One command. From feature idea to tested code.
Forge scans your repo, detects your stack, and generates a full dev pipeline - architecture guardrails, quality gates, test traceability. Your AI stops guessing and starts following your rules.
◆ Gate 1: Architecture Audit
Stack: NestJS 11.0 + TypeScript 5.8 + Prisma 6.4
Architecture: Clean Architecture (3 layers)
Database: PostgreSQL 16 | CI: GitHub Actions
Auth: JWT + Passport | Events: Bull queues
Parallel: 7/10 | Conflicts: 2 zones
✓ Gate 2: Development Pipeline
Git: trunk-based · Commits: conventional (92%) · Mode: balanced
✓ Gate 3: Task Tracking → GitHub Issues + 3 scripts
✓ Gate 3.5: Script Verification → 3/3 syntax + smoke tests
◆ Gate 4: Quality Checklist
Capabilities: api · database · auth · events · docker
✓ Loaded: api.md + infra.md + shared.md → 53 patterns
◆ Gate 5: Naming Conventions
✓ Services: *Service (8/8) ✓ DTOs: Create*Dto (12/15)
⚠ Events: inconsistent → adopted *CreatedEvent
✓ Gate 6: Knowledge Export → CLAUDE.md updated, 15 rationalization detectors
◆ Gate 6.5: Skill Reconciliation
dev 35% → Replace (5/13 core features)
deploy 62% → Augment (+resume, +tasklist, +self-review)
testing 90% → Keep (12/13 core features)
◆ Generation: Dev-skills & Config
.claude/
├── CLAUDE.md ← architecture + conventions
├── settings.json ← 3 execution modes
└── skills/dev/
├── SKILL.md ← 6-phase TDD pipeline
├── step-catalog.md ← 52 scoped steps
└── quality-patterns/ ← api + infra + shared
✦ Setup complete. 9/9 gates passed. AI is now architecture-aware.
Works with: Claude Code · Cursor · Windsurf · any AI agent
AI writes code fast. You fix it faster.
“Your AI forgets your architecture every session. You spend 20 minutes re-explaining what NestJS guards do.”
“Your AI patches over bugs with quick fixes until the entire module is unmaintainable spaghetti.”
“I ran 3 agents in parallel. Spent 4 hours resolving merge conflicts. Net productivity: negative.”
“Unit tests pass with 100% coverage. Every assertion uses mock data. Zero real behavior tested.”
“A simple rename task burned 50K tokens because the AI re-read the entire codebase. Every. Single. Time.”
The real problem isn’t AI coding speed - it’s unstructured AI creating technical debt at scale.
Architecture-aware pipeline for AI agents
Understands your project
7-gate wizard auto-detects stack, layers, architecture patterns, and conventions. No manual config.
Enforces your standards
50+ quality patterns, naming conventions, 15 LLM rationalization detectors. Prevention, not detection.
Structures the workflow
Design → Build → Verify. Scoped phases with quality gates. Requirement traceability end to end.
One forge setup → your AI codes like a senior engineer who’s read every file in the repo.
What should Forge build for your project?
Choose your setup. Watch the pipeline assemble.
Product Health
No guardrails - AI generates code without following your conventions.
No test generation from specs - coverage depends entirely on manual effort.
AI and task tracker disconnected - status updates remain manual.
No regression testing for prompts - changes go unverified.
No parallel orchestration - features queue up sequentially.
No structured validation process - product decisions built on assumptions.
No SEO audit process - technical gaps and content strategy unaddressed.
No retention framework - churn causes and growth levers unexamined.
Turn on capabilities to see how your coverage changes.
Your AI forgets your architecture. Forge doesn’t.
7-gate wizard auto-detects your stack, layers, architecture patterns, and conventions. Generates project-specific dev-skills. No manual config. No prompt engineering.
◆ Gate 1: Architecture Audit - Results
Stack: Astro 5.7.0 + TypeScript 5.8.0 + Tailwind CSS 4.1.0
Build: Vite │ Linter: Biome 2.0.0 │ Hosting: Cloudflare Pages
Architecture: FSD-adapted (4 layers)
Layers:
pages: src/pages/ (15 files) - assembly & routing, i18n (en + ru)
widgets: src/widgets/ (32 files, 21 widgets) - composite blocks, zero cross-imports
shared: src/shared/ (41 files) - ui(13) + config(22) + i18n(3) + lib
layouts: src/layouts/ (2 files) - BaseLayout
Conflict zones: none (0 files)
Parallel readiness: 9/10 - highly parallel
Sibling repos:
| Repo | Stack | Role | Forge? |
|---|---|---|---|
| acme-web | React 19.2+TanStack | frontend | ✓ 4.6 |
| acme-api | NestJS 11+Prisma | backend-api | ✓ 4.12 |
$ forge autopilot "auth, payments, notifications"
✓ Triage: 3 features classified
✓ Conflicts: auth ↔ payments (shared User model)
✓ Wave 1: [auth] [notifications] - parallel
✓ Wave 2: [payments] - after auth merges
⚡ Deploying 2 teammates to worktrees...
