Skip to content

1.7.0-beta.5

Pre-release
Pre-release

Choose a tag to compare

@TiveCS TiveCS released this 08 Jul 00:26
· 0 commits to main since this release

Krust Studio 1.7.0-beta.5

Follow-up to the AI Schema Sync beta line, from live testing on real projects.
Adds typed Data changesets, gives the AI a way to query big tables, and
fixes several editor/viewer papercuts. Opt in via Settings → Updates.

Changesets — now Schema and Data (ADR-0023)

  • Data changesets. INSERT / UPDATE / DELETE can now be grouped into a
    Data changeset and exported as a .sql for the DevOps handoff — the same
    capture-and-hand-off workflow you already had for schema DDL, applied to
    seed/reference-data changes.
  • Cleanly separated by kind. A changeset is Schema (DDL) or Data
    (DML); the two never mix in storage. The History rail groups them into a
    Schema and a Data section, each with its own active changeset and its
    own Unassigned inbox; "Move to changeset" only offers same-kind targets.
  • Two optional active slots. Set a Schema changeset and/or a Data changeset
    active independently; matching captures auto-attach only when a slot is active
    (nothing auto-collects until you say so). Destructive DML (TRUNCATE,
    no-WHERE DELETE/UPDATE) never auto-attaches.
  • Export together. Pick a Schema + a Data changeset and export one .sql
    with the statements interleaved in execution order — for when a data
    backfill needs to sit between two DDL steps.

AI / MCP

  • read_rows can filter now. The read tool takes structured
    filter / orderBy / columns (no raw SQL), so an agent can target rows in a
    large table instead of paging blind through it. Any column used in a
    filter/sort/projection must be allowlisted and unmasked — a masked column can't
    be probed indirectly.

Fixes & polish

  • Changeset actions are visible. The per-changeset menu (set active, rename,
    export, delete) was hover-only and easy to miss — it's now persistently shown.
  • Copy SQL by selecting it. Read-only SQL displays (history detail, DDL
    preview, routine viewer) are now mouse-selectable with a visible highlight, so
    you can drag-select and copy without hunting for the Copy button.
  • Smoother tab scroll. Vertical mouse-wheel over the tab strip now glides
    horizontally instead of jumping a notch at a time.
  • JSON viewer FK follows the selection. Expanding a foreign key in the row
    JSON panel now re-fetches when you pick another row (it used to stick on the
    previously-expanded row's parent). The panel's resize handle is also easier to
    grab.

Verification still needed

The Data-changeset and read_rows-filter paths build clean but want a real run:
a live agent reading a filtered big table, DML auto-attaching to an active Data
changeset, and an export-together .sql in the right order. Manual promotion of a
destructive DML entry into a Data changeset is not yet behind a typed confirm —
that lands next. Please report through the 1.7.0-beta tracking issue.