Claude Code plugin that scaffolds the full .claude/ agentic engineering stack for TypeScript/monorepo projects.
Defense-in-depth: CLAUDE.md, path-scoped rules, specialist agents, lifecycle hooks, and scoped permissions.
/plugin marketplace add pax-k/paxify-plugins
/plugin install agentic-init@paxify/paxify:agentic-initThe plugin analyzes your codebase and generates a complete .claude/ directory tailored to your project.
Project constitution (~80-100 lines). Commands, workflow, key rules. Acts as a router to detailed context — not a novel.
Path-scoped instructions that load only when Claude touches matching files:
- code-style.md — TypeScript strict, naming, imports (always loaded)
- git-conventions.md — conventional commits, PR workflow (always loaded)
- testing.md — test runner, assertion rules (loaded for test files)
- security.md — OWASP, input validation, no secrets (always loaded)
- frontend/react-patterns.md — framework-specific patterns (loaded for frontend files)
- backend/api-design.md — API design, DB access layer (loaded for backend files)
Specialist subagents with isolated context and scoped tools:
- architect.md — read-only planner, produces ADRs
- implementer.md — TDD workflow: tests first → implement → verify
- reviewer.md — adversarial reviewer with zero prior context
- debugger.md — systematic root-cause debugging
- docs.md — documentation updater
- judge.md — neutral evaluator for competing implementations
Reusable workflow templates:
- compete/SKILL.md — runs 3 competing implementations (minimal, thorough, creative), judge picks the winner
Shell scripts at lifecycle events. Exit 2 = blocked, no negotiation:
- pre-bash-firewall.sh — blocks dangerous commands, enforces package manager
- protect-files.sh — blocks edits to .env, lock files, node_modules
- auto-format.sh — runs formatter after every edit
Permissions allowlist + hook wiring. Generous permissions for safe operations, hooks catch the dangerous ones.
Phase 1: Analyze — A read-only codebase-analyzer agent explores your project and produces a structured profile: package manager, monorepo structure, framework, test runner, formatter, linter, database layer, git conventions, file patterns.
Phase 2: Generate — The /agentic-init skill reads the profile and adapts template files to your project. Every placeholder is replaced with detected values. Conditionals handle monorepo vs single-project, Next.js vs other frameworks, etc.
| Category | Detects |
|---|---|
| Package manager | pnpm, bun, yarn, npm (from lock files) |
| Monorepo | Workspaces, workspace types (frontend/backend/library) |
| Framework | Next.js, Remix, Astro, Express, Fastify, Hono, tRPC |
| Test runner | Vitest, Jest, Bun test, Playwright, Cypress |
| Formatter | Prettier, Biome |
| Linter | ESLint, Biome, oxlint |
| Database | Drizzle, Prisma, TypeORM, Knex |
| Git | Commit style, protected branches, CI config |
This plugin generates the guardrails layer. For spec-driven workflow (plan → implement → review), use OpenSpec.
After generation, review and customize:
- Edit
CLAUDE.mdto add project-specific rules - Adjust path globs in rules to match your directory structure
- Add/remove agent capabilities as needed
- Tune hook scripts for your team's conventions
- Modify
settings.jsonpermissions for your tooling
MIT