Agent Teams without merge hell
Agent Teams are powerful - until 3 agents edit the same file. forge-autopilot orchestrates autonomous teammates like Kubernetes orchestrates containers: backlog triage, conflict detection, wave execution, auto-merge.
Tests that trace back to requirements. Not AI guesswork.
forge-product creates verifiable acceptance criteria BEFORE code. forge-qa generates tests FROM those criteria. Every test proves a requirement.
// checkout.spec.ts
describe('US-3: Order Confirmation', () => {
// AC: US-3.AC-1
it('creates order with valid items
and shipping address', async () => {
const order = await checkout(validCart);
expect(order.status).toBe('confirmed');
expect(order.items).toHaveLength(3);
});
// AC: US-3.AC-2
it('sends confirmation email within
5 minutes', async () => {
await checkout(validCart);
const email = await getLastEmail(user);
expect(email.subject).toContain('Order');
expect(email.sentWithin).toBeLessThan(300);
});
// UX: loading-2
it('shows spinner during payment
processing', async () => {
const { getByRole } = render(<Checkout />);
fireEvent.click(getByRole('button'));
expect(getByRole('progressbar')).toBeVisible();
});
});
One setup. Project-aware AI forever.
INSTALL
npm i -g @reumbra/forge
Global CLI install
ACTIVATE
forge activate <license>
License key from purchase
SETUP
forge setup
7-gate wizard analyzes project, generates dev-skills
BUILD
/forge:hub "add payments"
Describe what you want - Hub resolves prerequisites and runs the right pipeline
Generated artifacts work independently. Remove the plugin - your rules stay.
7 dev pipeline modules. 8 advisory. One ecosystem.
◇forge-core
Architecture audit & AI guardrails
7-gate
wizard
50+
quality patterns
15
rationalization detectors
3
template families
◇forge-product
Feature design in business language
- ✓ Product archetypes
- ✓ User flow generation
- ✓ User stories with acceptance criteria
- ✓ UX criteria library
◇forge-qa
Tests from requirements
- ✓ Multi-framework test generation
- ✓ Acceptance criteria to unit tests
- ✓ Use cases to E2E tests
- ✓ Requirement traceability matrix
◇forge-tracker
Your tracker talks to your AI agent
- ✓ GitHub integration
- ✓ ClickUp integration
- ✓ Linear integration
- ✓ Notion integration
- ✓ Jira integration
◇forge-autopilot
Agent Teams without merge hell
- ✓ Backlog → conflict detection → wave execution
- ✓ Autonomous teammates in isolated worktrees
- ✓ Auto-merge with conflict prevention
- ✓ 3 autonomy modes: supervised, semi-auto, full-auto
◇forge-worktree
Parallel branches, zero port conflicts
- ✓ 3 templates: Docker backend, Frontend, Generic
- ✓ Docker port isolation per worktree
- ✓ Auto-cleanup on branch merge
◇forge-prompts
Manage prompts like code
- ✓ Prompt inventory and audit
- ✓ CO-STAR, RISEN, TIDD-EC frameworks
- ✓ LLM-as-judge regression tests
- ✓ Learning loop evolution
◇forge-discovery
Validate before you build
- ✓ Ideation and competitive research
- ✓ Problem validation
- ✓ Interview synthesis
- ✓ Cognitive bias protection
◇forge-marketing
From problem to go-to-market
- ✓ Positioning strategy
- ✓ Pricing framework
- ✓ Launch planning
- ✓ Content strategy
◇forge-analytics
Plan measurement before you code
- ✓ Tracking plans
- ✓ Event schema generation
- ✓ Dashboard specs
- ✓ Naming conventions
◇forge-onboarding
From signup to aha moment
- ✓ Activation flow design
- ✓ Empty states strategy
- ✓ First-run experience
- ✓ Onboarding audit
◇forge-copy
Consistent voice across every surface
- ✓ UI microcopy generation
- ✓ Email and notification templates
- ✓ Error message design
- ✓ Brand voice enforcement
◇forge-seo
SEO grounded in your product, not tricks
- ✓ Technical SEO audit
- ✓ Intent-based keyword research
- ✓ Schema markup generation
- ✓ Content gap analysis
◇forge-growth
Retention loops before acquisition tactics
- ✓ Retention loops
- ✓ Referral mechanics
- ✓ Upgrade triggers
- ✓ Churn prevention
◇forge-ab
Test with rigor, not hunches
- ✓ Structured hypothesis design
- ✓ Sample size pre-commitment
- ✓ Statistical analysis
- ✓ Documented learnings
The dev pipeline handles code from spec to deploy. Advisory modules cover product strategy, discovery, marketing, and growth. Every module works standalone.
