id = "data-pipeline"
name = "Data Analysis & Transform"
description = "Clean, transform, analyse, and summarise raw data. Handles messy inputs: CSV, JSON, tables, or free-form text."
category = "engineering"
tags = ["data", "analysis", "transform", "etl"]

[i18n.zh]
name = "数据分析与转换"
description = "清洗、转换、分析并摘要原始数据，支持 CSV、JSON、表格或自由文本等杂乱输入。"

[[parameters]]
name = "raw_data"
description = "The raw data to process (paste CSV, JSON, table, or any structured text)"
param_type = "string"
required = true

[[parameters]]
name = "output_format"
description = "Desired output format for the transformed data (json, csv, markdown-table)"
param_type = "string"
required = false
default = "json"

[[parameters]]
name = "analysis_goal"
description = "What you want to understand or extract from the data"
param_type = "string"
required = false
default = "general summary and key insights"

[[steps]]
name = "profile"
prompt_template = """
You are a data engineer. Profile the following raw data:
1. Detected format and structure
2. Number of rows and columns/fields
3. Data types per field
4. Missing value count per field
5. Obvious data quality issues (duplicates, inconsistent formatting, out-of-range values)
6. Sample of first 5 rows for reference

Raw data:
{{raw_data}}
"""

[[steps]]
name = "clean_transform"
prompt_template = """
Clean and transform the raw data based on the profile below. Apply:
- Trim whitespace from string fields
- Normalise dates to ISO-8601
- Remove exact duplicate rows
- Convert empty strings to null
- Standardise inconsistent categorical values (e.g. 'Y'/'Yes'/'yes' → true)
- Flag but preserve rows with suspicious values (do not silently drop them)

Output the cleaned data in {{output_format}} format, followed by a change log listing every transformation applied.

Data profile:
{{profile}}

Raw data:
{{raw_data}}
"""
depends_on = ["profile"]

[[steps]]
name = "analyse"
prompt_template = """
Analyse the cleaned data to address the following goal: {{analysis_goal}}

Provide:
1. Key statistics (counts, totals, averages, distributions as appropriate)
2. Notable patterns or trends
3. Outliers or anomalies worth investigating
4. Correlations between fields (if applicable)
5. Top 3 actionable insights from the data

Cleaned data:
{{clean_transform}}
"""
depends_on = ["clean_transform"]