- Home
- Skills
- Data & Databases
- Schema Change Review: Approve or block with lineage risk
More screenshots
Works with the AI Agents you already use
Schema Change Review: Approve or block with lineage risk
A column rename looks harmless until the downstream tables and dashboards break. Turn your lineage facts into a risk score and an APPROVE / CONDITIONAL / BLOCK decision, with the conditions to meet before merge.
$12
Schema Change Review: Approve or block with lineage risk
Example session with this skill installed
Review renaming orders.total to gross_total in warehouse.orders. From the package folder, run python3 -I -B scripts/score_change.py references/fail.json --format markdown. The sample assumes 12 downstream assets, 3 critical consumers, 5,000 weekly queries, one incident, 75% coverage, rollback readiness, 14 deprecation days and two owners.
- Read your context and instructions
- Compiled the schema change review
LineageGuard decision: BLOCK
Entity: warehouse.orders
Change: rename_column
Risk score: 72/100
Blast radius
- Downstream assets: 12
- Critical assets: 3
- Weekly queries: 5000
- Open quality incidents: 1
Score components
- base risk: rename_column: +28
- critical consumers: +24
- downstream breadth: +20
- weekly usage: +14.8
- open quality incidents: +4
- test coverage mitigation: -7.5
- rollback-ready mitigation: -8
- deprecation-window mitigation: -3.73
Required conditions
- Reach at least 80% coverage for affected contracts and consumers.
- Validate every critical consumer before production rollout.
The script exits with 1, so a CI gate would fail here. Next steps: check the three critical consumers, get coverage from 75% to 80%, and agree a compatibility window and rollback plan with the owners. The sample facts are made up for the demo.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem You want to rename a column, and the catalog says 12 downstream assets read it, 3 of them critical. Your reviewer wants to know who breaks, what has to be tested first and how you'd roll back. The SQL diff can't tell them any of that, and every team answers it a little differently.
See it before you buy: real input, real output Command (from the package folder, exit code 1, 0.025 s):
python3 -I -B scripts/score_change.py references/fail.json --format markdown
Before: The change and the lineage facts you supply (excerpt of references/fail.json)
"entity": "warehouse.orders",
"kind": "rename_column",
"field": "total",
"new_field": "gross_total",
"downstream_assets": 12,
"critical_assets": 3,
"weekly_queries": 5000,
"test_coverage": 0.75,
"rollback_ready": true,
"owners_identified": 2
After: lines copied from the script's output (excerpt):
# LineageGuard decision: BLOCK
**Risk score:** 72/100
- base risk: rename_column: +28
- critical consumers: +24
## Required conditions
What the numbers show
- Ran on 5 bundled samples and 3 made-up diff scenarios; all 7 valid scores matched hand arithmetic against the written formula.
- A broken input (coverage set above 100%) is refused with exit 2 instead of a crash.
- Honest limit: no production lineage was used. This shows the script follows its policy, not that the policy fits your warehouse.
What you get
- A small Python script that scores 8 kinds of change, from adding a nullable column to dropping one.
- A short JSON format for the facts that matter: downstream and critical consumers, weekly queries, open incidents, owners, test coverage, rollback readiness and deprecation window.
- A score from 0 to 100, a decision (APPROVE, CONDITIONAL or BLOCK) and the list of points that added up to it, so nobody has to trust a black box.
- Reports in JSON or Markdown, read from a file or stdin.
- Instructions that tell your agent how to collect the facts, explain the result, and suggest consumer tests, rollout steps and a rollback plan.
- 22 checks you can re-run, 7 saved sample reports, and 3 example SQL diffs with the facts filled in.
Measured on real data (full detail) I don't have production lineage to test on, so be clear on what this is: I ran the script on the 5 bundled sample files and 3 made-up diff scenarios, then checked every score by hand against the written formula.
| Sample | Change | Score | Decision |
|---|---|---|---|
| pass | add nullable column | 0 | APPROVE |
| fail | rename column | 72 | BLOCK |
| conditional | drop column | 35 | CONDITIONAL |
| unknown | most facts missing | 54 | CONDITIONAL |
| bad-input | coverage set above 100% | — | rejected (exit 2) |
| crm-nullable-country | add nullable country code | 0 | APPROVE |
| payments-narrow-amount | DECIMAL(18,4) to (10,2) | 61 | BLOCK |
| events-drop-device | drop device_type | 39 | CONDITIONAL |
All 7 valid scores matched my hand arithmetic, and the broken file was refused with a clear message instead of a crash. Each run took well under a tenth of a second, Python startup included. This shows the script does what its policy says. It doesn't show the policy is right for your warehouse.
Known limitations
- Someone has to fill in the facts. The script doesn't read SQL and doesn't discover lineage on its own.
- The score comes from fixed rules. It isn't a probability of breakage, and APPROVE doesn't prove consumers will cope.
- Missing facts add penalty points, but a small change can still get APPROVE. The agent is told to call out the gaps.
- The listed conditions are a starting point, not a full migration plan.
- JSON output repeats whatever notes you put in. Redact anything sensitive, or use Markdown, which leaves them out.
- I haven't tested it against a live DataHub instance or real production data.
Who it's for Data and analytics engineers who can get lineage and usage numbers for a table and want the same review every time someone changes a schema.
Requires Python 3.8 or newer and one small JSON file per change. Your agent needs to read files and run Python. Pulling facts from DataHub is optional and needs your own read-only connection.
The script makes no network calls and writes no files. It prints its report and that's it.
Price and guarantee $12 one time, yours to use with any agent that can run Python. Covered by Agensi's money-back guarantee if it doesn't do what this page promises.
Scores are only as good as the evidence in the manifest; missing production signals are scored as risk, not ignored. The skill reviews and plans — it does not execute migrations. DataHub integration requires your own DataHub MCP access.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 15 days ago
Requires
Permissions
Uses file reading and a terminal to run Python. Reads only the supplied JSON file or stdin and prints a report; the scorer writes no files. No network access, no credentials, and nothing is sent anywhere by local scoring. Optional catalog collection is separate and requires an authorized read-only connection.