One command. Full ecosystem visibility.
After setup, /forge:setup becomes your dashboard - plugin versions, project profile, integrations, and available actions. Everything your AI needs to know, always up to date.
● Forge Ecosystem Dashboard - acme-web
| Component | Version | Status |
|---|---|---|
| forge-core | v4.13.0 | Configured |
| forge-product | v3.9.0 | Configured |
| forge-qa | v3.7.0 | Configured |
| forge-tracker | v3.3.0 | Configured |
| forge-autopilot | v1.4.0 | Active (6 batches) |
| forge-worktree | v1.2.0 | Active |
Project Profile
| Property | Value |
|---|---|
| Stack | React 19 + TanStack Start + Tailwind v4 + Vite + TypeScript strict |
| Architecture | FSD (5 layers, 13 features) |
| Pipeline | Parallel, trunk-based, conventional commits |
| Git workflow | Trunk-based with worktrees |
| CI/CD | Vercel (SSR Docker) |
| Backend | acme-api (bidirectional propagation) |
Ecosystem is fully configured and up to date. Available actions:
– /forge:hub - describe what you want, Hub routes to the right pipeline
– /forge:tracker - task management via Notion
– /forge:autopilot - batch development pipeline
– /forge:qa - testing and coverage
What developers say
"We had this recurring nightmare where Claude would 'helpfully' refactor our entire auth layer every time someone asked it to add a field to the user model. Forge's architecture audit basically draws a line and says 'this is load-bearing, don't touch it.' QA module catches the rest. Haven't had a rollback in 6 weeks."
Marcus T.
Senior Backend Engineer - Series B healthtech
"29 euros. 20 min setup. Now Claude actually knows my project has a monorepo with shared packages before it starts generating import paths to nowhere. That's it. That's the review."
Anja Kowalski
Freelance Full-Stack Developer
"Autopilot is the reason I bought Complete. We run 3 worktrees in parallel for sprint work and before Forge the merge conflicts were genuinely worse than just doing it sequentially. Now the agents actually coordinate. It's not magic - you still review everything - but I stopped mass-closing PRs on Monday mornings."
David Chen
Tech Lead - Seed-stage dev tools startup
"Showed the product design module to our PM and she asked if she could use it too. The QA traceability alone justified Pro for us - we went from 'Claude wrote tests that test nothing' to actual acceptance criteria coverage. Still annoying that I can't get per-seat billing but whatever, it works."
Priya S.
Engineering Manager - Mid-size fintech
"Honestly was skeptical about paying for a Claude plugin when Claude itself costs money. But I kept losing 15 minutes at the start of every session re-explaining my stack. Forge just... remembers. Architecture, conventions, the whole thing. Starter is enough for my use case."
Jake Morrison
Indie Dev / OSS Maintainer
"The prompt management module is lowkey the most underrated part. I had like 9 different system prompts copy-pasted across repos with subtle differences. Now there's one source of truth and it actually evolves based on what works. Also the tracker integration with Linear saves me from context-switching to update tickets manually."
Tomas Herrera
Senior Frontend Engineer - Series A edtech
"I was mass-evaluating AI dev tools for the team and most of them are either vaporware or just a prettier wrapper around 'paste your codebase into context.' Forge is the only one that actually enforces architecture decisions downstream. The advisory modules are hit or miss depending on your stack but the core pipeline is solid. We standardized on Pro for ICs and Complete for leads."
Lena R.
CTO - B2B SaaS, 15 engineers
"lol it literally detected my NestJS project structure and stopped Claude from generating Express-style route handlers. worth it for that alone tbh"
Sam K.
Backend Developer - Contract, NestJS
Transparent pricing. No token billing.
STARTER
For solo developers
Includes 1 year of updates
Stop re-explaining your stack. One setup, forever.
