My AI Toolkit
My always-up-to-date list of tools & workflows that actually work in real life.
I’m addicted to minmaxing. Diablo 4 builds, building track cars, writing fast code, I have a deep need to optimize. My AI tooling is no different. I’ll share with you what tools that I consider my daily drivers and why I use them.
Claude Code
We’ll just get this out of the way. Claude Code is my agentic platform of choice (for now). I desperately want OpenCode and an open source model to be viable. However, as of right now, this is what the economics look like based on my real usage:
| Setup | Total/day | Monthly (22d) |
|---|---|---|
| Claude Code Max (subscription) | — | $200 |
| DeepSeek V3.1 | $83.70 | $1,841 |
| Qwen3 Coder 480B | $91.26 | $2,008 |
| DeepSeek V3.2 Exp | $96.32 | $2,119 |
| Kimi 2.6 only (no executor) | $154.08 | $3,390 |
| Claude API (list price) | $261.63 | $5,756 |
Context7
This is one of the lowest hanging fruits, highest yield tools I have. It provides an MCP for agents to verify real documentation during planning & execution phases (preferably during planning).
I find that just adding a link to the docs in my CLAUDE.md or AGENTS.md file doesn’t work as well or as consistently as installing context7. It tends to give me far more accurate implementions the first time around. I attempted to create something of my own called refdocs-cli, but simple grep & a local folder blew it out of the water.
It’s (mostly) free until you hit multi-user team kind of stuff.
Pull up-to-date, version-specific documentation and code examples for any library directly into Cursor, Claude Code, Windsurf, and other AI coding tools.
(opens in a new tab)Claude Super Powers
Claude Superpowers is basically a ton of current best practices baked into one suite of skills.
It gives you:
- Brainstorming
- Worktrees
- Planning
- Sub-agent orchestration
- TDD (Test driven development)
- Code Reviews
There’s a more detailed breakdown from the repo.
An agentic skills framework & software development methodology that works. - obra/superpowers
(opens in a new tab)Bully
This is my own skill that hooks into the PostToolUse hook in Claude Code to enforce rules.
I can enforce:
- Rules from linters & static analysis tools
- Raw bash script or
ast-grepscripts - Semantic LLM judgement calls
The goal of this tool is to short circuit code that goes against deterministic static rules
before they hit the codebase. It’s the enforcer arm of my CLAUDE.md file. This is so I can
keep my prompts shorter and tigher, so that I don’t use Claude as a linter (bad).
Contribute to dynamik-dev/bully development by creating an account on GitHub.
(opens in a new tab)