If your AI coding tool doesn't have safety gates, you don't have a tool. You have a liability.
AI-generated code has 41% more bugs than human-written code. These 5 skills won't fix that on their own — but they're part of a system that does.
This repo contains 5 production-grade skills extracted from the Mrs. Kitty AI Controller. They work standalone in any Claude Code setup. No dependencies, no lock-in.
Browser automation for AI agents. Navigate pages, fill forms, click buttons, take screenshots, scrape data — all from Claude Code. Structure-first verification that cuts token usage by ~98%.
Try it: Install the plugin, then ask Claude to "check if my site has any console errors"
✍️ Humanizer
Strips AI-generated writing patterns from text. Based on Wikipedia's "Signs of AI writing" guide — 24 specific patterns detected and fixed. Your copy reads like a person wrote it.
Try it: Paste any AI-generated text and say /humanizer
Builds conversion-optimized interfaces. Fuses Hormozi's value equation with GSAP/Lenis animations and OSMO-level techniques. Mandatory 12-point quality gate blocks generic output.
Try it: Say /frontend-design and describe a landing page
Self-orchestrating parallel browser agents. Need to verify 10 pages at once? Research 8 products simultaneously? This decomposes tasks into micro-agents, opens all tabs in one call, and dispatches reasoning agents in parallel.
Try it: Edit 3+ frontend files, then say "verify all my changes"
🖼️ Poe Media
Generate images, videos, edit photos, and upscale — all through Poe's API. Smart model routing picks the right generator for your task. Supports 9 image models, 4 video models, and chained workflows (image → upscale → animate).
Try it: Say /poe-media and describe what you want to create
claude plugin install CalebDane7/kittyaiOr clone and point Claude at it:
git clone https://github.com/CalebDane7/kittyai.git
claude --plugin-dir ./kittyaiThe full Mrs. Kitty AI Controller includes:
- 33+ safety gates that enforce research, planning, testing, and review before any code edit
- 97% confidence threshold — AI won't touch your code until it proves it understands the problem
- Adversarial plan review — a second AI tears apart the first AI's plan before you see it
- Persistent memory — learns from every session, never repeats the same mistake
- Voice dictation — speak at 150 WPM instead of typing at 40
- Parallel browser testing — tests your pages simultaneously in real Chrome
- Token compression — Rust-powered proxy saves 80%+ on API costs
One installer. 8 minutes. Free.
MIT — use these skills however you want.
Built by @CalebDane7. If these skills save you time, star the repo.