What's this about?
People have started publishing and curating agent skills the way they once published plugins. OrangeAI began a regular series of recommended Claude skills, opening with five aesthetically-focused ones: an editorial photo abstraction skill, a HeyTea-style poster generator, a minimal zine-style poster maker, a personal-IP illustration generator, and a skill that redraws ordinary snapshots as poetic hand-drawn illustrations. Separately, Theo praised a "grill-me" skill for interrogating him until the agent is aligned with how he actually thinks. OpenAI's enterprise data points the same direction: top-decile firms use skills six times as often as typical ones.
What exactly is a "skill"?
A skill is a reusable package of instructions — and sometimes scripts and reference files — that you drop into an agent to teach it a specific way of working. Rather than re-explaining your poster design conventions every time, you write them down once and the agent loads them when relevant. The plugin analogy is apt but incomplete: a plugin adds a capability the program lacked, while a skill mostly shapes judgment and taste in a capability the model already has.
Why is this worth noticing?
- Skills are becoming shareable artifacts with curators and taste-makers — the earliest sign of an ecosystem rather than a feature.
- The "grill-me" example is a different genre from the design skills: it isn't a capability at all, it's a protocol for extracting the user's intent, which suggests skills are being used to fix the alignment-with-me problem as much as the do-this-task problem.
- The 6x usage gap between top firms and typical ones suggests skills are a competence that separates organizations, not just a convenience — and today's SkillZip paper on compressing large skill libraries indicates the research community sees the same growth curve.