx skills add kelegele/oh-my-pm -a <agent> |
All 12 Skills — the complete PM workflow capability (intent routing, evidence discipline, execution profiles all live in the skills layer) | Works with any agent: Claude Code, Codex, Cursor, Copilot... |
| Claude Code marketplace plugin | Everything in the 12 Skills, plus 6 Subagents (isolated execution + persistent memory) and 4 slash Commands (quick entry points) | Enhanced experience for Claude Code |
# Clone repository git clone https://github.com/kelegele/oh-my-pm.git cd oh-my-pm
Note: This plugin installs via
--plugin-diror plugin marketplace. Skills trigger automatically through natural language—no manual configuration needed.
4 workflow Commands can be invoked directly:
# Direct invocation /quick-prd "User profile redesign" taobao jd /full-pm-cycle "New project management tool" /feature-launch "User registration flow" # Namespace invocation /ompm quick-prd "Dark mode" /ompm full-pm-cycle "AI assistant feature" /ompm feature-launch "Cart redesign" /ompm help # Show help
/ompm is a shim that activates the ompm skill — the global intent router: invoked with no arguments or with a natural-language request, it reads the current context plus docs/product/.ompm/current-workflow.json and recommends the most valuable next step.
12 Skills trigger automatically through natural language:
# Direct conversation—system identifies and calls the relevant Skill "Analyze the competitive differences between Notion and Feishu docs" # → competitive-analysis "Write a PRD for user profile redesign" # → prd-gen "How did this feature perform after launch?" # → retrospective "How should we position our product?" # → product-strategy
Note: Skills do not support /skill-name explicit invocation—they trigger through natural language or workflows.
The main line runs Perception → Strategy → Design; validation skills close the feedback loop. The former Delivery layer is folded in: PRD pre-review is now a prd-gen tail gate, and release checklists / rollout strategy live in feature-launch. On top sit 4 workflow skills: the ompm global intent router (front door) plus three orchestrators (quick-prd / full-pm-cycle / feature-launch).
┌─────────────────────────────────────────────────────────┐
│ Main Line 1: Perception 3 Skills │
│ Competitive Analysis · Market Research · Clarify Requirements│
└─────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────┐
│ Main Line 2: Strategy 1 Skill │
│ Product Strategy (positioning · prioritization · roadmap)│
└─────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────┐
│ Main Line 3: Design 2 Skills │
│ PRD Generation (with pre-review gate) · Prototype Design│
└─────────────────────────────────────────────────────────┘
↓
┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ┐
Auxiliary: Validation 2 Skills
Retrospective (impact + iteration) · Feedback Synthesis
──→ feedback loop back to Perception
└ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ┘
Subagents are specialized AI agents that run in isolated contexts with custom system prompts, specific tool access, and independent permissions. When Claude encounters tasks matching a subagent's description, it delegates to that subagent for processing.
| Feature | Skills | Subagents |
|---|---|---|
| Context | Shared with main conversation | Independently isolated |
| Model Selection | Inherits from main conversation | Configurable (haiku/sonnet/opus) |
| Tool Access | Unlimited | Configurable allowlist/denylist |
| Persistent Memory | None | Supports cross-session knowledge accumulation |
| Use Case | Simple prompt injection | High-capacity output isolation |
| Subagent | Model | Features | Trigger Examples |
|---|---|---|---|
market-researcher |
haiku | Worktree isolation | "Analyze X market", "Industry trends" |
competitive-analyst |
sonnet | GitHub code analysis | "Competitive analysis", "Compare XX and YY" |
user-interviewer |
sonnet | User research memory | "Create user persona", "User interviews" |
| Subagent | Model | Features | Trigger Examples |
|---|---|---|---|
impact-analyst |
sonnet | Performance analysis | "Launch performance", "How is it performing" |
feedback-collector |
haiku | Feedback aggregation | "User feedback", "What users are saying" |
| Subagent | Functionality | Trigger Examples |
|---|---|---|
pm-orchestrator |
Orchestrates complete PM cycle with parallel execution | "Complete product planning", "0-1 product" |
Each subagent has an independent memory directory (.claude/agent-memory/) for cross-session knowledge accumulation:
.claude/agent-memory/
├── market-researcher/ # Market data accumulation
├── competitive-analyst/ # Competitor knowledge base
├── user-interviewer/ # User research patterns
├── impact-analyst/ # Impact analysis framework
├── feedback-collector/ # Feedback theme patterns
└── pm-orchestrator/ # Workflow best practices
Subagents are automatically recognized and called by Claude Code—no explicit triggering needed:
