Skip to content

fix(app): the sign-in, the mailboxes page, the top bar, the list, the search and the open message use Kumo's controls, keeping their look - #98

Open
Madraka wants to merge 6 commits into
mainfrom
fix/kumo-first-screens
Open

Madraka wants to merge 6 commits into
mainfrom
fix/kumo-first-screens

Conversation

@Madraka

@Madraka Madraka commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor

Part of #82: six screens, one commit each, each lowering its own debt in the controls check. These are the screens a person meets first, as the order on #82 puts them. A move to Kumo keeps what a screen looks like, compared with main; where Kumo cannot give the shape, the control stays as it was, with its reason.

  • TO_MOVE goes from 43 places to 26, the counts in its entries summed on main and on this head.
  • SIZED_BY_HAND goes from 6 to 3.
  • ON_PURPOSE goes from 1 line to 9.
  • A new list, ROUND_ON_PURPOSE, has 2 lines.

The commits

  1. The sign-in's second step (048228f). "Use a recovery code instead" (and back) is Kumo's Button, the ghost kind, across the form's width under Continue.
  2. The mailboxes page (8957b2b). Sign out is Kumo's Button at the small size, after the address and the "Sign-in and security" link.
  3. The top bar (0aee97a).
    • The search box stays as on main, one ON_PURPOSE line naming what Kumo lacks.
    • Its Clear search and Search options buttons are Kumo's: ghost, round, small, 26 px.
    • The rail's shade on a phone stays native, one ON_PURPOSE line.
    • The check now holds InputGroup and its parts to Kumo's sizes.
  4. The list (dd9e343).
    • The box above the list and a line's box are Kumo's Checkbox.
    • The star is Kumo's Button: ghost, square, small, 26 × 26.
    • Shift still ticks a range.
    • The box above the list keeps its 12 px click area off the menu's button beside it.
    • A line stays native, one ON_PURPOSE line.
  5. The search (21041e7).
    • The quick filters are Kumo's Button at the small size, 26 px, and stay fully round.
    • The controls check gains ROUND_ON_PURPOSE: a Kumo control made fully round with rounded-full, and nothing else written by hand, at a listed place with its reason and exact count.
      • Padding beside the corner, another corner, a breakpoint, the important mark or a style attribute are still refused there.
      • Its tests say so; three of them fail when the exception is taken out.
    • docs/status.md and the podbay-ui skill name the list. The skill also says a move to Kumo keeps what a screen looks like.
  6. The open message (99d633d).
    • Reply, Reply all and Forward are Kumo's and stay fully round, one ROUND_ON_PURPOSE line.
    • "Unsubscribe" is Kumo's ghost button at the extra small size, grey and underlined. Its margins give back its extra 4 px of height and its 6 px of padding, so the head keeps its size and the word stands 8 px after the name.
    • The head stays as on main: the picture, the name and the line of who it went to. Each is one ON_PURPOSE line saying what Kumo's Button cannot be there:
      • the picture's own colours: Kumo's Button sets its text colour with an important mark and its own background under the pointer;
      • a line of text: Kumo's Button is a box of 20 px or more with its own padding.
    • The name's line wraps. With main's layout, a mailing list's long sender beside Unsubscribe was drawn 0 px wide on a phone, and the browser test found it hidden. Unsubscribe now moves under the name when both do not fit.
    • test/e2e/reading.spec.ts checks four things at 375 px: the name is cut, it is shown, it keeps its line above Unsubscribe, and it ends before the date. It fails without the wrap.
    • The line of a folded message and the line for the middle of a long conversation stay native, two ON_PURPOSE lines.

Your findings

  • 1–3 (the name, the star, the search box): fixed at the last head; held by measurement in your note.
  • 4. One's own picture: native again, as on main. The initial is oklch(0.97) on the brand colour oklch(0.5772 0.2324 260). Under the pointer it keeps both colours and takes a ring.
  • 5. The head: as on main, the same in every message. At 1280 px:
    • the name is 13 px, 181.4–196.6, level with the picture's top at 181;
    • the line under it is 12 px, 197–213;
    • the head is 40 px.
    • One's own message has the same pair: name 13 px at the picture's top, line under 12 px, head 40.
  • 6. The round shapes: the answers and the quick filters are fully round again, through ROUND_ON_PURPOSE.
  • Unsubscribe: ghost, extra small, underlined, 85 × 20.
    • At 1280 px it sits beside the name and the head is 40 px.
    • At 375 px, for a mailing list's message, it moves under the name and the head is 47.3 px. With main's layout the name had no room at all there.

