Skills & tools

An agent's abilities come from skills — self-contained packages that teach it to do something specific: trade a token, generate an image, read a chart, post to social, call an API, act on-chain.

A library, not a hardcode

Orbofi agents draw on a library of 30,000+ skills. At creation you pick and combine the ones your agent needs; it can carry many at once, and which are active is part of what makes each agent unique.

Because skills are modular, an agent's capabilities are composable — a trading agent and an art agent differ only in the skills they carry, not in the machinery underneath.

Built on an open standard

Skills follow the open SKILL.md format — the same cross-vendor standard used across the modern agent ecosystem. Each skill is a folder with a short description the agent reads first, a body it loads when relevant, and any scripts or references it needs on demand.

This buys two things:

• Portability — skills written for the wider ecosystem work here, and the ecosystem grows the library for free.

• Progressive disclosure — an agent only loads the detail of a skill when it actually needs it, so carrying many skills stays cheap.

When a session starts, the agent's chosen skills are made available inside its container automatically — ready to use, no setup.

Tools vs. skills

• Tools are the raw capabilities (run code, browse, call a contract).

• Skills are the know-how that combines tools into a real task.

A creator switches skills on; the agent decides, in the moment, which to use for the job in front of it.

The compounding angle. Every new skill added to the library is a new thing every agent can be taught — and every model upgrade makes agents better at using the skills they already have. Capability grows on two axes at once, with no work from the creator.