An empty read window (byte cap below the first selected line: `lines: []`
with `totalLines > 0`) dropped `offset` from the persisted presentation
meta, so a replayed read card could not report where the window starts or
where a continuation resumes. Carry `offset` on `FsReadMeta`,
`ReadResultView`, and the `presentationMeta` projection, and validate it in
`readMetaFromMeta` (1-based integer; the first line number may not fall
below it). Re-record the ACP fixtures and the cordis api catalog.
Also correct the Note's `parallel-file-reads` golden path
(examples/tui-agent -> apps/cli) and record the pre-card replay-degradation
tradeoff in the Decision section.
A filename whose extension is an Object.prototype key (foo.constructor,
foo.__proto__) resolved to the inherited member through the plain-object index,
so a function reached the read card's lang hint and failed the tool-output JSON
validation, failing an otherwise successful read. Look the extension up as an
own property only. Added rejection tests, converted the zh Note headings to the
all-English sibling convention, and named the parallel-file-reads terminal
golden as the TUI-unchanged evidence in the Testing section (both languages).
readMetaFromMeta narrows the opaque persisted meta boundary, so beyond
shape it now rejects replayed JSON that is well-typed but semantically
invalid: line numbers must be 1-based integers, totalLines a non-negative
integer, and line numbers must strictly increase without exceeding
totalLines. Any violation declines to the generic fallback.
Sync the public ToolResultView contract across the core/tools and
tool-fs READMEs and docs/core-data-structures/tools for the fourth
result-view member and the ReadFileLine vocabulary, and expand the
Agent Note Testing section with the new rejection paths and the snapshot
evidence this PR carries.
A read result now carries card:'read', but render()'s genericContent gate was
card==='generic' only, so the read body kept its text yet lost the dim-Markdown
dimBody treatment the generic card gave it. Admit card:'read' to that gate so
its content fallback takes the same dim path, restoring read's TUI rendering to
what it was before the read card existed. Refresh the parallel-file-reads TUI
golden accordingly and correct the Note's TUI claim on both language sides.
The read tool now projects presentationMeta ({path, lines, totalLines}) onto
its tool/result, so every scenario with a read call carries that meta; the
cordis-inspect snapshot's embedded type surface gains ReadResultView /
ReadFileLine / the widened ToolResultView. Model-facing text is unchanged.
Refreshed keyless via test:snapshot:refresh. The unrelated goal.snapshot
SQLite ExperimentalWarning failure is pre-existing on clean master.
The read tool's result carries structured numbered lines, but only the
model-facing envelope text reached the client. Add a card:'read' result view
(ReadResultView) projecting {path, lines, totalLines, lang} through the tool's
output.presentationMeta so presentResult reproduces it on live and replay
paths; the pending call stays a generic read card. A UI without the read
capability falls back to the envelope-stripped content, so the TUI is
unchanged. The web consumer that renders the line-numbered view is a follow-up.
CI's coverage gate rejected `ansi.ts` after the state normalization: the new
fold carries shapes the existing cases never reached.
Added, each expectation checked against a real terminal where the terminal is
the authority — three of my first guesses were wrong and the implementation
was right:
- `48;2;R;G;B` (the extended-background arm and its 4-parameter span), a bright
foreground and background (`91`/`101`, which resolve to `--dsw-*` tokens
rather than literal rgb), a `38` with no kind byte, re-opening an attribute
already in force, and the bare `\x1b[m` reset.
- Clearing a wide pair from the spacer side: `中x\r\bA` shows `A x`, since the
backspace clamps at column 0 and writing the lead blanks its spacer instead
of letting the `x` slide left. An erase reaching the lead does the same, so
`中x\x1b[1K|` shows three blanks before the bar.
Nine findings, one critical. Terminal cases verified in a real terminal first.
CRITICAL: cells held the accumulated SGR history, so every state boundary
re-emitted the whole chain — output switching color without a full reset
emitted O(n^2) characters. Measured: 3200 such cells produced 25 MB, and the
reviewer's ~90 KB alternating-color case is well under bash's own output cap.
State is now a normalized record (foreground, background, attribute set) with
one canonical sequence per boundary, so the emitted text is linear in cells;
the 90 KB case parses in 36 ms. That also makes the attribute closers every
chalk-based tool writes actually close: `\x1b[1mbold\x1b[22mplain` leaves the
following write PLAIN, which a real terminal confirms.