Measurements on this head

Throwaway instance, at 1280 and 375 px.

  • Top bar: the search box is 48 px, top 8, centre 32, fully round. Unfocused it is filled with no line; focused it is white with a line and a ring. Its buttons are 26 × 26, centred at 32.
  • List, every line 40 px, the columns the same on every line:
    • 1280 px: box 273–289 (target 261–301), star 301–327, sender at 335, subject at 539.
    • 375 px: box 16–32 (target 4–44), star 44–70, sender at 78.
  • Search: the quick filters are 26 px tall and fully round.
  • Answers: Reply 81 × 36, Reply all 99 × 36, Forward 98 × 36, all fully round.
  • Heads at 375 px:
    • one's own message: 40 px;
    • a long sender: 40 px, the name 180–195.3 and cut;
    • a mailing list's message: 47.3 px, Unsubscribe under the name.

Checks

  • On 8ebf685, which has the same tree as this head (only the last commit's message differs):
    • pnpm check passed: 22 static checks, 1,017 unit tests.
    • pnpm e2e passed, 92 tests.
  • On this head:
    • node scripts/commits.ts: six commits, no problems.
    • The controls check and its unit tests pass at each of the six commits.

Failed runs on this work before this head:

  • A full pnpm e2e failed one test at the first draft: the box above the list covered the menu's button.
  • pnpm check failed once, on formatting.
  • test/unit/controls.test.ts failed one test while the new list was being placed: a test called the check without it. It now passes an empty list.
  • While writing the new browser test, I saw it fail several times:
    • it first ran without signing in;
    • then it measured Kumo's wrapping span instead of the name;
    • then it used a threshold of 48 px I had picked; 41.6 px was the real width;
    • then, with main's head layout, it found the name hidden.
    • I also ran it without each fix, on purpose, and saw it fail.

CI: run 44476080477 on ac6c562 passed, and 44481251919 on d3496fe passed. The run on this head is pending.

Pictures

On the orphan branch pictures, in pr-98/ at 4361d92. All 40 are from a throwaway local instance with made-up mail, each under 80 KB.

1. The sign-in's second step

Light Dark
Desktop, 1280 px
Phone, 375 px

2. The mailboxes page

Light Dark
Desktop, 1280 px
Phone, 375 px

3. The top bar, without focus

Light Dark
Desktop, 1280 px
Phone, 375 px

3. The top bar, with focus and text

Light Dark
Desktop, 1280 px
Phone, 375 px

4. The list, two lines ticked and one starred

Light Dark
Desktop, 1280 px
Phone, 375 px

5. The search, two quick filters on

Light Dark
Desktop, 1280 px
Phone, 375 px

6. A conversation that ends with one's own message

Light Dark
Desktop, 1280 px
Phone, 375 px

6. The same, the pointer on one's own picture

Light Dark
Desktop, 1280 px
Phone, 375 px

6. A mailing list's message, details shown

Light Dark
Desktop, 1280 px
Phone, 375 px

6. A message from a long sender

Light Dark
Desktop, 1280 px
Phone, 375 px

🤖 Generated with Claude Code

Madraka and others added 2 commits November 19, 2026 03:15
…s button

"Use a recovery code instead", and "Use a code from the app instead", is an action, so it is
Kumo's Button: the ghost kind, at the form's own size and across its width, under Continue.
The controls check's debt loses SignInSteps.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
Signing out is an action, so it is Kumo's Button at the small size, after the address and the
link to sign-in and security, instead of a word underlined in the middle of the sentence. The
controls check's debt loses home.tsx.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
@Madraka

Madraka commented Nov 19, 2026

Copy link
Copy Markdown
Contributor Author

audit: not passed yet, at ac6c562.

I opened this branch from the local ref before it was pushed and measured each screen at 1280 px and 375 px. The note is on the issue: #82 (comment)

Before merge:

  1. The sender's name in an open message no longer shortens at phone width (137.3 px button in a 102.9 px column; the name ends at x=181.3 and the date begins at 182.9). On main it was cut with an ellipsis. Wanted: a browser test at phone width with a long sender, and the picture.
  2. The list's star is a 14 × 14 px target (20 × 20 on main), and the box's target ends exactly where the star begins. Wanted: no smaller than before, with the list's columns measured afterwards.

