My Toolkit
The plugins, skills, CLIs, and hooks installed on my laptop, and why each one stays.
This page is mine. The guide tells you how to work with AI agents. Here is the concrete setup I run it on. Everything here is installed on one laptop, around Claude Code as the main agent.
- Workflow: how a change goes from idea to merged PR, stage by stage.
- Skills: plugins and skills that change how the agent works.
- Tools: CLIs, MCP servers, other agents, and the hooks that wire them in.
- Hot takes: opinions about working with agents, and why I hold them.
How it fits together
| Layer | Job | What I use | Page |
|---|---|---|---|
| Process | Take a change from idea to merged PR | superpowers, /ponytail-review, proving-it-works, no-mistakes | Workflow |
| Behaviour | Change how the agent thinks and talks | ponytail, caveman, superpowers | Skills |
| Review | Catch design and code mistakes | swe-guidelines, arch-*, grilling | Skills |
| Design | Get from idea to a UI I can point at | Figma plugin, ui-ux-pro-max, /mock-options | Skills |
| Skill library | Reusable ways of working | pstack | Skills |
| Context savers | Keep junk out of the context window | rtk, codegraph | Tools |
| Gates | Stop bad code before it leaves the machine | no-mistakes, CodeRabbit | Tools |
| Other agents | Second opinions and local models | Codex, Cursor agent, Ollama | Tools |
| Workspace glue | See what every agent is doing | herdr, iTerm status hooks, status line | Tools |
The rule behind the whole list: a tool earns its place by either saving context or catching a mistake. Anything that does neither gets turned off, because every enabled plugin and skill costs tokens in every session just by being listed.