Field rename verification

Field Rename Verification

Wrong skill? General code modification → blast-radius-check. Pre-commit checks → pre-commit-gate. Completion verification → verification-before-completion.

When to Use

• Renaming a field, column, or property across one or more services
• Changing an API contract (request/response shape, query parameters, headers)
• Refactoring a data model that crosses service boundaries
• Splitting or merging fields (e.g., trackingLine → inboundLine/outboundLine)

Why Renames Are Dangerous

Field renames and API contract changes are the #1 source of production incidents from "complete" work that wasn't actually complete.

Core principle: Trace the FULL data flow, not just the obvious touchpoints.

Incident that created this skill: PROJ-XXX (Feb 2026) - Renamed trackingLine to inboundLine/outboundLine across 6 PRs, but missed the telephony client adapter that actually calls voice-service. Quality gates (lint, typecheck, tests) all passed because each service was internally consistent.

The Iron Law

NO FIELD RENAME IS COMPLETE UNTIL YOU VERIFY EVERY PATH:
READ → STORE → PASS TO OTHER SERVICES

If you only traced READ and STORE, you missed PASS.

The Three Paths

Every field has THREE data flow paths that must be traced:

Path	Description	Example
READ	Where the field is read FROM	Config endpoint, database query
STORE	Where the field is written TO	Database insert, cache storage
PASS	Where the field is sent TO ANOTHER SERVICE	HTTP client, message queue, event payload

The PROJ-XXX mistake: Traced READ (settings-service → post-lead.ts) and STORE (lead table), but missed PASS (telephony-client.adapter → voice-service).

Pre-Rename Checklist

BEFORE claiming a field rename is complete:

1. Grep Exhaustively

# Search ALL affected repos for the old field name
grep -rn "oldFieldName" --include="*.ts" --include="*.tsx" --include="*.js" repo1/ repo2/ repo3/

# Also search for the CamelCase, snake_case, and UPPER_CASE variants
grep -rn "old_field_name\|OLD_FIELD_NAME\|OldFieldName" --include="*.ts" repo/

2. Trace Every Service Boundary

For each service that handles this field:

Question	If Yes → Check
Does this service call another service?	HTTP clients, SDK calls, API adapters
Does this service emit events?	Event payloads, message queue producers
Does this service store data?	Database schemas, cache keys
Does this service expose an API?	Request/response types, OpenAPI specs
Does this service have NPM packages?	Published type definitions

3. Verify Type Definitions

# Check if old types are still published
grep -rn "oldFieldName" node_modules/@yourorg/*/dist/*.d.ts

If old types exist in node_modules, consumers may compile against wrong contract.

4. Integration Test the Full Path

Unit tests pass but integration fails because:

• Service A sends oldFieldName
• Service B expects newFieldName
• Both services' unit tests pass (internally consistent)
• E2E test fails (contract mismatch)

REQUIRED: Run at least one test that crosses the service boundary you changed.

Common Failure Patterns

Pattern	What Gets Missed	Prevention
"I updated the route handler"	HTTP client that calls the route	Trace callers, not just callees
"I updated the type definition"	Runtime code still uses old name	Grep before commit
"Tests pass"	Tests mock the boundary, don't test it	Add integration test
"Schema is updated"	Transform/adapter layer unchanged	Check every layer
"Config reads correctly"	Config is passed to another service wrong	Trace PASS path

Service Boundary Verification

For cross-service field renames, verify at BOTH ends:

Service A (sender)        Service B (receiver)
─────────────────         ────────────────────
HTTP client sends    →    Route handler expects
message.fieldName         body.fieldName
     ↓                         ↓
   MUST                      MUST
   MATCH                     MATCH

PROJ-XXX failure:

• voice-service (receiver) expected inboundLine/outboundLine ✅
• acquisition-service (sender) still sent trackingLine ❌

Red Flags - STOP

If you find yourself thinking:

• "The database column uses the old name, that's fine" → Maybe, but verify callers
• "I just need to update the schema" → No, trace the full flow
• "The tests pass so it's done" → Tests may not cross the boundary
• "I updated the obvious files" → The non-obvious ones break production
• "It's just a field rename" → Field renames touch everything

The Gate Function

BEFORE claiming a field rename is complete:

1. GREP: Run exhaustive search for old field name in ALL affected repos
2. LIST: Enumerate every service boundary the field crosses
3. TRACE: For each boundary, verify BOTH sender and receiver
4. TEST: Run at least one integration test crossing each boundary
5. ONLY THEN: Claim the rename is complete

Skip any step = incomplete work

Companion Skills

• verification-before-completion — General verification discipline
• systematic-debugging — When the field mismatch causes runtime errors
• link-verification — For verifying API endpoint URLs still exist

Failure Modes

Failure	Recovery
Missing a consumer across service boundary	Search ALL repos, not just current. Check API clients, shared libs.
Renaming in code but not in config/env files	Check .env files, docker-compose, CI configs, deployment manifests
Database column rename without migration	Column renames need a migration. Check ORM models match DB schema.