AI Skills
Skills in this context is the ‘skill’ that can be given to an LLM using the LLM Harness, in the case of Claude Code it is in a .claude/skills folder which is either install globally or locally to the folder/repo you are currently using.
A skill is something that you can invoke an pass a parameter on, i.e. the call shape is something to the below:
# the skill is the /sa-coach the `lets work on an architecture kata today` is the prompt/parameter
/sa-coach lets work on an architecture kata today
# you can call a skill on its own but it primarily depends on what the skill does.
/sa-coach
I found that skills are similar to a Prompt as it is basically just that, it is an instruction to a LLM to construct a consistent artifact/outcome/interaction that would normally get by just prompting a LLM.
A skill to me is something that you arrive at the same way you would when you have a repeated item/flow that you find yourself repeating the same prompt against your file system.
Great Skills across the Web
- GitHub - mattpocock/skills: Skills for Real Engineers. Straight from my .claude directory. · GitHub
- GitHub - blader/humanizer: Claude Code skill that removes signs of AI-generated writing from text · GitHub
📇 Additional Metadata
- 🗂 Type:: note
- 🏷️ Tags::
- 📡 Status:: #status/🌱