Width follows emoji presentation, not the U+2600-U+27BF block: `A✓B` redrawn
with `XY` shows `XYB`, so the check every progress line writes is ONE column.
Taking the block as wide misaligned exactly the output this card exists for.
Writing over either half of a wide pair blanks the other, since a terminal
cannot leave one cell of a two-cell glyph standing.
`line\n\x1b[0m` does not end in a newline as a string yet its last parsed line
holds nothing visible, so the terminator check now reads the parsed lines — it
had added a blank row and inflated the collapse count.
A line with no cursor movement no longer builds a column buffer at all; only
its SGR is folded, so an `ls -R` or a 5k-line log allocates nothing per
character.
The `.terminalDescription` rule had been inserted into an existing grouped
selector, silently giving `.codeBody` description typography and changing its
bottom margin from 4px to 0 — a pre-existing surface this PR does not own.
Split out, `.codeBody`'s margin restored.
Three comments contradicted their code: the fixture's exit-marker line (still
claiming recovery from a marker deliberately removed), `bash-sample`'s header
(still routing a click to the details panel, and calling the consumer's cap the
block's own), and a DetailsPanel comment stacked above the wrong rule. The
ui-primitives README documented only the CR/BS half of the replay, so a reader
would expect `OK0%` where `100%\r\x1b[KOK` renders `OK`.
CI's coverage gate caught `ansi.ts` at 95.31% branches. My local check had
scoped coverage to that one file, which measures a different test set than
the gate does — the gate is the authority and it was right.
Three branches. Two are now pinned, each verified in a real terminal first:
`ab\r` + U+0301 + `x` shows `xb` — the redraw leaves the cursor at column 0,
so a combining mark has no cell to attach to and the terminal shows nothing
for it. It also revealed that the mark was kept when `cursor` was 0, which is
fixed.
`ab\rX\x1b[31m` then a plain line shows `Xb` unstyled and the next line RED:
the mirror of the reset case, where the scan ends styled while the last cell
is not, so the convergence has to OPEN the run at the line end for it to
reach the following line.
The third was `?? ''` on a `String.split` result, which always yields at
least one element — removed rather than tested, since no input can reach it.
Eight findings, each terminal case verified in a real terminal first:
`\x1b[32mdone\rok\x1b[0m` then `plain` shows `okne` green and `plain` in the
DEFAULT color. The replay returned the last written cell's state, so a reset
landing after the final write vanished from both the text and the returned
state — and every build tool writes exactly that shape, so the color leaked
onto all later output. The replay now converges to the state the scan ended
in, which is also what it hands to the next line.
`abcd\b\x1b[1K|` shows ` |`: CSI 1K erases THROUGH the cursor column, and
the loop stopped before it. The erase mode also reads only the first
parameter now, since a terminal treats `1;2K` exactly as `1K`.
`éx\rYZ` shows `YZ`: a combining mark takes no column, so it attaches
to the cell already written instead of advancing the cursor and leaving the
`x` standing.
`中x\rA` shows `A x`: overwriting a wide character's lead cell leaves its
spacer as a blank rather than closing the gap, which would shift everything
after it one column left.
The banner lost its span when the gutter became padding — a plain block child
only reaches the content box, so the reserved column was painted in the body
color and the card's top-left radius drawn in it. Invisible in the light
theme, where banner and body share a token; visible in the dark one. The
header now pulls back across the gutter and re-insets by the same amount.
The replay trigger matches the same CSI shape the parser accepts, so a form
like `\x1b[1;2K` can no longer skip its own erase, and `replayLine`'s JSDoc
documents its new parameter and returned pair.
Docs: four places still described the dot as sitting left of the card surface,
which stopped being true when the gutter became the card's own padding, and
two fixture comments still referenced the exit marker that was deliberately
removed.
Six findings. Each terminal-semantics case was checked in a real terminal
(tmux, reading back the painted screen) before changing anything:
`100%\r\x1b[KOK` shows `OK`. Modelling the `\r` without its erase left the
previous frame's tail standing — a regression against the old truncate, since
`\r\x1b[K` is the single idiom every spinner and progress bar writes. Erase is
now part of the same replay, in all three parameter forms.
`a\tb\rXY` shows `XY b`. Counting a tab as one column produced `XYb` and
destroyed the alignment this card exists to hold, so the cursor now advances
by terminal columns: tabs reach the next 8-column stop and a wide character
takes two cells.
