Fundamentals
The concepts behind a taskset — what RL data looks like, how Datagen Factory authors a pack, and how to drive it from Claude Code.
Read this section when you want to understand the product. If you want a fast path to a working taskset instead, skip to Start.
What you need to know
Task
Every file in a Harbor task directory, section by section of task.toml.
Rubrics
What makes a rubric training-grade, the criterion fields, and how a rubric compiles into the verifier.
Brief
The plain-language description you write to kick off a taskset.
The authoring flow
Brief to preview to delivery, end to end. Where iteration happens and where it doesn't.
Claude Code workflow
Pointing a coding agent at the CLI. Why the CLI is your docs.
Resources and sandboxes
How a database or file corpus is wired into per-rollout copies of your real state.
Dataset formats
What you receive — JSONL, Parquet, the quality report, and the optional HuggingFace mirror.
Where to go next
For the command surface, see the CLI reference. For the browser experience, see the UI reference.