Skills
A Skill is a packaged, reusable set of instructions that gives an Agent a real capability. It is how the system learns to do something well, written down once and reused everywhere.
What a Skill is
Think of a Skill as a tight playbook for one job. It carries the know-how to do that job at a high standard, and any Agent can load it when the work calls for it.
Each Skill has two parts. A short description that signals when the Skill applies, always visible. And a body of detailed instructions, loaded only when it is actually needed.
This keeps Agents light. They do not carry every detail at all times. They pull in the right Skill for the moment, then set it down.
Capability lives in Skills. Orchestration lives in Agents.
Here is the core idea: capability lives in Skills, and orchestration lives in Agents.
An Agent is an autonomous worker that owns a domain and runs the work. A Skill is the packaged capability it reaches for. The Agent decides what to do and in what order. The Skill supplies the how.
So the system is not a sprawl of narrow bots. It is a few powerful Agents, each equipped with world-class Skills. A handful of genuine masters, not many narrow specialists pretending.
How the system grows: add a Skill
When a new capability is needed, you add a Skill. You do not spawn a new narrow Agent.
This is the rule that keeps the system clean. Skills are how the system grows. Agents are how it operates.
Adding capability through Skills means the same few Agents keep getting sharper, instead of the team multiplying into a crowded org chart of tiny bots.
Sub-Skills
Sometimes a capability needs to go deeper on one subtopic. That is what a Sub-Skill is for.
A Sub-Skill is a Skill that deepens another Skill for one specific area. It loads alongside the main Skill, never instead of it.
A Sub-Skill stays disciplined. It does not repeat what the main Skill already says. It reinforces only the one to three points that matter most for its subtopic, plus the details unique to that area.
For example, a research Skill can have a sub-Skill that deepens research on one topic. The main Skill covers the general approach. The Sub-Skill adds the specialized edge.
How every Skill and Agent gets built
Skills and Agents are not thrown together. Each one is built through a disciplined flow, so the result feels inevitable, like an elite operator and an AI refined it together for hours.
The default progression runs like this:
| Step | What happens |
|---|---|
| Line it up | Pin down the exact outcome the work should produce. |
| Research | Study the problem and gather real, cited evidence. |
| Plan | Decide the shape of the build before writing it. |
| Build | Write the actual Skill or Agent. |
| Test | Run it and watch what it really does. |
| Refine | Tighten and sharpen, keeping what already works. |
| Deploy | Put it live. |
| Check | Verify it works, with evidence, not assertion. |
| Learn | Capture what the build taught, so the next one is better. |
The flow is a default, not a rigid march. The build composes the shortest effective path for the work at hand, rather than forcing every step every time.
The bar: tight and world-class
Every Skill is crafted to the standard of an elite operator and kept tight. No padding, no decorative complexity. Every line earns its place.
The aim is a Skill that feels inevitable: you read it once and operate from it. And because Skills are written down, not baked into a model, they can be inspected, refined, and improved over time.
This is how the whole ecosystem compounds. Each new Skill makes the next build easier and the whole system sharper.