Join upcoming workshop on Claude skills at AI Builder Club: https://www.aibuilderclub.com/     See Less

it's basically an attempt to standardize local tools ? Code agents will sometimes create tools that are discarded after being used. Like this it'll be possible to reuse them. It can     See More to parse logs, tidy the code (e.g. remove the horrible tailing whitespaces generated by Claude models) and any other recurrent task.    See Less

Isn't this the same as memory bank?     See Less

I think the token savings over MCP isn't given enough credit, it's insane news for indie game developers. That's the best use-case that immediately comes to mind. One could vibe     See More tire competent game in something like RPG Maker now without ever having to open the actual program (as even with MCPs a lot of game engines have files that are 5-6-figures of Tokens making them almost impossible for Agents to work with consistently).    See Less

When i wonder if something is overhyped or not, i wait for jason to talk about it. If he talks about it, I start looking at it more seriously. This trust has been cemented time and time agai     See More     See Less

your channel is like benchmark for me to take serious implementation of new release and protect myself from overhype.     See Less

Whatever other GenAI companies does, Anthropic does it better.     See Less

Well done, very practical review, as usual. To me, when MCP was out, there were two uses: one was tool-calling, the other was providing context. The former worked well; the latter did not. N     See More ill, "a zip file containing text and scripts" is a far better replacement, serving as executable context rather than plain text. Eventually this “skill” will become an operational sub-agent that users and the system share.    See Less

Having all these tools is great but at the end of the day it all depends on how well the LLM model works.     See Less

1. The name “skill” is confusing. 2. “Better than mco? Really? This doesn’t even compare to what mcps bring to the game. 3. 4k to 70 tokens? There is no way, as long as that long tex     See More in the “skill” is sent to the llm, the text itself is much more than 70 tokens, and that doesn’t even include the output.    See Less