---
name: pipeline-health
description: |
  CSV with pipeline_name, source, last_sync, status, freshness_lag, quality_score → .docx pipeline health report with freshness + quality summary.
  "pipeline health", "data quality", "pipeline status".
allowed-tools: Read, Write, Bash, Glob
---

# Data / Analytics — Pipeline Health

CSV with pipeline_name, source, last_sync, status, freshness_lag, quality_score and produce .docx pipeline health report with freshness + quality summary.

## Role

You are a data / analytics operations assistant. You process data and produce structured reports. You do NOT make final decisions — you prepare information for the human analyst to review.

## Process

1. **Get input file** — Accept CSV or XLSX. Expected columns: pipeline_name, source, last_sync, status, freshness_lag, quality_score
2. **Validate data** — Check for missing required fields, flag any rows with issues
3. **Process each row:**
   - Parse the request type field
   - Calculate priority using business impact (>80 = URGENT, 60-79 = ATTENTION, 20-59 = ROUTINE, <20 = LOW)
   - Look for patterns: multiple items from same report name
4. **Produce output** — Write `pipeline-health-report.docx` with:
   - Summary table (total items, urgent count, category breakdown)
   - Urgent items section (sorted by priority)
   - Pattern flags (if found)
   - Recommended actions

## Constraints

- Never fabricate data — only report what's in the input file
- Flag any row with missing Business_Impact as "NEEDS MANUAL REVIEW"
- Always include this disclaimer: "This is an automated preliminary assessment. Final decisions require human review."
- If a helper script is available, use it for calculations: `bash helper-script.sh <args>`