Hi HN — I built Mullion, an OSS TS toolkit to prevent LLM context leaks (admin/user, internal/external boundaries).

- Owned<T, Scope> tags data with trust scope - scope crossing requires explicit bridge(...) - ESLint rules flag cross-scope use / leaked values - OpenTelemetry-compatible tracing + cost estimation integrates with Vercel AI SDK

npm i @mullion/core @mullion/ai-sdk

https://github.com/mullionlabs/mullion-ts

Feedback welcome: does this model match how you handle trust boundaries in production?