# Auto-identify and delegate to market-researcher "Analyze the market size for AI writing assistants" # Auto-identify and delegate to competitive-analyst "Compare feature differences between ClickUp and Asana" # Auto-identify and delegate to pm-orchestrator "Do complete product planning for a new project management tool"
| Skill | Install | Function | Trigger Examples |
|---|---|---|---|
competitive-analysis |
--skill competitive-analysis |
Competitive feature comparison & strategic opportunity identification | "Analyze competitors", "Compare XX and YY" |
market-research |
--skill market-research |
Market trends, industry analysis & user research / personas | "Market analysis", "Industry trends", "User research" |
clarify-requirements |
--skill clarify-requirements |
Requirement clarification & information gap analysis | "Clarify requirements", "What info is missing" |
| Skill | Install | Function | Trigger Examples |
|---|---|---|---|
product-strategy |
--skill product-strategy |
Positioning, prioritization (RICE/MoSCoW) & roadmap in one pass | "Product positioning", "Feature priority", "Product roadmap" |
| Skill | Install | Function | Trigger Examples |
|---|---|---|---|
prd-gen |
--skill prd-gen |
Structured PRD generation (scenario-driven, with pre-review tail gate) | "Write PRD", "Requirements document" |
prototype-design |
--skill prototype-design |
HTML prototype generation (iteration + new product modes) | "Design prototype", "HTML prototype", "UI flow" |
| Skill | Install | Function | Trigger Examples |
|---|---|---|---|
retrospective |
--skill retrospective |
Post-launch impact analysis & next-iteration planning | "Impact analysis", "Post-launch review", "Plan next sprint" |
feedback-synthesis |
--skill feedback-synthesis |
User feedback aggregation & analysis | "Feedback analysis", "User feedback" |
| Workflow | Install | Function | Trigger Examples |
|---|---|---|---|
ompm |
--skill ompm |
Global intent router (front door): reads context + current-workflow.json, recommends the most valuable next step | "What should I do next", "Continue", any PM request |
quick-prd |
--skill quick-prd |
Thin preset of full-pm-cycle: competitive analysis + PRD | "Quick PRD", "Requirements with competitive analysis" |
full-pm-cycle |
--skill full-pm-cycle |
Complete product lifecycle (0-1) | "Complete product planning", "0-1 product" |
feature-launch |
--skill feature-launch |
Feature launch workflow (release checklist & rollout strategy) | "Feature launch", "Launch coordination" |
/prd-gen supports intelligent scenario recognition:
| Scenario | Required Info | UI Extraction Method |
|---|---|---|
| Iteration Update | Feature description + UI state + iteration goals | Screenshot / HTML / Link |
| New Feature | Product architecture + design spec + entry point | Read from docs/product/ |
| 0-1 New Product | Product context + resource constraints + reference products | User input + competitive analysis |
Core Principle: No random assumptions—output based on industry best practices.
oh-my-pm/
├── agents/ # Subagent definitions (6) - at root!
│ ├── perception/ # Market research, competitive analysis, user research
│ ├── validation/ # Impact analysis, feedback aggregation
│ └── workflows/ # PM orchestrator
├── skills/ # Skill plugin directory (12)
│ ├── competitive-analysis/ # layer: perception
│ ├── market-research/ # layer: perception
│ ├── clarify-requirements/ # layer: perception
│ ├── product-strategy/ # layer: strategy
│ ├── prd-gen/ # layer: design
│ ├── prototype-design/ # layer: design
│ ├── retrospective/ # layer: validation
│ ├── feedback-synthesis/ # layer: validation
│ ├── ompm/ # layer: workflow (global intent router)
│ ├── quick-prd/ # layer: workflow
│ ├── full-pm-cycle/ # layer: workflow
│ └── feature-launch/ # layer: workflow
├── .claude/
│ ├── agent-memory/ # Subagent memory system
│ └── settings.local.json
├── .claude-plugin/ # Plugin configuration
│ ├── plugin.json
│ ├── agents.yaml
│ ├── skills.yaml
│ └── marketplace.json
├── docs/product/ # Context passing directory
├── docs/ # Design documents
├── tests/ # Test scripts
└── CLAUDE.md # Project configuration
| Version | Goal | Status |
|---|---|---|
| v0.1.0 | MVP (4 Skills) | ✅ |
| v0.2.0 | Complete 5-layer architecture (19 Skills) | ✅ |
| v0.3.0 | Subagent hybrid architecture (8 Subagents + memory system) | ✅ |
| v0.4.0 | Commands integration (19 Skills + 4 Commands) | ✅ |
| v0.5.1 | Remove Figma dependency, switch to HTML prototypes (19 Skills) | ✅ |
| v0.6.0 | HTML prototype generation (iteration + new product modes) | ✅ |
| v0.9.0 | Flat skill directory, standalone install per skill | ✅ |
| v0.9.3 | Fix context-passing contract (flat .ompm model) + doc version sync | ✅ |
| v1.0.0 | Simplified architecture: 3-layer core + validation loop (11 Skills, 6 Subagents) | ✅ |
| v1.1.0 | Portability overhaul: new ompm intent-router skill (12 Skills total); npx skills add on any agent delivers full capability |
✅ |
See Project Board for full planning.