My skill-maker tool actually outperforms Anthropic's official skill-creator tool. If anyone want to create any custom skills for their AI, let me know, and I can help you get set up.

https://skill-maker.pages.dev/

skill-maker outperforms the official Claude skill-creator skill. In a test of some of the most challenging dev tasks (DB Migrations, PDF toolkit, etc), skill-maker generated skills learned to perform at 100% whereas Claude's skill-creator achieved 93.1% performance.

Read more here: skill-maker.pages.dev

#ai #skills

Just added a bunch of nostr-related AI skills built with skill-maker: https://github.com/accolver/skill-maker

bunx skill-maker install --all

My new skill-maker tool increases the success rate of medium-high complexity tasks from as low as 4% up to 100%

Complex tasks like DB migrations, PDF data extraction, robust API implementations, etc are evaluated (harshly) by LLMs. Initially their base evaluation is on a simple prompt, like "extract the data from this PDF". skill-maker turns this naive prompt into a thorough prompt with custom scripts and tools. After several iterations a final skill is made. All skills I've created so far, pass at a 100% rate after using agent skills generated by skill-maker.

Check it out at https://skill-maker.pages.dev/

Deep dive into the analytics: https://github.com/accolver/skill-maker/tree/main/examples

Open source and completely free.

I created an agent skill that will help you make your own agent skills. https://github.com/accolver/skill-maker

Using opus-4.6 WITHOUT this repo gives a 57% quality score to one-shotting your skill; with the repo it's 100% success rate.

The fact that so many things still require a wet signature over a digital signature is baffling.