rnow-train-jsonl
Installation
SKILL.md
train.jsonl Format
One JSON object per line. Each entry is a training example.
Fields
| Field | Required | Description |
|---|---|---|
messages |
Yes | Conversation array |
rewards |
RL only | List of reward function names |
metadata |
No | Data accessible via args.metadata in rewards |
variables |
No | Template variables via args.variables |
tools |
No | Filter which tools are available for this entry |
docker |
If sandbox | Docker image for sandbox execution |
docker_env |
No | Environment variables for sandbox |