k77040062 / data-janitor-evaleval
Measures whether an agent using the data-janitor skill cleans CSV files safely: profiling first, planning before changing, preserving the original, protecting identifiers, never guessing ambiguous dates, and delivering a complete audit trail.
csvdata-cleaningdata-qualityevaluationaudit-trail
0
manifest.yaml719 B
version: "1.0.0" metadata: name: "DataJanitor Eval - Safe CSV Cleaning" id: data-janitor-eval description: "Measures whether an agent using the data-janitor skill cleans CSV files safely: profiling first, planning before changing, preserving the original, protecting identifiers, never guessing ambiguous dates, and delivering a complete audit trail." author: "Kartik Ranjan Singh" license: MIT tags: [csv, data-cleaning, data-quality, evaluation, audit-trail] dataset: format: jsonl file_path: ./dataset.jsonl features: input: "input" expected_output: "must_contain" evaluation_type: method: llm_as_a_judge rubrics_directory: rubrics metrics: - name: pass_rate threshold: 0.8