Brief
The plain-language description you write to kick off a dataset.
A Brief is plain English describing your agent and the work it does. The system reads it and turns it into a three-sample preview.
What to include
- What your agent does in production. The tools it uses, the problems it solves, the kind of input it gets.
- What the training data should teach it. The capability you're sharpening, expanding, or filling in.
- The shape of the work, if you know it. A typical input, a desired output format, a grading criterion that matters to you.
You don't need to specify the rubric or the task structure. We build those.
Example
I'm training a research agent that answers questions about SEC 10-K filings. Each task should pose a multi-step question where the answer requires reading at least two sections of a filing and synthesizing across them. Grading should check that the agent cites the section it pulled each fact from, not just the final answer.
Next
You don't have to get the Brief right on the first pass. Gaps surface in the preview, and feedback drives revisions.