A skill is a named bundle — instructions, tools, prompts, optional data — that the model can load on demand. Build one skill, and it works the same way in chat, in workflows, in agents. The pattern your team uses every Monday becomes a verb the model already knows.
A skill bundles the thing the model needs to know with the things it needs to call. Loaded by name, available everywhere.
Plain‑language guidance the model reads when the skill loads — what good output looks like, which edge cases to handle, which formats to prefer. Written once, applied every time it's used.
A short list of the exact tools and connected servers the skill needs. The model picks from this list, not the kitchen sink — a focused set means fewer wrong calls and faster reasoning.
Reusable prompt building blocks — the opening, the structure, what to do with the result. Skills work hand in hand with prompts: a skill can teach the model how to use a saved prompt template — what to fill in, what to expect back, how to check it.
Optional: a data source, a slice of vector memory, an attached set of docs. The skill carries its own reference material, so the model doesn't have to go re‑search the basics every time.
Skills aren't tied to chat or to workflows — they're a building block that works across every surface. Define earnings‑prep once; every team member who loads it gets the exact same behavior.
The model calls load_skill("X") when the conversation calls for it, or you can pin a skill to a session yourself. Its guidance folds into the model's instructions, and its tools join the ones already on hand.
The go‑to example: workflow‑author is the skill that teaches the model how to write a workflow. Load it before you save a chat as a workflow, and the result comes out in the right shape, every time.
An agent can take a skill as its "operating mode" — the model starts the run with that skill's instructions and its set of allowed tools already in place. Different skill, different agent personality, same engine underneath.
Admins publish skills across the org through a shared catalog — anyone on the team can load them. You can tweak your own copy (different model, different prompt) without touching the shared original.
A skill lives in one place. Update it, and every chat, workflow, and agent that uses it picks up the change the next time it loads. No copy‑paste, nothing drifting out of sync.
One skill can call load_skill on another. Build small building blocks — "format as memo", "cite sources" — and have your bigger skills pull them in.
The skills concept earns its keep when teams have repeated workflows. A handful of patterns we've seen people define and load constantly:
The repeated workflows your team runs — board reports, code reviews, standups — deserve to be building blocks, not screenshots in a doc. Skills are how they become verbs your AI knows.