The dashboard calls the muse gateway route Meta Muse Code (it was "Meta Muse"): Muse Code subscriptions are the supported target, not the plain Muse API. The route id, stored usage and org model rules are unchanged.
hnl gateway connect omp no longer points omp's plain Meta Muse API provider (providers.meta) at the gateway; it wires muse-code only. A meta entry an earlier release wrote is left as it is, and hnl gateway disconnect omp still restores or removes it.
The Telegram link tables are dropped (control migration 0055, ADR-82) and account deletion and identity merge no longer touch them. Memories banked from Telegram keep their label.
bun run deploy:prod refuses to apply a pending migration marked -- hnl:destructive until HNL_CONFIRM_DESTRUCTIVE_MIGRATIONS names it, and says how to back up first. Migration 0055 carries the marker.
The weekly-digest playbook skill describes what the digest sends today (stats and the weekly retrospective) instead of an AI summary and a monthly ask quota, which ADR-80 removed. docs/CLOUD_SETUP.md drops the AI Gateway section, and the production Worker no longer declares the unused AI binding. Superseded search and judge ADRs (27, 28, 36, 45, 78) carry a status line pointing at ADR-57 and ADR-80.

The cloud e2e suite (cd cloud && bun run test:e2e) passes again (22 of 22; it was 13-15). The sync driver's gateway-usage fixture lacked cacheWrite1h, so gateway_usage refused the row and five sync and read-plane tests failed after it (cloud/e2e is not type-checked, which is how it drifted). On a machine with a HarnessLink managed enrollment (ADR-74), every throwaway e2e org's policy was correctly refused as another org's, failing the policy-push tests; each fresh e2e machine now gets its own empty system root (HNL_MANAGED_SYSTEM_ROOT, the seam unit tests already use).

Flag notifications