Command Editing Is Becoming a Real Workspace
Commands now feel much closer to a proper editing environment instead of a one-off form. Humm ships a chat-based command editor with approval and apply flows, command history restore, manual editing for permitted users, better diff rendering, and a searchable command list that holds up better once a workspace has a lot of saved commands.
The workflow is also less fragile than it was a week ago. We fixed persistence and approval regressions in the command editor, improved fallback behavior when the model returns empty output, cleaned up list scrolling and search, and documented the updated command flow in customer docs so the shipped UI now has matching guidance.
Permissions and Admin Workflows Are More Predictable
A set of changes this week made shared workspaces easier to manage. RBAC is now the authoritative source for permissions, organization membership writes stay in sync with RBAC roles, schedule owners can add subscribers directly, and the Members page behaves more consistently when you switch organizations or create new users from a filtered view.
That should translate into fewer cases where the UI and backend disagree about who can do what. It also reduces the amount of state that could previously drift between membership records and actual access control.
Improvements
- Chat input now shows a context-capacity ring so users can see how close a conversation is to its summarization threshold.
Bug Fixes
- Fixed scheduled command admin routes so they request the correct API audience and stop failing on missing access tokens.
- Fixed stale scheduled-run blockers so old in-progress rows stop preventing legitimate retries.
- Reduced unnecessary OAuth credential read lock contention, which should make credential reads and refreshes less likely to stall each other.
- Hardened data-sync OAuth and retry routes with better session requirements, null-body handling, and regression coverage.