SQL_debug_env_v1 / openenv.yaml
sai1912's picture
Upload folder using huggingface_hub
078e08b verified
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