Steveskills — Development skill for Claude Code

steveskills

A directory of agent skills by Steve Solun. A **skill** is a folder with a `SKILL.md` that an agent reads when a task matches its description, plus whatever scripts and reference files the instructions need. The agent keeps only the one-line description in context until it needs the rest, so a skill costs nothing until it fires.

Every skill here follows the same rules:

• One job. Each folder does one thing and says in its description exactly when it should and should not fire.
• Numbers, not advice. Where a skill makes a claim about tokens, time or correctness, it ships the script that measures it and reports before and after numbers.
• Fixes live in the environment. A skill leaves behind a flag, a script, a check or a config, never a paragraph in a steering file that every session has to re-read.
• One copy, every harness. Written to the open Agent Skills format, with Codex and ChatGPT picker metadata beside it, so the same folder works in Claude Code, Codex, ChatGPT, Copilot, Cursor and anything else that reads the standard.

The directory grows one folder at a time. The table below is the index.

Skills

Skill	What it does
tool-diet	Puts tool results on a diet. Measures which tool, CLI, script or MCP results in a session cost the most tokens for the least use, cuts the fat at its source, and proves the saving with before and after numbers.

More folders will land here. Each one carries its own README with details and sample output.

Install

Pick one route per agent. Installing a skill twice makes it show up twice.

Any agent, with the skills CLI

npx skills@latest add stevesolun/steveskills --skill=

The CLI asks which agents to install for and puts the files where each one looks. With `tool-diet` as the example:

npx skills@latest add stevesolun/steveskills --skill=tool-diet -g            # user-wide instead of this project
np