Claude Code Emergent Behavior: When Skills Combine
vibeandscribe.xyz
vibeandscribe.xyz
It's more than a single prompt, but less than an entire agent. I find skills to be the tools you use on the fly. Like how I might have a wrench,screw-driver, hammer in my tool box.
tools vs skills is all about context efficiency from what I see. and yes, this isn't novel of claude. but they are the first to offer this abstraction.
My point is that Skills are not the first to do this. Well written MCPs are dynamic workflow engines. Skills are like a more user focused and slimmed down version of MCPs.
It'd be interesting to see a comparison of a well written MCP compared to a skill in terms of task competency.
Now that you mention it, i can see a future where claude may offer a "skills" feature and codex offers a "talent" feature. where they are essentially the same things, but specific to that vendor.
reminds me how each cloud has the same offerings but different products.
Ideally I would build a bunch of atomic skills that combine well and claude just uses them naturally when the situation arises.
At some point I plan to do some ralph wiggum loop stuff maybe to hash out the best way for triggers to work.
Here's some prior work that might be interesting: https://scottspence.com/posts/how-to-make-claude-code-skills...
Edit: I realise it might look kind of weird I posted this link in 2 comments on this thread - disclaimer I am not the author of that random blog post, just sharing what I found!
That's it
(I've rephrased the sentence by removing "just" -- the novel thought here is that the agent's use of a skill can automatically trigger another skill, which is somewhat emergent.)
During testing today I asked a task I knew should have activated a skill and claude just did it without the skill instead.
This might help: https://scottspence.com/posts/how-to-make-claude-code-skills...
E.g. you could have:
- a set of skills to use design patterns of a library
- a skill to add to this skill-set -- either when prompted by user or autonomously via a stop-hook
E.g. I set up this combination for design patterns for the Langroid[1] LLM-Agent framework:
https://github.com/pchalasani/claude-code-tools/tree/main/pl...
That's what I have been doing as I'm building these skills out. There is a really fun testing/tdd loop that runs these skills through multiple prompts with baselines and tests it all out.
I was really curious to read it given the comments + full disclosure my co-founder recently wrote about a similar topic ("To Tool or Not to Tool") https://blog.codeyam.com/p/to-tool-or-not-to-tool .
I wanted to see how this is similar or different with the focus on Claude Code + Skills in a more literal sense vs. tools in a more abstract sense.
(year fixed below)