Nathaniel Whittemore

That's what Anthropic has been doing here, building a unified system that connects dots.

Simon Willison discussed in his blog how skills are awesome and maybe a bigger deal even than MCP, which is, of course, in bold words.

First, he described what skills are.

He said that conceptually, they're extremely simple.

They're a markdown file telling a model how to do something, optionally accompanied by extra documents and pre-written scripts that the model can use to help it.

However, he writes, there's one extra detail that makes this a feature, not just a bunch of files on disk.

At the start of a session, Claude's various harnesses can scan all available skill files and read a short explanation for each one.

This is very token efficient.

Each skill only takes up a few dozen extra tokens with the full details only loaded in should the user request a task that the skill can help solve.

He discussed how he would build a data journalism agent using the feature.

The skills that he would build were things like how to access and parse census data, how to load data from SQL and DuckDB with associated Python code provided, how to publish data online, and how to figure out an interesting story to tell based on the data.

Simon concluded, The core simplicity of the skills design is why I'm so excited about it.

Skills are marked down with a tiny bit of YAML metadata and some optional scripts in whatever you can make executable in the environment.

They feel a lot closer to the spirit of LLMs, throw in some text and let the model figure it out.

They outsource the hard parts to the LLM harness and the associated computer environment.

Given everything we have learned about LLM's ability to run tools over the last couple of years, I think that's a very sensible strategy.

Put differently, I think skills are in some ways a different user experience pattern for getting at agent creation.

That's really what this is about.

On the one hand, you've got the end-to-end style agent workflow builder, but this is basically where you articulate component parts and then can use natural language to help the LLM itself figure out which of those parts it needs to put together.

Again, in Simon's example, he's got a skill for parsing census data, a skill for loading data, a skill for publishing data online, and a skill for figuring out an interesting story.