Reduce token costs - in internal testing, the Highly Opinionated Agent Skill Fixer reduced Skill token costs an average of 20%.
Get predictable outcomes - by using structured systems like templates, assets, references, and evals, AI has less wiggle room and more guardrails.
Convert any existing Skill, Gem, GPT, or prompt into a powerful Skill!
What's in the box?
You receive the Skill Fixer skill, which you can install into any software that supports the Agent Skills specification, and includes but is not limited to: - Claude Cowork / Claude Code - ChatGPT Work - OpenAI Dots - Gemini Spark - Google Antigravity - Microsoft Copilot Cowork - OpenCode/Pi/OpenWork - Grok Bot - Meta Muse
Frequently Asked Questions
What does Skill Fixer do for me? It repairs an AI skill so that it passes about 100 checks. A skill is a folder of instructions. It teaches your assistant one task. Skill Fixer finds every fault, fixes it, and shows you what it changed. It can also start from one text file. Give it a prompt, a Gem, or GPT instructions. It builds a skill from the file, then repairs it. It does not move GPT knowledge files or actions. Why not ask my assistant to clean up the skill? A model does not give the same answer every time. A script does. Skill Fixer uses a script wherever a script can do the job. Scripts find the faults, fix the mechanical ones, and count the tokens. Scripts check each stage of the repair. The assistant does only the judgment work: restructure, rewrite, and review. You get a repeatable repair, not a one-time guess. Can it damage or lose my work? It protects your text in three ways. It saves a snapshot of your skill before it changes anything. A script lists every old line that is not in the new skill. The assistant gives a reason for each line. You approve the list. It never deletes text to pass a rule. It moves the text word for word. It adds no advertisement or credit line to your skills. The line check compares words, not punctuation. Keep a backup of your own as well. How will I know that it worked? It writes a report. The report shows the errors and warnings before and after. It also shows the token cost before and after. An optional measure step runs three tests on the repaired skill. Activation: does the skill start when it should? Outcome: does it do its job as well as the old skill? Efficiency: how many tokens and how much time does it use? The measure step uses model time. Skill Fixer shows the number of model calls and asks you first. If you skip the step, the report says "not measured". It does not hide this. Do I need to write code, and will it work in my tool? You need no code. You type one request, such as "Use skill-fixer to repair this skill." Your assistant runs the scripts. It works in tools that support Skills. Examples: Claude Code, Claude Cowork, ChatGPT Work, Gemini Spark, Google Antigravity, Microsoft Copilot Cowork, and OpenCode. It does not work in web chat tools such as ChatGPT chat or Gemini chat. The scripts need Python 3.11 or later. If your tool cannot run Python, the skill continues by hand. That is slower and less repeatable.