My agentic toolbox. Batteries included. Node.js excluded.
┌─────────────────────────────────────────────────────────────┐
│ "What if coding agents were opinionated about everything?" │
└─────────────────────────────────────────────────────────────┘
| Agent | Status | Notes |
|---|---|---|
| Claude Code | ✅ Supported | Full integration |
| Droid | 🔜 Planned | — |
| Cursor | 🔜 Planned | — |
| Codex | 🔜 Planned | — |
| Opencode | 🔜 Planned | — |
| Gemini CLI | 🔜 Planned | — |
| Goose | 🔜 Planned | — |
| Kilocode | 🔜 Planned | — |
| Aider | 🔜 Planned | — |
| Cline | 🔜 Planned | — |
.claude/
├── agents/ # The workforce
│ ├── idea-scaffolder.md → Builds specs through interrogation
│ └── skill-generator.md → Creates skills that create skills
│
├── skills/ # The slash commands
│ ├── generating-skills/ → /generating-skills - meta as it gets
│ └── scaffolding-ideas/ → /scaffolding-ideas - death by questions
│
├── settings.json # The rules
│ ├── hooks → Bun or bust
│ ├── permissions → Trust but verify
│ └── deny list → node? npm? npx? lol no
│
└── settings.local.json # The secrets (gitignored)
One runtime to rule them all. This repo enforces Bun exclusively. Try to run node or npm and watch the hooks stop you cold:
HOOK BLOCKED: Node.js command detected. IMMEDIATELY retry with bun equivalent:
node script.js → bun script.js
npx package → bunx package
npm run X → bun run X
Write code with Node.js imports? The pre-write hook has opinions about that too.
A skill that generates skills. Yes, it generated parts of itself.
/generating-skills reviewing-code --depth=standard
Creates a complete skill scaffold:
- SKILL.md with proper frontmatter
- manifest.json for machines
- examples/ for humans
- Validation that actually validates
For when you need to think harder. Uses gated phases to force you through structured ideation:
/scaffolding-ideas a distributed cache invalidation system
Then answer questions until you've accidentally written a complete spec. Uses Exa and Firecrawl for research-backed validation because your assumptions are probably wrong.
Gate phrase to advance: This is the perfect spec for this phase! Go to the next question section
Gate phrase to escape: This is the perfect scaffolding
| Agent | Purpose | Vibe |
|---|---|---|
idea-scaffolder |
Elicits specifications through relentless questioning | Socratic method meets product manager |
skill-generator |
Creates new skills with proper structure | Recursive self-improvement |
Pre-tool-use hooks that enforce the Bun-only policy:
| Tool | Hook | Effect |
|---|---|---|
Bash |
Blocks node/npm/npx | Suggests bun equivalents |
Edit |
Validates imports | No fs, no path, no Node.js |
Write |
Validates imports | Same deal |
Allowed: Everything useful (Read, Write, Bash, MCP tools, etc.)
Denied:
sudo(obviously)rm -rf /(nice try).envfiles (secrets stay secret)- Every Node.js package manager ever invented
# Clone it
git clone <this-repo> agentix
cd agentix
# Install deps (with bun, obviously)
bun install
# Copy env and add your keys
cp .env.example .env.local
# Validate all skills
bun run validate:all
# Check for forbidden Node.js imports
bun run check:node| Variable | Required | Purpose |
|---|---|---|
EXA_API_KEY |
For research | Semantic search via Exa |
FIRECRAWL_API_KEY |
For research | Web scraping via Firecrawl |
Without these, /scaffolding-ideas works but can't validate assumptions against the real world.
bun run validate # Validate a skill
bun run validate:all # Validate all skills
bun run check:node # Find Node.js imports (and judge them)
bun run typecheck # TypeScript without the runtimeConfigured for:
mcp__exa__*— Semantic search for researchmcp__firecrawl__*— Web scraping for documentation
This repo is self-documenting, self-validating, and mildly self-aware:
- Skills validate themselves via
validate-skill.ts - Hooks enforce the rules that hooks are written in
- The generating-skills skill can update itself
- The scaffolding-ideas skill was scaffolded by scaffolding-ideas
It's agents all the way down.
Soon™
──────
[ ] /git → Commit workflow that follows conventions (you have one, right?)
[ ] /reviewing-code → Opinionated code review. Your code has opinions about itself.
Later™
──────
[ ] /refactoring → Guided refactoring with before/after validation
[ ] /debugging → Hypothesis-driven investigation. Socratic method for stack traces.
[ ] /testing → Test generation that understands your patterns
[ ] /migrating → Schema/API/dependency migrations without the tears
Eventually™
───────────
[ ] /researching → Deep technical research via Exa/Firecrawl
[ ] /benchmarking → Performance testing with Bun's built-in benchmarking
[ ] /analyzing-deps → Security audit, bundle impact, "why do I have 847 packages"
[ ] /hooking → Meta-skill to generate hooks. Hooks generating hooks.
[ ] /prompting → Prompt engineering. Prompts about prompts.