The list item tooltip taught multi-select on every hover and sat over
the sidebar New Note button. Drop it; selection still works without the
hint, and action-button tooltips stay.
Drive light and dark from one token table so Tailwind utilities follow
:root.dark assignments. Drop the leftover dark: color variants and the
[class~="…"] inversion layer, and keep PDF paper as a white surface.
Introduce DIAGRAM_READABLE_MIN_SCALE and use it after auto layout instead of shrinking the whole graph. Architecture diagrams pin to the top of the canvas so the extra headroom is gone.
Add a bulk Markdown ZIP action for selected notes, reuse the existing
readable export layout, and keep the desktop bulk action card visible
when a diagram note is open.
Lay out each architecture boundary independently, place ungrouped terminals in the group-level graph, stack groups when a left-to-right row would overflow, and route architecture edges orthogonally.
Phone viewers no longer shrink a whole diagram into an unreadable overview. The canvas fills the remaining viewport, pinch-zooms, and focuses a topic on tap, without zoom chrome.
Sidecar memo.list ignored the tag parameter, so choosing a tag showed every note. Match json_each tag values case-insensitively and keep prefix-overlapping names out of the result.
Official plugins already update silently in the background. Drop the blocking "got it" notice so a successful auto-update does not steal focus from the workspace.
NAS GUI upgrades that drop container environment variables could leave
login working while AI and object-storage credentials could no longer be
encrypted. Persist the encryption root on /data, generate one when an
existing instance has no secret, and stop the add-service form from
looking fillable when encryption is still unavailable.
Fixes#370
First-party canvases, including public share, parse diagram IR and draw
with read-only X6. Embedded Mermaid fences in ordinary notes stay on
Mermaid. Native viewers no longer inject a hidden Mermaid document when
the IR is valid.
Light component cards mix about 12% accent into white so types stay
readable without turning into saturated chips. Icon wells stay slightly
stronger than the fill; strokes and dark surfaces are unchanged.
Same-type flowchart nodes share a fixed size, sequential links stay
vertical after auto-layout, and tall flows open at reading size instead
of shrinking to fit. Architecture node paint moves into a shared style
module so X6 and Mermaid stay on one palette.
Mind maps pick a whole-map structure (balanced maps and right-growing
logic charts) from thumbnail cards, and a named 6-color palette from
vivid/classic strips. Flowchart notes keep a dedicated paper surface.
Mind maps now pick structure (underline map vs boxed topics)
independently from color (forest green vs spectrum). The editor
sizes the paper to every node so two-sided branches are not
cropped while editing.
Give mind maps a shared three-level appearance across Web, mobile, and
iOS: capsule roots, stronger first-level topics, lighter nested nodes,
and side-anchored tapered branches. Keep flowchart and architecture
edges unfilled so SVG does not paint them black.
tianma-if