super-dev

Orchestrates a linear 8-stage workflow (Documentation → Spec → Red Team Review → Quality Gate → Code Review Guide → AI Prompt → CI/CD → Migration) using a WorkflowEngine that enforces a mandatory 80+ quality score threshold at Stage 4 before proceeding to implementation stages. Each stage generates artifacts that feed into the next, creating an auditable chain of custody from requirements to production-ready code. The pipeline uses scenario detection and domain-aware context to adapt generation strategies based on project type and tech stack.

The DocumentGenerator class produces three categories of human-readable artifacts (PRD, Architecture, UI/UX) by leveraging domain knowledge (6 business domains × 4 tech platforms × common patterns) and project analysis results. Generation is context-aware: it detects project type (e.g., SaaS, mobile app, API service) and tech stack (e.g., React + Node.js + PostgreSQL) and adapts templates and content accordingly. Uses Claude to synthesize requirements into structured documents with sections for acceptance criteria, non-functional requirements, and architectural constraints.

Stage 5 of the pipeline that generates detailed code review guidelines and checklists specific to the project's architecture, tech stack, and quality standards. The guide includes acceptance criteria from specs, architectural compliance checks (e.g., microservices isolation, API contract validation), performance benchmarks, security requirements, and testing expectations. Formatted as a structured document that human reviewers or AI tools can follow during code review, with specific checks tied to the generated specifications and architecture documentation.

Super Dev operates in two distinct modes that share core engines: (1) CLI tool for standalone artifact generation (specs, docs, prompts, CI/CD, migrations), and (2) Agent Skills for integration with Claude Code and other AI IDEs via OpenClaw/MCP protocols. The dual architecture enables both batch processing workflows (CLI) and interactive development workflows (agent skills). Both modes use the same underlying components (DocumentGenerator, ProjectAnalyzer, QualityGateChecker, etc.) but expose different interfaces and integration points.

A WorkflowContext system that maintains state across the 8-stage pipeline, tracking artifacts, quality scores, approvals, and decisions at each stage. Implements an enforcement layer that ensures mandatory quality gates are met before stage progression and prevents skipping stages. Uses a memory system to persist workflow state (local or cloud-based) and enable resumption of interrupted workflows. Provides audit trails of all decisions, approvals, and quality checks for compliance and traceability.

Implements a spec-first development model where specifications are generated before code, and changes are tracked as delta specifications rather than code diffs. The SDD workflow manages a directory structure that separates specs, designs, and code artifacts, and tracks the lifecycle of each change (proposed → reviewed → approved → implemented). Uses OpenSpec format (machine-readable specification standard) to enable AI tools to consume specs directly. Supports incremental updates via delta specifications that describe only what changed, reducing context bloat for iterative development.

A design asset repository system that indexes design patterns, components, and tokens using BM25+ full-text search, enabling semantic retrieval of relevant design assets for new features. The engine generates design systems and design tokens (color palettes, typography, spacing scales) based on project context and tech stack. Uses a Design Asset Repository to store and retrieve design patterns, and a Design System Generator to synthesize tokens and component specifications from project analysis and domain knowledge.

An expert system that models domain expertise through expert personas (e.g., Backend Architect, Frontend Engineer, QA Lead) with associated knowledge bases and skills. Each persona has specialized knowledge for their domain and can be invoked as an agent skill in Claude Code or other AI IDEs. The system integrates with agent skill frameworks (OpenClaw, MCP) to expose expert personas as callable functions that AI tools can invoke during development. Uses a knowledge base per persona to provide context-specific guidance and best practices.

A QualityGateChecker system that evaluates generated artifacts against 80+ quality dimensions (code quality, architectural compliance, security, performance, testability, etc.) and produces a composite quality score. Uses scenario detection to identify project-specific quality concerns (e.g., high-concurrency systems require performance checks, healthcare projects require compliance checks) and adapts quality checks accordingly. Enforces a mandatory 80+ score threshold at Stage 4 before proceeding to implementation stages, with detailed feedback on failing checks to guide remediation.

A ProjectAnalyzer component that inspects existing codebases to automatically detect tech stack (languages, frameworks, databases, deployment platforms) and classify project type (SaaS, mobile app, API service, monolith, microservices, etc.). Uses pattern matching on file structures, dependency files (package.json, requirements.txt, pom.xml, etc.), and code samples to infer tech stack. Feeds analysis results to downstream components (DocumentGenerator, QualityGateChecker, DesignEngine) to enable context-aware generation and quality checks.

Generates AI-ready prompts (Stage 6 of pipeline) formatted specifically for 15+ AI coding platforms (Claude Code, Cursor, Windsurf, Aider, ChatGPT, GitHub Copilot, etc.) from specifications and documentation. Each platform has a distinct prompt format and capability set; the generator adapts prompt structure, context injection, and instruction style to match each platform's strengths. Produces prompts that include specs, architecture docs, code review guidelines, and expert persona knowledge, enabling AI tools to implement features with full context and quality standards.

Generates CI/CD pipeline configurations (GitHub Actions, GitLab CI, etc.) and database migration scripts based on project analysis and detected ORM type (SQLAlchemy, Django ORM, Sequelize, etc.). The CI/CD Generator produces workflow files that implement quality gates, automated testing, and deployment stages aligned with the 8-stage pipeline. The Migration Generator detects ORM patterns and generates migration scripts for schema changes, with support for common ORMs and databases. Integrates with Stage 7 (CI/CD Configuration) and Stage 8 (Database Migration) of the pipeline.

Stage 3 of the pipeline that performs adversarial review of generated specifications and designs to identify potential risks, security vulnerabilities, performance bottlenecks, and architectural issues before implementation. Uses expert personas (Security Architect, Performance Engineer, etc.) to evaluate specs from multiple perspectives and generate detailed risk reports with mitigation recommendations. Feeds risk findings into Stage 4 (Quality Gate) to inform quality scoring and may trigger spec revisions before proceeding to implementation stages.