coding.sh dataset

coding.sh dataset <action>

  Manage versioned datasets of cases

Commands
  add           Add cases to a dataset (creates it when missing)
  import        Import cases from a JSONL, JSON or CSV file
  list          List datasets

coding.sh dataset add <NAME> [options]

  Add cases to a dataset (creates it when missing)

Arguments
  name          Dataset name

Options
  --case <JSON>    Case as JSON: {"input":…,"expected":…} (repeatable)
  --session <ID>   Case from a captured session (repeatable)
  --split <SPLIT>  Split for the added cases (default: stable hash of the case id, 60/20/20)
  --kind <KIND>    Kind for the added cases: bad fails before the fix, guard and control pass before and after
  --note <TEXT>    Version note

Examples
  coding.sh dataset add smoke --case '{"input":"2+2","expected":"4"}'
  coding.sh dataset add labelled --session <ID> --split test
  coding.sh dataset add prompt-fix --kind control --case '{"input":"What is 2+2?","expected":"4"}'

coding.sh dataset import <FILE> [options]

  Import cases from a JSONL, JSON or CSV file

Arguments
  file          Path to the file

Options
  --name <NAME>  Dataset name (default: file name)
  --note <TEXT>  Version note

coding.sh dataset list

  List datasets