A Claude Code plugin that transforms Product Requirement Documents (PRDs) into production-ready code through AI-powered spec generation, automated task execution, and multi-stage quality review.
- PRD to Specs: Automatically generate technical specifications from requirement documents
- AI-Driven Development: Execute development tasks with specialized AI agents
- Multi-Stage Review: Spec compliance check + code quality review (Codex integration)
- Change Management: Support requirement changes with version tracking
- Lessons Learned: Automatic documentation of architectural decisions and gotchas
- Single Repository Mode: Unified docs and code structure
- Claude Code CLI or Desktop App
- Git repository for your project
# Clone this repository
git clone https://github.com/jaredchao/forge.git ~/.claude/plugins/forge
# Or install via Claude Code (when published)
claude plugin install forgecd your-project
/forge:initThis creates:
your-project/
├── docs/
│ ├── prd/ # Place your PRD documents here
│ └── specs/ # Generated specs (auto-created)
└── .claude/
├── CLAUDE.md # Project context
└── rules/ # Coding standards
Place your requirement document in docs/prd/, then:
/forge:prd ~/path/to/your-projectForge will:
- Read PRD documents from
docs/prd/ - Analyze project architecture and tech stack
- Generate
requirements.md,design.md, andtasks.mdindocs/specs/{N}.{feature-name}/
/forge:ai ~/path/to/your-projectForge will:
- Execute tasks sequentially with specialized agents
- Perform spec compliance review after each task
- Run Codex code quality review
- Mark completed tasks and record lessons learned
- Continue until all features are implemented
PRD Document
↓
[/forge:prd] → Generate Specs
↓
requirements.md + design.md + tasks.md
↓
[/forge:ai] → Automated Development
↓
For each task:
1. Match specialized skill (frontend/backend/QA)
2. Dispatch implementer subagent
3. Spec compliance review
4. Codex quality review
5. Mark done + record lessons
↓
Production-Ready Code
Generate development specifications from PRD documents.
New Feature:
/forge:prd ~/code/my-appChange Request:
/forge:prd --change 1.user-auth "Add WeChat login support"
/forge:prd --change 2 "Update validation logic"Execute automated development workflow.
/forge:ai ~/code/my-appInitialize Forge project structure (creates docs/ and .claude/ directories).
cd your-project
/forge:initForge includes specialized agents for different development tasks:
- forge-implementer: General-purpose implementation
- forge-frontend-engineer: React/Vue/frontend development
- forge-backend-engineer: Backend/API development
- forge-database-engineer: Data modeling, migrations, query optimization
- forge-contract-engineer: Smart contract development (EVM/Solana/Move)
- forge-qa-engineer: Testing and quality assurance
- forge-doc-syncer: Documentation synchronization
Auxiliary skills (invoked on demand, not dispatched per task):
- forge-spec-writer: Requirement docs → requirements/design/tasks specs
- forge-code-reviewer: Two-round code review + security scan (Telegram alert on findings)
- forge-test-runner: Auto-generate and run unit/E2E tests + visual regression
- forge-security: METATRON-based penetration testing of deployment targets
Each engineering discipline (frontend/backend/database/contract/qa) is also exposed as a subagent (agents/forge-*-engineer) with its own tool sandbox and model: sonnet. N2 dispatches them per-discipline for parallel work with natural context isolation; each subagent loads its same-named skill as the single source of truth.
your-project/
├── docs/
│ ├── prd/ # Input: requirement documents
│ │ └── feature-spec.md
│ └── specs/ # Output: generated specs
│ ├── 1.user-auth/
│ │ ├── requirements.md # Functional requirements
│ │ ├── design.md # Technical design
│ │ └── tasks.md # Task breakdown
│ ├── 2.payment/
│ └── LESSONS.md # Lessons learned
├── .claude/
│ ├── CLAUDE.md # Project context
│ └── rules/ # Coding standards
└── src/ # Your code
Every task goes through two mandatory review stages:
- Spec Compliance Review: Verifies implementation matches requirements exactly
- Codex Quality Review: Checks code quality, security (OWASP Top 10), and best practices
Failed reviews trigger automatic fixes before proceeding.
Define your tech stack, architecture, and conventions:
# Project: My App
## Tech Stack
- Frontend: React + TypeScript
- Backend: Node.js + Express
- Database: PostgreSQL
## Architecture
- Monorepo with pnpm workspaces
- REST API with OpenAPI specsAdd project-specific rules:
security.md- Security requirementstesting.md- Test coverage standardsstyle.md- Code style guidelines
See examples/ directory for sample projects:
examples/simple-webapp/- Basic web applicationexamples/api-service/- REST API serviceexamples/monorepo/- Multi-package monorepo
When requirements change:
/forge:prd --change 1.user-auth "Add OAuth2 support"Forge will:
- Update
requirements.mdwith version tracking - Modify affected sections in
design.md - Mark changed/dropped/new tasks in
tasks.md - Preserve completed tasks
Tasks without dependencies are executed in parallel automatically.
- Subagents provide natural context isolation between tasks
- Controller executes
/compactat 80% context usage /clearbetween features to reset context
Place your PRD documents in {PROJECT_DIR}/docs/prd/ before running /forge:prd.
Ensure your project has package.json, Cargo.toml, go.mod, or pyproject.toml.
Check the review output for specific issues. Forge will automatically attempt fixes.
Contributions welcome! Please read CONTRIBUTING.md for guidelines.
MIT License - see LICENSE for details.
- Issues: GitHub Issues
- Discussions: GitHub Discussions
- Support for more specialized skills (mobile, DevOps, ML)
- Integration with project management tools (Jira, Linear)
- Visual progress dashboard
- Multi-language PRD support
- Automated E2E test generation
Built with Claude Code | Powered by Claude 4.6