Repository: twentyhq/twenty — YC-backed open-source CRM alternative to Salesforce

The problem: A developer TODO lived inside Twenty's frontend — unsafe index typings on SettingsDataModelFieldEditFormValues, plus a strict TypeScript enum incompatibility between the string union SettingsFieldType and its underlying GraphQL counterpart FieldMetadataType.

What I shipped:

• ✦ Introduced SettingsDataModelFieldEditFormValues as a proper discriminated union covering every field variant (text, number, select, relation, currency, etc.)
• ✦ Resolved the SettingsFieldType → FieldMetadataType enum casting incompatibility under strict: true
• ✦ Achieved 0-error local compilation from a previously type-unsafe module

// Before (unsafe) ↓
type SettingsDataModelFieldEditFormValues = Record<string, unknown>;

// After — PR #21060 ↓
type SettingsDataModelFieldEditFormValues =
  | SettingsDataModelFieldTextFormValues
  | SettingsDataModelFieldNumberFormValues
  | SettingsDataModelFieldSelectFormValues
  | SettingsDataModelFieldRelationFormValues;
//  ↑ Exhaustive. Discriminated. Zero escape hatches.

Repository: nhost/nhost — Open-source Firebase alternative

The problem: Multiple inconsistent spinner implementations across the dashboard caused UI flicker on fast network responses and non-uniform sizing/styling across the onboarding flow and project pages.

What I shipped:

• ✦ Upgraded core <Spinner /> with size prop (sm | md | lg), configurable className overrides, and a delay prop to suppress flicker on sub-200ms responses
• ✦ Migrated all dashboard loader instances (onboarding + project pages) to the new unified <Spinner /> standard
• ✦ Single source of truth — eliminated every ad-hoc loader implementation

// PR #4391 — New Spinner API
<Spinner size="md" delay={150} className="text-purple-500" />
//                 ↑ delay prevents flicker on fast connections

Repository: formbricks/formbricks — Open-source product survey & analytics suite

The problem: A cluster of critical self-hosted Docker deployment failures affecting Postgres 18 and Node.js healthchecks in production compose stacks.

# The silent killer — Node.js healthcheck on IPv6-first stacks:

# Before (fails silently):
CMD node -e "require('http').get('http://localhost:4000/health', ...)"

# After (explicit IPv4 bind):
CMD node -e "require('http').get('http://127.0.0.1:4000/health', ...)"