The agent skills we actually use

We talk a lot about working in the new age of AI tooling. This is some of what that actually looks like, in the open.

agent-skills is our curated collection of skills for agentic coding assistants - packaged instructions and scripts that give an agent a repeatable way of working. It is not a demo. It is the toolkit we reach for on real client and product work every day.

The collection has grown into a pipeline rather than a pile of tricks: discovery sharpens a loose idea into a goal and scope, plan turns that into a concrete implementation plan, and the implement and oneshot skills execute it through a disciplined, git-centric workflow with the human in the loop. There is also agent-feedback, for when an agent has wandered off and you need to work out why and steer it back. Our April-into-May releases reshaped all of it around that plan-then-build spine.

It is free, open source under the MIT licence, and lives on GitHub. Borrow whatever is useful.

Alongside the open source, a few larger things are taking shape behind the scenes: a total rewrite of DragnDrop.me, a microhosting manager we call HostHub that we are already leaning on heavily internally, a client portal, and a small-business server for macOS. They are not ready to show yet, but they are why it has been quiet here. We will write properly about each as it lands.