`\x1b[31mabc\rX\nnext` paints BOTH lines red. A newline does not reset the
graphic state, so state threads from one replayed line to the next instead of
closing at each line end.
Only `m` accumulates into a cell's style now. Folding cursor and erase
sequences in grew the state string per redraw and emitted boundaries anser
had to discard.
The empty check reads the parsed lines the card renders rather than the raw
text: output that is only escapes or control bytes survives `trim()` yet
parses to nothing, and drew blank rows plus a copy control for invisible
bytes instead of the placeholder.
The gutter is the card's own left padding rather than a margin. Every render
site rewrites `margin` wholesale for its own indent, which silently cancelled
the reservation and let a container clip the dot.
The fixture sample no longer carries an `[exit code: 1]` line: the real bash
presenter consumes that marker precisely because the card shows the exit as
its own pill, so the built-bundle snapshot had pinned a frame showing it
twice — one the product path cannot produce.
Carriage return and backspace only MOVE the cursor; neither erases. Both of
my earlier approximations were wrong, and I checked each case against a real
terminal rather than reasoning about it:
`100%\rOK` shows `OK0%`, not `OK` — the redraw is shorter than the frame
beneath it, so the tail stands. `abc\b` still shows `abc`, not `ab` — a
trailing backspace has nothing to overwrite. `\x1b[31mgone\rkept` paints
`kept` RED, because a carriage return does not reset the graphic state, which
one of my own tests had asserted the opposite of.
Both now replay into a per-line column buffer with SGR state stamped per
column, as a terminal stores it per cell. That gives the partial-overwrite
case its real result too: red `bad`, three backspaces, then `ok` shows `okd`
with the `d` still red, since `ok` reached only two of the three cells.
The presenter description now also renders at every site. An expanded row
draws it itself — the collapsed summary is hidden while open, so otherwise
the description was visible only collapsed, the opposite of "above the card"
— and the details panel draws it above the card as well.
Three of my own tests encoded the wrong semantics and were corrected with
their behavior, and the emit loop's gap-filling arm was removed as
unreachable: `\r` and backspace only move left, so no column can be unwritten.
Four review findings. Two are defects the previous two rounds introduced,
which the existing tests did not catch:
A backspace erased raw bytes, so one landing after an SGR reset ate part of
the escape: `\x1b[31mabc\x1b[0m\b\bXY` left `\x1b[` and repainted the rest of
the line with whatever the remainder parsed as. Backspaces now resolve over
VISIBLE characters — a CSI sequence is one indivisible unit a backspace steps
over on its way to the last printed character, so the surviving text keeps
the color its run authored.
The cwd normalizer popped a UNC share root: `\\server\share` with a `..`
became `/server`, losing the separators too. A UNC path's server and share
are its root, and Windows cannot climb above a share, so they are split off
and the remainder collapses against that root.
The other two are gaps the earlier fixes left:
The render-site fallback row still passed the args-derived summary, so any
terminal-declaring tool without its own keyed row (`terminal_send`) lost the
contract's above-card description. It now prefers the description exactly as
BashRow does.
A settled call read `call?.cwd`, which cannot tell "the call omitted a cwd"
from "the paging window dropped the call head". The second case has no cwd
anywhere and the original call may have used an explicit workdir, so it now
draws a bare `$` instead of naming the session workspace.
A multi-line command repeated the cwd label on every prompt row, which
states something the view does not know: it carries ONE working directory —
where the call started — and a `cd` in the command moves later lines
elsewhere. `cd ~` then `ls` rendered both rows labelled with the session
workspace while `ls` actually listed the home directory.
The label now appears on the first row only, and later rows keep a bare `$`
so they still read as prompts. Same reasoning as the run-state dot: neither
a per-line directory nor a per-line exit status exists to report.
The built-bundle snapshot records the effect on fixture turn 60's two-line
command (`fixture echo done` becomes `$ echo done`).
Three review findings, each verified against the presentation contract:
The call view's `description` was dropped, so a presenter that authors one
(`terminal_send` declares `Terminal <id>`) lost the contract's above-card text
and the row fell back to an unrelated args-derived summary. It now rides the
same derivation and outranks that summary.