To decide in this pull request: the quick filters' height (26 → 20 px), how Unsubscribe reads as an action, and the reason the sender's picture stays on purpose.

Not opened by me: the sign-in's second step (the example has no two-step account); its picture is wanted here.

I open the fixed screens again before passing.

@Madraka

Madraka commented Nov 19, 2026

Copy link
Copy Markdown
Contributor Author

audit: one more before merge, and it is one I let through in my first note. The project lead saw it in the running app.

  1. The top bar's search box lost its shape. On main it is the widest thing in the bar and reads as the place to search: 48 px tall, fully round, filled (bg-kumo-fill) with no line, turning to the page's colour with a ring when it has the focus. On this branch it is Kumo's InputGroup at its large size: 40 px tall, 8 px corners, the page's colour with a line around it (measured at 1280 px: 672 × 40 at x=248, y=12, in a 64 px bar). It now looks like a form's field.

    The cause is the move itself: InputGroup has sizes and nothing else (no round or filled variant in 2.18.1), and the controls check refuses a height, a corner or padding written on a Kumo control. So the shape could not be carried over, and the skill's line was rewritten to fit the component ("one wide, rounded search box (48px)" became "Kumo's InputGroup at its large size (40px)"). That is the wrong way round: the skill says what the screen is, and the component serves it.

    Wanted: the search box as it is on main (48 px, round, filled, the same focus), the skill's line as it was, and the buttons inside it still Kumo's. If Kumo cannot give that shape, the field stays on purpose with its reason, like the rail's shade, and the reason says what Kumo lacks. A picture of the bar at 1280 px and at 375 px, in both themes, with and without the focus.

A rule for the rest of #82, which I am adding to the issue: a move to Kumo does not change what a screen looks like unless the pull request says so and shows it; where the skill describes a control, the description is not edited to fit the move.

Madraka and others added 2 commits November 19, 2026 10:21
…umo's buttons inside it

The search box stays the field it is: one wide box 48 px tall, fully round and filled, with no
line, that turns to the page's colour with a ring when it has focus. Kumo 2.18.1's Input and
InputGroup give a size alone, 40 px at their largest, with Kumo's corner and line, and a height,
a corner or padding written by hand is refused; so the field is one ON_PURPOSE line that says
what Kumo lacks. Its two buttons, "Clear search" and "Search options", are Kumo's: the ghost
kind, round, at the small size, where they stood.

The shade behind the rail on a phone stays a native button, as one ON_PURPOSE line with its
reason: it is the screen around the rail, out of the tab order, and closes the rail when pressed.

The controls check now holds InputGroup and its parts as it holds Kumo's other controls, so a
size written by hand on one is refused; two cases in its tests, which fail without it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
…n purpose

The box above the list is Kumo's Checkbox, which says "some" with its own indeterminate state
instead of a property set on the element after each paint. A line's box is Kumo's Checkbox too,
and its star Kumo's Button, the ghost kind, square, at the small size: 26 px, no smaller a
target than the 20 px the star had. Shift still ticks every line from the last box used: Kumo's
box takes no click handler, so the line reads Shift from the click or the key that passes
through it on the way to the box.

Kumo's box takes clicks 12 px around itself, so the box above the list keeps that room from
the menu's button beside it, which it covered.

A line itself stays one native button, as one ON_PURPOSE line with its reason: it is the
conversation, with who wrote, the subject, the text, its labels and the date, laid out for the
width of the list, and Kumo's Button holds a label and an icon in one row.

The browser tests find these boxes by role and name.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
@Madraka
Madraka force-pushed the fix/kumo-first-screens branch from ac6c562 to d3496fe Compare November 19, 2026 07:38
Madraka added a commit that referenced this pull request Nov 19, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
@Madraka

Madraka commented Nov 19, 2026

Copy link
Copy Markdown
Contributor Author

audit: not passed, at d3496fe. This time every changed screen was opened on this head and on main (8089522) in the same example, at 1280 px and 375 px, and the two were compared.

Fixed, by measurement

  • The search box is as on main: 672 × 48 at y=8 in the 64 px bar, fully round, filled, no line. Its buttons are 26 × 26 (28 and 24 on main); accepted. The reason in ON_PURPOSE names what Kumo lacks; the skill's line is back.
  • The list's star is 26 × 26 with a 16 px icon; the box's target (261–301) ends where it begins (301). The sender's column moves 6 px to the right against main; accepted.
  • The sender's name is cut at 375 px (text 68–158.9, the date begins at 182.9), nothing scrolls sideways. reading.spec.ts read.
  • The quick filters are 26 px tall again.

