Fusengine Claude Code Plugins
A plugin ecosystem that turns Claude Code into a supervised, multi-agent development environment. Expert agents write code, hooks enforce quality in real-time, skills inject framework-specific knowledge, and intelligent cartography auto-maps plugins and projects — so Claude never guesses, never duplicates, and always follows your architecture.
Scripted demo — a faithful reconstruction of the real hook → agent → sniper cycle. Regenerate with vhs docs/demo/hook-sniper.tape.
What It Does
Without plugins: Claude Code writes code based on its training data. It can hallucinate APIs, duplicate existing code, ignore your project structure, and produce inconsistent quality.
With Fusengine plugins:
- Expert agents detect your project type (Next.js, Laravel, React, Astro, Swift...) and load framework-specific documentation via MCP servers before writing a single line
- 60+ hook checks intercept Write/Edit/Bash and agent/session lifecycle events to enforce file-size limits (<100 lines), block code duplication (DRY), require SOLID reference reads, gate documentation lookup, and block destructive shell commands
- 196 skills inject copy-paste-ready templates, architecture patterns, and best practices directly into agent context — no hallucination needed
- APEX workflow structures every task through Analyze → Plan → Execute (TDD) → Review → Validate — preventing the "just write code and hope" approach
- Sniper agent — a 7-phase post-edit quality check (explore → research → grep usages → lint → fix → zero errors) the APEX workflow invokes after every modification
Quick Install
# Add marketplace
/plugin marketplace add fusengine/agents
# Install all plugins
/plugin install fuse-ai-pilot fuse-commit-pro fuse-laravel fuse-nextjs fuse-react fuse-astro fuse-swift-apple-expert fuse-solid fuse-tailwindcss fuse-design fuse-prompt-engineer fuse-shadcn-ui fuse-security fuse-changelog fuse-lessons fuse-tanstack-start fuse-typescript fuse-php fuse-rust fuse-go
# Setup (hooks + API keys + MCP servers)
~/.claude/plugins/marketplaces/fusengine-plugins/setup.sh # macOS / Linux
~\.claude\plugins\marketplaces\fusengine-plugins\setup.ps1 # Windows
Statusline (optional):
bun --cwd ~/.claude/plugins/marketplaces/fusengine-plugins/plugins/core-guards/statusline run config
How It Works
User prompt → Hook detects project type → Expert agent activated
→ Hook loads SOLID references → Agent reads docs via MCP
→ Hook blocks if DRY violation → Agent writes code
→ Hook checks file size → Agent runs Sniper validation
→ Commit gate blocks on lint/type errors → Commit with version bump
Every step is intercepted. Claude cannot skip research, cannot duplicate code, cannot exceed file size limits, and cannot commit with lint or type errors.
Plugins
Development — Framework Expert Agents
Each plugin provides an expert agent that auto-activates when it detects the framework in your project. The agent has access to official documentation via MCP servers and follows SOLID principles enforced by hooks.
| Plugin | Detects | What the agent does |
|---|
| fuse-nextjs | next.config.* | App Router, RSC, Prisma 7, Better Auth, proxy.ts patterns |
| fuse-laravel | composer.json + artisan | Eloquent, Livewire, Blade, queues, Sanctum, Stripe Connect |
| fuse-react | package.json + React | React 19 hooks, TanStack Router/Form, Zustand stores |
| fuse-astro | astro.config.* | Islands, Content Layer, Actions, SEO, Starlight, i18n |
| fuse-swift-apple-expert | Package.swift | SwiftUI, concurrency, all Apple platforms (iOS → visionOS) |
| fuse-tailwindcss | tailwind.config.* | v4.1 CSS-first config, @theme, @utility, OKLCH colors |
| fuse-design | Any UI task | Gemini Design MCP + shadcn/ui + WCAG 2.2 accessibility |
| fuse-shadcn-ui | components.json | Radix/Base UI detection, registry, theming, migration |
| fuse-tanstack-start | @tanstack/react-start | Server functions, selective SSR, middleware, server routes, Nitro deploy |
| fuse-typescript | tsconfig.json (no framework) |
Quality & Security — Automated Enforcement
| Plugin | What it enforces |
|---|
| fuse-ai-pilot | APEX workflow orchestration, 7-phase sniper validation, DRY detection via jscpd, 4-level cache (60-75% token savings) |
| fuse-security | OWASP Top 10 scanning, CVE research via NVD/OSV.dev, dependency audit, secrets detection, auth patterns audit |
| fuse-solid | Files < 100 lines, interface separation, JSDoc mandatory — auto-detected for TypeScript, PHP, Swift, Go, Java, Ruby, Rust |
Productivity — Automation
| Plugin | What it automates |
|---|
| fuse-commit-pro | Conventional commits with security check, auto version bump, CHANGELOG, git tag — with a pre-commit secret-pattern scan |
| fuse-prompt-engineer | Prompt creation with CoT/Few-Shot/Meta-Prompting, A/B testing, 50+ template library, agent design |
| fuse-cartographer | Intelligent cartography: auto-generates navigable maps of plugins and projects at SessionStart with merge-preserving enrichment — agents navigate via linked index trees, /map --enrich completes descriptions from source frontmatter |
| fuse-changelog | Monitors Claude Code updates, detects breaking changes in plugins, gathers community feedback via Exa |
| fuse-lessons | Per-project "never reproduce" memory — writes compact timestamped lessons to MEMORY/LESSON.md after code edits and force-reads them at session/subagent start so mistakes don't recur across sessions |
Core (auto-installed, always active)
| Plugin | What it guards |
|---|
| core-guards | Blocks git push --force, rm -rf, npm install without lock, enforces SOLID file limits, tracks session state, DRY duplication blocking — also bundles token-optimization hooks (MCP verbosity caps, MCP + WebFetch disk cache, SessionStart cleanup) |
| claude-rules | Injects APEX/SOLID/DRY rules into every prompt so Claude never forgets the methodology |
Key Features
| Feature | How it prevents mistakes |
|---|
| APEX Workflow | Forces Analyze before coding — no more "let me just write this real quick" |
| Agent Teams | Parallel agents with exclusive file ownership — no merge conflicts |
| DRY Detection | Blocks Write/Edit if function/class name already exists — forces import or shared extraction |
| SOLID Hooks | Denies file save if >100 lines or missing SOLID reference read |
| Sniper Validation | 7-phase post-edit check: explore → research → grep → lint → fix → zero errors |
| 4-Level Cache | Caches exploration, docs, lessons, tests — 60-75% token savings |
| Token-efficient by default | Caps MCP verbosity (Context7/Exa numResults≤3, tokens≤2000), disk cache for MCP results + WebFetch with TTL (48h / 24h) and ripgrep lookup, auto-cleanup at SessionStart — typically saves ~150-200 KB per research-heavy session |
| 23 MCP Servers | Context7, Exa, Astro docs, Gemini Design, shadcn, fuse-browser, and more |
| Smart Commits | Security scan before commit, auto version bump, CHANGELOG, shields.io badge sync |
Documentation
| Section | Content |
|---|
| Getting Started | Installation, configuration, first steps |
| Workflow | APEX methodology, agents, skills, commands |
| Plugins | Per-plugin documentation and skills |
| Reference | Architecture, hooks, MCP servers, cache system |
License
MIT
