A plugin ecosystem that turns Codex 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 Codex never guesses, never duplicates, and always follows your architecture.
Without plugins: Codex 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
- 148 hooks intercept every Write/Edit/Bash/MCP call in real-time to enforce file size limits (<100 lines), block code duplication (DRY), require SOLID references, and validate security
- 131 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 validation runs a 7-phase quality check after every modification: explore → research → grep usages → lint → fix → zero errors
# Add marketplace
/plugin marketplace add fusengine/codex-agent
# 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
# Setup (hooks + API keys + MCP servers + feature flags)
~/.codex/.tmp/marketplaces/fusengine-plugins/setup.sh # macOS / Linux
~/.codex/.tmp/marketplaces/fusengine-plugins/setup.ps1 # Windows (run via PowerShell)Statusline (optional):
bun --cwd ~/.codex/plugins/cache/fusengine-plugins/core-guards/<version>/statusline run configUser 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 → Sniper validates quality
→ Hook blocks secrets → Commit with version bump
Every step is intercepted. Codex cannot skip research, cannot duplicate code, cannot exceed file size limits, and cannot commit without security validation.
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 |
| 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 |
| Plugin | What it automates |
|---|---|
| fuse-commit-pro | Conventional commits with security check, auto version bump, CHANGELOG, git tag — blocks secrets from commits |
| 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 Codex updates, detects breaking changes in plugins, gathers community feedback via Exa |
| 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 |
| codex-rules | Injects APEX/SOLID/DRY rules into every prompt so Codex never forgets the methodology |
| 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 |
| 28 MCP Servers | Context7, Exa, Astro docs, Gemini Design, shadcn, Playwright, and more |
| Smart Commits | Security scan before commit, auto version bump, CHANGELOG, shields.io badge sync |
.codexignore |
Bloque les writes/edits sur fichiers sensibles (gitignore-style) — .env, secrets/, *.key jamais touchés |
| Transcript watcher | Capture les outils natifs Codex non-hookables (read_file, web_search, spawn_agent) via tail du rollout JSONL — alimente les guards APEX |
| Cross-platform TTS | Notification audio fin de tâche : macOS (afplay), Linux (paplay/aplay/mpv/ffplay), Windows (SoundPlayer) |
| Codex feature flags | Audit 0.130 : hooks, tool_search, personality, multi_agent, fast_mode, shell_snapshot, enable_request_compression, skill_mcp_dependency_install, memories, goals (defaults) + 6 prompts (memories, apps, approval_policy, sandbox_mode, web_search, model_reasoning_effort) |
| Codex 0.130 compliant | Token-economy model mapping (opus → gpt-5.5 high, sonnet → gpt-5.4 medium, haiku → gpt-5.4-mini low), subagent filenames <plugin-slug>-<agent>.toml, per-agent MCP scoping (research-expert, design-expert, websearch), [agents] global config, opt-in approval_mode prompt, 131 skills with strict name+description-only frontmatter |
| 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 |
MIT