Before merge

  1. The picture of one's own message has a dark initial on the brand colour. Kumo's button carries !text-kumo-default, which wins over text-kumo-inverse: the computed colour is oklch(0.205 0 none) on oklch(0.5772 0.2324 260). On main the initial is white. Seen in a conversation of two messages. The button also carries not-disabled:hover:bg-kumo-tint: say what the picture does under the pointer and show it; on main it kept its colour and took a ring.

  2. The head of an open message is taller, and two messages of one conversation no longer match. With the name as Kumo's button at its default size the name is 14 px in a 36 px box:

    1280 px main this head
    The name's text 13 px, begins at y=181.4, level with the picture's top (181) 14 px, begins at y=189
    The line under it y=197–213 y=217–237
    The head 40 px, the picture's height 56 px

    At 375 px a mailing list's message has a head of 82 px (40 on main): the name, Unsubscribe on a row of its own, then the line of recipients. And in one conversation the message of a stranger (name a button: 56 px, 14 px type, the name 8 px under the picture's top) stands above one's own (name a plain text: 40 px, 13 px type, level with the top). Wanted: the name and the line under it as a pair beside the picture, the same in every message, at the size main has.

  3. Two more round shapes went the way the search box had gone. The answers under a conversation (Reply, Reply all, Forward) are fully round on main and have an 8 px corner here. The quick filters are fully round on main and have a 6 px corner here, beside the folder's label in the result's line, which is still round. The rule that brought the search box back holds for these: the shape stays, on purpose and with a reason where Kumo's Button has no such shape. If the check has no way to say "this corner is on purpose" for a Kumo control, propose one in this pull request; do not take the shape away to satisfy the check.

To decide here

  • Unsubscribe is a bordered button 104.9 × 26 beside a name that has no border (underlined 12 px text, 86.6 × 16, on main). It weighs more than the name, and on a phone it costs a row. Kumo's ghost kind at the small size with the underline kept would read as the text action it was; say what you choose and show both widths.

Accepted as changes of look: Sign out as a small button on the mailboxes page (the line is 26 px, was 15.3); the picture's initial at 14 px.

Not opened by me: the sign-in's second step; its pictures are read from the pictures branch next.

From now on I compare each control with main, not only with its neighbours; the first note did not, and that is how findings 3, 5 and 6 got through.

@Madraka

Madraka commented Nov 19, 2026

Copy link
Copy Markdown
Contributor Author

audit, from the pictures on the pictures branch (not opened by me): the sign-in's second step. On main "Use a recovery code instead" is a line of small underlined text under Continue; here it is a bar as wide as Continue, and in both pictures it is drawn filled although its kind is ghost. Say in the description that this look is wanted and why the bar is filled at rest (or show it at rest if the picture caught it under the pointer). The top bar with the focus, in the dark theme, reads well.

On the count of the debt: 43 on main and 26 here, as you wrote; my 46 was wrong.

Madraka and others added 2 commits November 19, 2026 10:47
…pose

Each quick filter above the results is Kumo's Button at the small size, 26 px as the chips
were, which still says whether it is on with aria-pressed: on, it is the primary kind; off, the
outline kind. They stay fully round, as before and as the folder's label in a result's line is.
The controls check's debt loses search-results.tsx.

Kumo's Button is round only as a circle for an icon alone, and the check refused a corner written
by hand on a Kumo control. It now lets one shape through where it is listed: a Kumo control made
fully round with rounded-full and nothing else written by hand, at a place of ROUND_ON_PURPOSE
with its reason, the count exactly as listed. Padding beside the corner, another corner, a
breakpoint, the important mark or a style attribute are still refused there; its tests say so,
and fail without the exception. docs/status.md and the podbay-ui skill name the list, and the
skill says that a move to Kumo keeps what a screen looks like.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
…ns, and its head stays on purpose

Reply, Reply all and Forward under the last message are Kumo's buttons and stay fully round, as
before: one line of ROUND_ON_PURPOSE with its reason. The controls check's SIZED_BY_HAND loses
EmailPanel.

