Lightwork Studio

Enforced orchestration runtime for AI-assisted development.

32 rules. 20 hooks. 7 agents. Quality enforcement by code, not hopes.

Works with: Claude Code | VS Code | Cursor | Gemini CLI | Codex | GitHub Copilot

What Is Lightwork?

AI coding tools generate code 10x faster. But 66% of developers say the output is "almost right, but not quite." 45% contains known security flaws. 37.6% more vulnerabilities with each iteration cycle.

Speed without quality is just faster failure.

Lightwork Studio transforms Claude Code into a rigorously enforced production system. Install it, scaffold the framework, and every AI interaction runs through 32 battle-tested rules enforced by 20 hooks.

Who Is Lightwork For?

Non-technical founders: You're the CEO with an idea. Lightwork is your CTO — it handles engineering discipline, quality enforcement, and technical decision-making so you can focus on what to build, not how to build it.

Solo developers: You know the discipline is important but context-switching between coding and quality checking kills your flow. Lightwork enforces the discipline automatically so you stay in the zone.

Senior engineers who use AI tools: You've seen the quality problems firsthand. Lightwork gives you the enforcement hooks to keep AI-generated code at your standards without manually reviewing every line.

Features

Free Tier

• Build Lightwork — One command scaffolds the entire framework into any project
• Pipeline Visualizer — See current phase, task, session count in the sidebar
• Status Bar — Live indicator showing Lightwork state
• 32 Rules — Browse all rules with categories in the sidebar
• 3 Agents — Researcher, Developer, Reviewer
• State Management — Automatic state file creation and tracking

Pro ($29/month — save with quarterly, 6-month, or yearly plans)

• 20 Enforcement Hooks — Rules enforced by code. Can't stop without tests. Can't skip steps.
• All 7 Agents — Adds Planner, Tester, Security, Architect with strict tool separation
• Constraint Snapshot — Evaluate decisions against all constraint fields before implementing
• Quality Dashboard — Live visualization of knowledge connections, weights, and field health
• Intent Capture — Structured intent extraction saved to CONTEXT.md
• Knowledge System — UUID-indexed entries with weight-based retrieval
• Task Triage — PATCH/SMALL/MEDIUM/LARGE classification scales the cycle to complexity
• Haiku Semantic Evaluation — Each /proceed cycle fires an independent evaluation of all 32 rules. Uses your Claude API quota (~1 short message per cycle). Degrades gracefully if unavailable.
• PR Quality Gate — GitHub Actions workflow that enforces Lightwork standards on every PR

Enterprise ($199/user/month — save with quarterly, 6-month, or yearly plans)

• Everything in Pro
• Cognition Engine — Dynamic knowledge graph with Hebbian learning
• Team Dashboard — Aggregated enforcement metrics across developers (trends, not surveillance)
• Branch-Scoped Enforcement — Cycle history audit trail across branches
• Cross-Repo Knowledge Sharing — Export/import knowledge entries between projects
• Role-Based CONTEXT.md — Product Owner, Tech Lead, Designer sections with ownership
• Team Onboarding — Complete TEAM.md with shared vs per-developer file management
• Dedicated Support

The 32 Rules

Every rule earned through failure in production development:

1. ONE CYCLE PER PROCEED
2. ZERO MANUAL STEPS
3. MAXIMUM EFFORT
4. OVERKILL GATE
5. R/T/P/B PIPELINE
6. SESSION BOOTSTRAP
7. STATE FILES CURRENT
8. ONE FEATURE PER CYCLE
9. CONTEXT.MD SACRED
10. RESEARCH IS A LIBRARY
11. GIT DISCIPLINE
12. REAL SCIENCE
13. SELF-IMPROVEMENT
14. VERIFY BEFORE BUILD
15. DISAGREE WHEN EVIDENCE DEMANDS
16. SELF-CRITIQUE
17. SPEC COMPLETENESS GATE
18. UNIVERSAL SECURITY
19. TRACE THE FULL PATH
20. TEST WITH THE HUMAN
21. DOMAIN RESEARCH FIRST
22. PLAYTEST FEEDBACK OVERRIDES
23. DEPENDENCY COMPLETENESS
24. ITERATION-AWARE SECURITY
25. LOOP DETECTION
26. COMPLETION VERIFICATION
27. INTENT VERSIONING
28. RATIONALIZATION DETECTION
29. DEPLOYMENT IS BUILD
30. REGRESSION PREVENTION
31. VALIDATE BEFORE BUILD
32. DOCUMENTATION INTEGRITY

Why Lightwork Exists

Without orchestration frameworks, measured AI code quality:

• 1.7x more issues per PR (CodeRabbit 2025-26)
• 37.6% vulnerability increase per iteration cycle (IEEE-ISTAS 2025)
• 45% of AI code has known security flaws (Veracode 2025)
• 31.7% fail in clean environments (Reproducibility Study 2026)
• +41% complexity increase with AI agents (GitClear/Cortex 2025)

Every Lightwork rule, hook, and agent exists to address a specific measured failure mode. Every rule earned through failure, not theory.

Prerequisites

• VS Code 1.85 or later (or Cursor)
• Claude Code — Either the Claude Code CLI (terminal-based) or the Claude Code Extension (VS Code native). Both work with Lightwork.

Quick Start

1. Install from VS Code Marketplace (or Open VSX for Cursor)
2. Open any project
3. Run command: Lightwork: Build Lightwork
4. Open Claude Code and type /proceed for your first enforced cycle

Research Basis

Grounded in 20+ peer-reviewed sources: DORA/Accelerate, Reflexion (Shinn et al. 2023), Swiss Cheese Model, Praetorian 8-layer architecture, CodeRabbit AI Code Report, IEEE-ISTAS Security, Veracode GenAI Security, and more.

Privacy

Lightwork Studio collects zero data by default. Optional telemetry (opt-in only) collects anonymous usage patterns — never code, file names, specs, or project content. Your CONTEXT.md never leaves your machine.

License

Extension: MIT | Framework methodology: CC BY-NC-SA 4.0

Built in Arkansas. By a human and an AI. Who decided that "good enough" isn't.