name: sql-debug-env
version: 1.0.0
description: >
  SQL Debug & Data Pipeline Repair — an OpenEnv environment where an AI agent
  diagnoses and fixes broken SQL queries and ETL pipelines executed against a
  live DuckDB instance. Seven tasks ranging from easy (syntax fix) to expert
  (chaos engineering).

author: sql-debug-env
tags:
  - openenv
  - sql
  - data-engineering
  - debugging
  - rl
entrypoint: uvicorn app:app --host 0.0.0.0 --port 7860
inference: python inference.py

tasks:
  - id: task_1_easy
    difficulty: easy
    max_steps: 5
    description: >
      Fix a SQL SELECT query with a missing comma between column names.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

  - id: task_2_medium
    difficulty: medium
    max_steps: 5
    description: >
      Fix a GROUP BY aggregation query — add GROUP BY clause.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

  - id: task_3_hard
    difficulty: hard
    max_steps: 5
    description: >
      Fix a RANK() window function missing PARTITION BY department.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

  - id: task_4_expert
    difficulty: expert
    max_steps: 5
    description: >
      Fix an invalid date literal (month 13) inside a CTE.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

  - id: task_5_optimization
    difficulty: expert
    max_steps: 5
    description: >
      Rewrite a CROSS JOIN query to use INNER JOIN.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

  - id: task_6_migration
    difficulty: expert
    max_steps: 5
    description: >
      Migrate denormalized table to 3NF schema safely.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

  - id: task_7_chaos
    difficulty: expert
    max_steps: 5
    description: >
      Fix a live ETL pipeline with duplicate entries and NULL emails.
    grader: SQLGrader
    grading_metric: accuracy
    baseline_score: 0.5

observation_schema:
  task_id: string
  description: string
  broken_sql: string
  schema_info: object
  error_hint: string

action_schema:
  fixed_sql: string
  explanation: string

endpoints:
  health: GET /health
  reset: POST /reset
  step: POST /step
  state: GET /state
  tasks: GET /tasks
  grader: POST /grader
  baseline: GET /baseline
  docs: GET /docs