Interactive Guide

Your Stack, Understood

forge setup reads your codebase - package.json, directory structure, config files - and builds a complete architecture profile. No questionnaires, no configuration files. It just works.

Backend Project

9 gates in under 10 seconds. Forge detects your full stack (NestJS, Prisma, PostgreSQL, JWT, Bull queues), audits naming conventions across all services, measures existing skill coverage, and generates a tailored dev-pipeline with 50+ quality patterns scoped to your exact capabilities.

Frontend Project

Same command, completely different output. On a Next.js + React 19 + Tailwind v4 project, Forge generates UI-specific quality patterns (accessibility, performance, component boundaries), configures component testing with Testing Library + Vitest, and adapts the entire pipeline to frontend conventions.

Ecosystem Discovery

Gate 0 scans your installed adapters and recommends additional modules based on your stack and workflow. Install them in seconds - they auto-register into the dev pipeline.

Tell It What You Want

Without Hub, you need to know that /forge:product should run before /forge:qa, which should run before /dev. With Hub, you describe what you want in plain English. /forge:hub checks what prerequisites exist, finds what's missing, and builds the right execution chain automatically.

Automatic Prerequisite Resolution

Ask for a payment system. Hub checks if product stories exist for payments - they don't. Checks if a test plan exists - it doesn't. Instead of jumping straight to code with missing context, Hub builds a 3-step chain: design the feature first, generate tests from acceptance criteria, then implement with TDD. Every prerequisite resolved before a single line of code is written.

The Dev Pipeline

When Hub routes to development, /dev takes over with a 6-phase TDD pipeline. All your modules work together: forge-worktree isolates in a branch, forge-product loads acceptance criteria, forge-qa generates tests, forge-tracker syncs status - full integration, zero manual coordination.

6-Phase Pipeline with Module Integration

Give it a feature description. Forge creates an isolated worktree, loads acceptance criteria from forge-product, generates failing tests via forge-qa (dispatched on Sonnet for cost efficiency), implements until green, runs self-review on Haiku, verifies AC coverage, and opens a PR - all in one command. Subagents run on cheaper models so you don't burn Opus tokens on mechanical work.

Rationalization Detectors

AI models are trained to be helpful - sometimes too helpful. When an AI agent rationalizes skipping best practices ("let's add tests later", "error handling can wait"), Forge's 15 rationalization detectors catch it mid-generation and block the shortcut. Patterns are scoped to your capabilities: API guards, database migrations, auth flows, event handling - not generic checklists.

Specialized Modules

Beyond setup and the dev pipeline, Forge has specialized modules for testing, task tracking, product design, prompt engineering, and fully autonomous development.

Test Generation (forge-qa)

Tests generated from acceptance criteria, not AI guesswork. forge-qa reads your feature specs from forge-product, maps each AC to concrete assertions, and generates tests across 11 modes (unit, integration, component, acceptance, e2e). Every test traces back to a requirement through 4-level traceability: AC, use cases, UX criteria, and LLM-as-Judge evaluation.

Task Sync (forge-tracker)

Bidirectional sync between your task tracker and the AI agent. Pull tasks from Linear, ClickUp, Jira, GitHub, or Notion - acceptance criteria flow directly into the dev pipeline. When implementation is done, forge-tracker marks the task complete and posts a summary. No copy-pasting ticket IDs, no manual status updates.

Agent Teams (forge-autopilot)

The endgame. Forge triages your backlog, detects file conflicts between features, groups them into parallel waves, and spawns AI agent teams in isolated git worktrees. A tech-lead agent coordinates while teammates execute the full /dev pipeline independently. 3 autonomy modes: supervised (you approve each PR), semi-auto (auto-merge green PRs), full-auto (hands-off sprint execution).

And More

Start Building Better

Everything you saw runs in under 10 seconds. The artifacts live in .claude/ - they work with Claude Code, Cursor, Windsurf, and any AI agent. Remove Forge after setup if you want - the artifacts are standalone.