"Unsubscribe" is Kumo's ghost button at the extra small size, grey and underlined, the text
action it was. Its margins give back the 4 px it is taller than the name's line and the 6 px of
its padding, so the head keeps the size it had and the word stands 8 px after the name, as it did.

The head of a message stays as it was: the sender's picture, the sender's name and the line of
who it went to, as a pair of lines 13 and 12 px beside the 40 px picture, the same in every
message of a conversation. Each is one ON_PURPOSE line with what Kumo's Button cannot be there:
the picture's own colours (Kumo's Button sets its text colour with an important mark and its
own background under the pointer), and a line of text (Kumo's Button is a box of 20 px or more
with its own padding).

The name's line wraps. With the head laid out as it was, a mailing list's long sender beside
Unsubscribe was drawn 0 px wide on a phone, so the browser test found it hidden. Unsubscribe now moves under the name when both do not fit.
A browser test at phone width with such a sender holds that the name is cut, is shown, keeps its
line above Unsubscribe and ends before the date; it fails without the wrap.

The line of a folded message and the line that stands for the middle of a long conversation stay
native too, each as one ON_PURPOSE line with its reason.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
@Madraka
Madraka force-pushed the fix/kumo-first-screens branch from d3496fe to 99d633d Compare November 19, 2026 08:05
Madraka added a commit that referenced this pull request Nov 19, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Madraka <8898640+Madraka@users.noreply.github.com>
@Madraka Madraka changed the title fix(app): the sign-in, the mailboxes page, the top bar, the list, the search and the open message use Kumo's controls fix(app): the sign-in, the mailboxes page, the top bar, the list, the search and the open message use Kumo's controls, keeping their look Nov 19, 2026
@Madraka

Madraka commented Nov 19, 2026

Copy link
Copy Markdown
Contributor Author

audit: at 99d633d, opened on this head and on main (8089522) at 1280 px and 375 px. Findings 4, 5 and 6 are fixed; two small ones are left, both of the same kind, and then this passes when the run on the last head is green.

Fixed, by measurement (this head = main unless said)

  • The head of an open message at 1280 px: 40 px; the name 13 px at y=181.4, x=333, 116.5 wide; the line under it at y=197; the body begins at y=238; the date at x=1070.9. Every number is main's.
  • A conversation of two messages: both heads 40 px, both names 13 px level with the picture's top. One's own picture: oklch(0.97) on the brand colour.
  • Unsubscribe: underlined, 12 px, grey; its word begins at x=457.5 as on main; its box is 100.9 × 20 (86.6 × 16 on main), so the target grew and the head did not.
  • At 375 px, a mailing list's message: the name is cut in its 102.9 px, Unsubscribe stands under it, the line of recipients under that; head 47.3 px; nothing scrolls sideways. The wrap is accepted, with the reason the description gives.
  • The answers: fully round, 36 px, at y=453 and the same widths as on main (88.1, 144.2, 64.6 in Turkish).
  • The quick filters at rest: fully round, 26 px, 12 px type.
  • The search box (672 × 48 at 1280 px, 233 × 48 at 375 px, fully round) and the star (26 × 26, beginning where the box's target ends, at both widths) hold as before. The dark theme was looked at on the open message at 375 px.
  • ROUND_ON_PURPOSE is the right size of exception: one shape, a listed place, an exact count, a reason. The sentence added to the skill is a rule, not a control's description; read and accepted.

Left

  1. A chosen quick filter. On main a filter that is on has a tint of the brand colour, a line in the brand colour and dark semibold text. Here it is filled with the brand colour and its text is white (oklch(0.704 0.163 260), measured after pressing "has an attachment"). The filters are also 9 px narrower each (8 px of padding for 12). The description does not say either. Keep main's look for a filter that is on (a colour is not a size, so the check lets it through), or say the change in the description with a picture of before and after.

  2. The sign-in's second step. Kumo's ghost kind has no fill at rest (hover:bg-kumo-tint), so the four pictures show the bar under the pointer, not as a person finds it. At rest it is a line of dark 14 px text with no underline under Continue; on main it is small grey underlined text. Give it what Unsubscribe got (grey and underlined, so it reads as the text action it was), and take the pictures at rest.

Accepted changes of look, as said in the description: Sign out as a small button; the search box's buttons at 26 px; the star at 26 px with the sender's column 6 px to the right; Unsubscribe's taller target.

Not opened by me: the sign-in's second step (the example has no account with a second step); finding 8 is read from the pictures and from Kumo's own classes.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant