Analyze any codebase and generate optimized Claude Code configurations — skills, agents, MCP servers, guardrails, CLAUDE.md, and settings — in seconds.
Catalyst deeply analyzes your codebase and generates:
- CLAUDE.md — Project context, conventions, commands, and critical rules so Claude understands your codebase from the first message
- Skills — Tailored workflow automations based on your actual tech stack (test runner, API builder, component builder, deployment, etc.)
- Agents — Specialized sub-agents for code review, architecture design, test writing, and security auditing — configured for your specific patterns
- MCP Servers — Model Context Protocol server configurations for databases, GitHub, cloud services, and other integrations detected in your codebase
- Guardrails — Safety configurations including deny rules, hooks, and protected file rules that prevent destructive actions while keeping Claude productive
- Settings — Granular permission allowlists so Claude can work efficiently without constant permission prompts
# Install globally
npm install -g @crashbytes/catalyst
# Or use npx
npx @crashbytes/catalyst init# Install Catalyst into your project
cd your-project
catalyst init
# Check installation status
catalyst status
# Remove Catalyst from your project
catalyst removeAfter installing, use these in Claude Code:
/catalyst Full analysis + generate everything
/catalyst analyze Analysis report only (no files generated)
/catalyst skills Generate skills only
/catalyst agents Generate agents only
/catalyst mcp Generate MCP server configs only
/catalyst guardrails Generate guardrails only
/catalyst claudemd Generate CLAUDE.md only
/catalyst settings Generate settings.local.json only
- Tech Stack — Languages, frameworks, libraries, build tools, databases, deployment targets
- Architecture — Code organization patterns, directory structure, entry points
- Code Patterns — API routes, database access, auth, error handling, validation, state management
- Conventions — Naming, imports, file organization, testing patterns, commit style
- Dev Workflow — Build commands, dev server, test runner, linter, deployment pipeline
- Git History — Branch conventions, commit style, active development areas
- Integrations — Database connections, cloud SDKs, external APIs, monitoring services, browser automation
- Security Posture — Secret management, existing hooks, protected files, sensitive data patterns, compliance markers
A concise (<200 lines) project reference containing:
- Tech stack summary
- Key directories and entry points
- Common commands (dev, build, test, deploy)
- Code conventions and critical rules
- Schema/validation warnings
| Detection | Generated Skill |
|---|---|
| Test framework | test-runner — run, write, and manage tests |
| API routes | api-builder — create endpoints following project patterns |
| UI components | component-builder — scaffold components with proper conventions |
| Database/ORM | db-ops — migrations, schema changes, queries |
| CI/CD pipeline | deploy — build, validate, and deploy |
| Content files | content-writer — create content with proper frontmatter |
| Docker | containerize — Docker operations |
| Multiple envs | env-manager — environment variable management |
| Detection | Generated Agent |
|---|---|
| Any project | code-reviewer — bug detection, security, conventions |
| Test framework | test-writer — write tests following project patterns |
| Complex architecture | code-architect — design approaches, evaluate trade-offs |
| Large codebase | code-explorer — trace execution paths, map dependencies |
| Auth/security | security-auditor — OWASP checks, auth review |
| Database/API | performance-analyzer — N+1 queries, missing indexes |
| Detection | Recommended MCP Server |
|---|---|
| PostgreSQL/MySQL/SQLite | database — query and inspect schema |
| GitHub workflows | github — PRs, issues, code review |
| AWS/GCP/Azure SDKs | Cloud-specific MCP — service management |
| Puppeteer/Playwright | puppeteer — browser automation |
| Slack SDK | slack — team communication |
| Sentry/Datadog | monitoring — error tracking, metrics |
| External APIs | fetch — HTTP requests to whitelisted domains |
| Redis | redis — cache inspection |
- Deny rules — Block destructive commands (force push, drop database, rm -rf) and protect critical files (lock files, production configs, CI/CD pipelines)
- Hooks — Pre/post-edit validation (linter checks, secret detection, format enforcement)
- CLAUDE.md safety rules — Protected file lists, forbidden actions, required validations, security boundaries
- Universal rules applied to all projects, plus language-specific and domain-specific guardrails
- Granular bash permissions for detected package manager, build tools, test runners
- Git operation permissions
- Web fetch domains for detected framework documentation
- Deny rules and hooks from guardrail generation
- Never auto-allows destructive operations
- Specific, not generic — every generated file references your actual paths, patterns, and conventions
- Quality over quantity — fewer, better skills beat many mediocre ones
- Merge, never overwrite — respects existing Claude Code configuration
- Convention-aware — detects and encodes your project's established patterns
- Security-first — never auto-allows destructive operations or broad permissions
catalyst/
├── bin/catalyst.js # CLI entry point
├── plugin/
│ ├── .claude-plugin/
│ │ └── plugin.json # Plugin manifest
│ ├── skills/catalyst/
│ │ ├── SKILL.md # Main analysis + generation skill
│ │ ├── skill-templates.md # Templates for generating skills
│ │ ├── agent-templates.md # Templates for generating agents
│ │ ├── mcp-templates.md # Templates for MCP server configs
│ │ └── guardrail-templates.md # Templates for guardrail generation
│ ├── commands/
│ │ └── catalyst.md # /catalyst command
│ └── agents/
│ └── codebase-analyzer.md # Analysis agent
├── package.json
└── README.md
MIT - CrashBytes