Public preview: free to install. The plugin is open source (MIT) and the orchestrator is source-available (FSL-1.1). Alpha testing is in progress, and interfaces may change.
Governance-native AI development.
For Claude Code; Codex CLI as a task runner; Gemini CLI support in development.
The only system that makes AI agents safer as they get more autonomous. Skills are commoditized. Every plugin gives you agents and commands. Only Forge gives you governance: drift detection that catches agents going off-spec (it needs a configured model), CRUCIBLE auditing that exposes hollow tests, blocking security guards that prevent dangerous operations, and shared memory that crosses sessions. 9 of 12 capabilities exist nowhere else.
claude plugin marketplace add nxtg-ai/forge-plugin && claude plugin install nxtg-forgeThe Last Mile Problem
Why does AI-generated code compile but fail to ship?
The illusion of correctness: code that looks polished but conceals serious flaws. The independent research below shows the gap is measurable, not anecdotal.
45% chose the insecure path
When faced with a secure and an insecure way to write the same code, AI models picked the insecure one in 45% of tests. Not edge cases — nearly half.
86% failed XSS defense
AI-generated code failed to defend against cross-site scripting (CWE-80) in 86% of relevant samples. The AI can explain XSS perfectly. It just doesn't enforce the defense.
The Kafka handler story
An AI-generated event handler passed unit tests and integration tests. Under production traffic, it failed. The tests never simulated the load patterns production actually generated.
Knowledge amnesia
Claude refactors your auth module. Codex updates tests against the pre-refactor interface. Both save. Tests fail. Neither tool knows the other exists.
19% slower, convinced they were faster
In a controlled trial, experienced developers took 19% longer to finish tasks with AI tools — while estimating AI had sped them up by 20%. Code compiles, tests pass, and the polished surface conceals the gap.
Source: METR 2025 RCT
One Product
Three Depths
Start where it hurts. Go deeper when you're ready.
Forge Plugin
Zero-dependency governance for Claude Code (today, health scoring needs Node.js 20+; build B3, target 2026-10-20).
- -4 blocking security guards: prevent rm -rf, .env access, eval() injection, SQL injection
- -33 specialized agents: builder, guardian, planner, detective, security, testing, and 27 more
- -23 slash commands: /nxtg-forge:feature, /nxtg-forge:status, /nxtg-forge:gap-analysis, /nxtg-forge:test, and 19 more
- -27 knowledge skills that auto-load based on context: architecture, security, testing, coding standards
- -12 governance hooks: 4 blocking PreToolUse security guards + 8 advisory governance checks
30 seconds to install. Nothing to configure.
+ Forge Orchestrator
Coordinates Claude Code and Codex CLI on one project.
- -File locking across AI tools: prevents simultaneous edit conflicts at the Rust level
- -Knowledge flywheel: captures decisions, patterns, and learnings from every session
- -Plan generation from specs: forge plan --generate turns a spec into a task board
- -Multi-AI support through a single state file: Claude Code, and Codex CLI as a task runner. Gemini CLI support is in development (build B1, target 2026-10-27).
- -TUI dashboard with live agent panes, health scores, drift detection (needs a configured model)
- -Headless autonomous mode: forge run for CI/CD pipelines
One binary. 6.4 MB. 572 Rust tests. Zero runtime dependencies.
+ Forge Dashboard
The Portfolio Architect's lens.
- -Real-time governance dashboard: health score A/B/C/D/F with 8 quality checks
- -Agent activity feed: watch 33 agents coordinate in real-time
- -Infinity Terminal: survives browser close, network drops, server restarts (today, it survives tab close and network drops for 5 minutes; build B5, target 2026-10-20)
- -Multi-device access: phone, tablet, remote machine. Sessions never die (today, a session lasts 5 minutes after you disconnect; build B5, target 2026-10-20).
71 React components. 4,500+ tests.
The Builder
Built by a Portfolio Architect
Asif Waliuddin has spent 23 years as a Portfolio Architect: building automation frameworks, governance systems, and team cultures for enterprise-scale programs. The coordination failures he watched happen with human teams (knowledge lost in handoffs, conflicting decisions, governance enforced after the fact) are exactly the failures that appear when you run multiple AI tools without coordination infrastructure.
The lineage is direct. Spreadsheets in the dot-com era. Access databases for rule encoding. Power Automate for workflow automation. Each era, the same goal: encode coordination into systems so teams can focus on building.
Every feature in Forge maps to a specific failure mode. File locking exists because teams lose days to conflicting edits. The knowledge flywheel exists because decisions evaporate between sprints. Governance hooks exist because quality checks happen at audit time instead of during the work.
Forge is the AI-native chapter of a 23-year automation lineage. Not a framework built by an engineer who read about coordination. Coordination infrastructure built by someone who has lived it.
Questions
Frequently asked questions
- What is NXTG-Forge?
- Forge is the only developer governance system that makes AI agent teams safer as they get more autonomous. It provides drift detection (needs a configured model), CRUCIBLE test quality auditing, blocking security guards, shared memory, and multi-tool orchestration for Claude Code, with Codex CLI as a task runner. Gemini CLI support is in development (build B1, target 2026-10-27).
- How do I install Forge?
- L1 (Vibe Coder): Run "claude plugin marketplace add nxtg-ai/forge-plugin && claude plugin install nxtg-forge" for governance in Claude Code. L2 (Pro Builder): Run "curl -fsSL https://forge.nxtg.ai/install.sh | sh" for multi-tool coordination. L3 (Ship Lord): Clone the forge-ui repo for visual dashboards.
- Why does AI-generated code compile but still fail in production?
- AI gets you to "it compiles" fast. The last 20% (tests, security, deployment, governance) is where it falls apart. Veracode's 2025 GenAI Code Security Report found that AI models chose the insecure coding path in 45% of tests, and failed to defend against cross-site scripting in 86% of relevant samples. Forge closes this gap with intrinsic governance built into every session.
- Does using AI coding tools actually make developers faster?
- Not automatically. A 2025 METR randomized controlled trial found experienced open-source developers took 19% longer to complete tasks when allowed to use AI tools, even though they believed AI had sped them up by 20%. The perceived-versus-actual gap is exactly why Forge measures test quality and drift (drift detection needs a configured model) instead of trusting the feeling of speed.
- Is Forge open source?
- The plugin is MIT licensed. The orchestrator and UI are FSL-1.1-ALv2: source-available, and each release converts to Apache-2.0 two years after it ships. State lives in plain files in .forge/, fully portable.
Get Started
Public preview: free to install. The plugin is open source (MIT) and the orchestrator is source-available (FSL-1.1). Alpha testing is in progress, and interfaces may change.