OP here. I wrote this because I feel the term "Agent Skills" is getting a bit mystical, when in reality it's just a nice pattern for progressive disclosure.
I implemented the Anthropic skill standard using claudette (thanks @jeremyphoward) to see how minimal I could make it. Turns out you only need about 100 lines of Python to support the full spec.
The code implements the three basic tools (skill, read, run) without any complex orchestration framework. Let me know what you think!
The scary part for me is uncommenting it, because the code has been removed from normal QA and testing (automatic or manual) and you don't know what evils you are unleashing upon the world.
I implemented the Anthropic skill standard using claudette (thanks @jeremyphoward) to see how minimal I could make it. Turns out you only need about 100 lines of Python to support the full spec.
The code implements the three basic tools (skill, read, run) without any complex orchestration framework. Let me know what you think!