- ✓ 7-gate audit detects your stack, layers, conventions
- ✓ 50+ quality patterns for AI-generated code
- ✓ Project-aware dev-skills - one setup, forever
- ✓ Scoped phases prevent runaway sessions
✓ 14-day money-back guarantee
PRO
For professional developers
Includes 1 year of updates
Full pipeline: Design → Code → Test → Track
- ✓ Everything in Starter
- ✓ Describe features in business language - AI builds from spec
- ✓ Tests traced to requirements, not AI guesswork
- ✓ Prompt engineering frameworks - consistent AI output across sessions
- ✓ 6 modules: core, product, qa, tracker, prompts, worktree
✓ 14-day money-back guarantee
COMPLETE
For developer-founders
Includes 1 year of updates
Agent Teams. Zero conflicts. Full automation from discovery to growth.
- ✓ Everything in Pro
- ✓ forge-autopilot - Agent Teams on your backlog, zero merge conflicts
- ✓ 8 product advisory modules - from discovery to growth
- ✓ Every future module included
✓ 14-day money-back guarantee
One-time purchase. Your artifacts stay forever.
Includes 1 year of updates. Renewal optional at reduced price.
No per-token charges. No usage limits. No surprise bills.
Devin: €240/yr mandatory. Cursor Pro: €240/yr. Forge: from €29 once.
By a developer, for developers
Martin Sikora
Senior Engineer · Founder, Reumbra
"I built Forge because I was tired of re-explaining my architecture to AI agents. Every project, every prompt, the same context from scratch. So I encoded 10 years of engineering decisions into a system that does it automatically."
10+ years
Experience
Senior Engineer
Role
Founder, Reumbra
Company
What is Forge DevKit?
Forge DevKit is a commercial plugin ecosystem for AI coding agents (Claude Code, Cursor, Windsurf). One command scans your repository - stack, layers, patterns - and generates architecture-aware guardrails, quality patterns, and dev-skills your AI follows in every session. 15 modules cover the full pipeline from architecture audit to test generation to Agent Teams orchestration. One-time purchase from EUR 29. No subscription. No per-token charges.
Frequently asked questions
How is this different from writing .cursorrules myself?
Auto-generated from architecture audit. Full pipeline, not just rules. Platform-agnostic - works in Cursor AND Claude Code AND Cowork.
Does it work with my stack?
8+ frameworks supported. 3 template families (NestJS/Clean Arch, generic backend, frontend). Stack-specific quality patterns auto-detected.
What if I remove Forge?
Disposable meta-tool: generated artifacts (dev-skills, tests, docs) work independently. Your code doesn't depend on Forge.
Does it work outside Claude Code?
.claude/ directory auto-detected by Cursor, Claude Desktop, Claude Cowork. Forge installs there - all platforms pick it up.
Is there a money-back guarantee?
Yes. 14-day money-back guarantee, no questions asked. If Forge doesn't improve your workflow, email us and we'll refund immediately.
Why is there no free trial?
Forge is a one-time purchase, not a subscription. A free trial doesn't make sense for a tool you buy once. Instead, you get a 14-day money-back guarantee - buy it, try it on your project, and if it doesn't work for you, get a full refund. Zero risk.
What's the ROI of using Forge?
One prevented production bug pays for the entire year. At €29-79, Forge costs less than a single hour of debugging AI-generated regressions.
How does Forge compare to Devin?
Devin is an autonomous AI developer - it writes code for you. Forge is infrastructure for YOUR AI agent - it makes your existing tools (Claude Code, Cursor, Windsurf) architecture-aware. Devin costs €240/yr mandatory. Forge starts at €29 one-time.
Does Forge work with Claude Agent Teams?
Yes. forge-autopilot is specifically designed to orchestrate Agent Teams - it handles backlog triage, conflict detection, wave planning, and auto-merge. Think of it as Kubernetes for AI agents.
What AI coding tools does Forge support?
Any tool that reads CLAUDE.md, .cursorrules, or similar config files. This includes Claude Code, Cursor, Windsurf, Cline, and any future AI agent. Forge generates platform-agnostic artifacts.
What happens after the 1-year update period?
Your generated artifacts continue working forever - they're files in your project, not a service. You just won't get new quality patterns or module updates. Optionally renew at a reduced price to stay current.
How does Forge reduce AI token costs?
Three ways. First: execution modes - economy mode skips optional quality steps, cutting ~60% of tokens on routine tasks. Second: pipeline phases communicate through 500-token summaries instead of passing full context (~49K) between phases. Third: subagents receive only their scope, not the entire codebase. Simple bug fix? 5-step economy mode. Complex feature? Full quality pipeline. You choose per task.
Your AI should respect your architecture.
50+ guardrails. 7-gate audit. Tests from requirements. One setup.
One-time purchase. 14-day money-back guarantee.