Skip to main content
This agent automatically reviews pull requests to detect breaking changes in your dbt models before they reach production. When a PR modifies a model, the agent compares the old and new versions, identifies removed columns or type changes, searches for downstream dependencies, and posts a detailed impact assessment directly on the PR.

​
Simple Example

name: breaking-change-detector
description: Catch breaking changes before they reach production

triggers:
  - type: pull_request
    on_changed_files: "models/**/*.{sql,yml}"

tools:
  preset: safe  # Read-only, can't commit or push changes

prompt: |
  Check for breaking changes:
  1. Compare old vs new: any columns removed?
  2. Find downstream models that use those columns
  3. Comment on PR with impact assessment
  
  Mark as "changes requested" if HIGH risk found.

​
More Robust Example

Production-ready with full impact analysis: 
name: breaking-change-detector
description: Detect and assess impact of breaking changes in PRs

triggers:
  - type: pull_request
    on_changed_files: "models/**/*.{sql,yml}"
    branches: ["main", "staging", "production"]

tools:
  preset: safe  # Read-only, can't commit or push changes

prompt: |
  Analyze this PR for breaking changes to data contracts.
  
  ## 1. Detect changes
  For each modified model:
  
  a) Columns removed or renamed?
     - Compare old vs new SELECT statements
     - Compare YAML column lists
     - Note any missing columns
  
  b) Data types changed?
     - Look for modified CAST statements
     - Check YAML type specifications
     - Flag: VARCHAR → INT, DECIMAL → INT, wider → narrower
  
  c) Filters added/changed?
     - New WHERE clauses (removes rows)
     - Modified JOIN conditions
  
  d) Grain changed?
     - New/removed GROUP BY columns
     - New DISTINCT clauses
  
  ## 2. Assess downstream impact
  For each breaking change:
  
  a) Find references:
     - Search for ref('model_name') in all models
     - Check which columns they select
     - Look for direct column references
  
  b) Calculate risk:
     - HIGH: Removed column + found usage downstream
     - MEDIUM: Type change + usage in calculations
     - LOW: Column added (non-breaking)
  
  ## 3. Generate PR comment
  
  Structure:
  Breaking Changes Detected
  
  HIGH RISK:
  • Model: stg_customers
    - Removed: email column
    - Impact:
      ✗ customers (line 45: SELECT email)
      ✗ user_360 (line 89: email_domain calculation)
      ✗ orders_enriched (line 23: LEFT JOIN on email)
    - Action: Update downstream models OR keep column with deprecation
  
  MEDIUM RISK:
  • Model: stg_orders
    - Changed: order_id type VARCHAR → INT
    - Impact:
      ⚠ order_metrics (line 67: CONCAT with order_id)
    - Action: Verify string operations still work
  
  ## 3. Set PR status
  - HIGH risk: Set to "changes requested"
  - MEDIUM risk: Set to "needs review"  
  - LOW/none: Set to "success"
  
  ## 4. Alert if HIGH risk
  Send Slack to #data-pipeline:
  "PR #{{pr.number}} has HIGH RISK breaking changes.
   Review required: {{pr.url}}"