A relative workdir was concatenated but never normalized, while the bash
executor resolves it before running: with session cwd `/w/app` and workdir
`..` the command runs in `/w`, yet the card displayed the label `..`. The
resolved path now collapses `.`/`..` segments, drops a `..` that would climb
past a root the way a filesystem does, and keeps a Windows path's separators
since the value is only ever displayed.
`run_code` sub-dispatches carry no presenter views on the shipped wire —
`session.ts` folds `tool/code-dispatch(-start)` with null views and the host's
`viewFor` presents only top-level call/result events — so a nested bash call
cannot reach a terminal card. The existing test only passed by injecting views
that path cannot produce; it now says so, and a second arm pins the no-view
shape the wire actually delivers.
Restoring master's fixture also fixed the todo snapshot lane, which my earlier
merge had broken by dropping the projection support the todo dock reads. The
terminal sample turn moved ahead of the todo turn, because the standing plan
retires at the next `turn/start` and a turn appended after it emptied the dock.
The card props are now nested under `card` so a render site spreads exactly the
primitive's own surface, and the fixture reads each sample's authored exit
status instead of re-implementing the bash tool's `parseExitStatus`.
A `command` carrying two shell commands on two lines rendered as one row:
`.command` had `white-space: nowrap`, so the two collapsed into a single
ellipsized line that read as one command with stray arguments.
Render one prompt row per command line, and move the run-state dot out of
flow into a gutter reserved to the left of the card surface, so it neither
indents its command nor depends on the command's text metrics to line up.
The dot stays exactly one per card, on the first row. The exit status the
view carries is the whole call's and bash reports no per-command status, so
a dot per line would assert, of a line that succeeded inside a failing call,
that the line itself failed. The single visually hidden label keeps the same
scope, since one label per row would read to assistive technology as several
distinct outcomes.
Fixture turn 60's command becomes two lines, so the built-bundle snapshot
pins the layout and its dot distribution (`dotsPerPromptRow: [1, 0]`), and
the e2e adds that the dot starts left of the card surface — geometry jsdom
cannot compute. Both READMEs now also record that this package's
user-facing copy is inline Chinese, since zero-cordis atoms have no route to
`ctx.locale`; extracting it belongs to the repo-wide localization work.
The check added last commit derived its elevated set from the sheets that
already rebind, which cannot catch the omission it targets: such a set only
confirms what someone already remembered, and a surface nobody has rebound
yet defines itself as unelevated. Review found the case that proves it —
TodoPanel scrolls in .list on a --dsw-specific-tip card, the same dark rung
as the menu surface, unrebound and with the derived check green.
Resolves the set from the palette's own dark elevation ladder instead: the
surface tokens whose dark value lands on bg-layer-2 or bg-layer-3, which is
the step the l1/l2 split encodes. A new palette token on an elevated rung is
in scope the moment it is defined.
Scope is by token family rather than geometry: only --dsw-alias-bg-* and
--dsw-specific-* name a surface. The button, interactive, and markdown
families reach the same rungs while naming a control or an inline span that
no scroll container renders a bar against, and shape cannot separate them
since a floating button carries a radius, a shadow, and a fixed size —
ChatView's .toBottom pill was the false positive that showed this.
Adds the missing TodoPanel rebind. Mutation-checked all four rebinds in
turn: each is named with its surface. The palette anchoring has its own
control — narrowing the family pattern turns it red on --dsw-specific-menu.
test:web ran build:web alone, which does not rebuild UI plugin client
bundles. Plugin CSS reaches the browser through packages/client/*/lib/
client.js, not apps/web/dist, so a changed *.module.css served its
previous bundle: the run exercised stale CSS and a removed declaration
still passed. That is how I first mistook a valid gutter test for a
vacuous one.
Root build already covers packages/*/*, so running it first is enough;
check-all already ordered build before build:web, so CI was never
exposed. Only the local script was, which is where a stale-bundle pass is
most likely to be believed.
Verified against the situation it fixes: mutate the source, rebuild the
bundle, restore the source, and the artifact is left without the
declaration while the source has it. Under the old script the run tested
that artifact; under the new one the artifact is rebuilt first (grep goes
0 to 1) and the scrollbar spec passes. The nine failing web files are the
pre-existing aria-golden set from f2c004524, unchanged by this.
Review found three scroll containers sitting on surfaces the rebinding
contract covers, none of which rebound: ui-primitives' shared Menu card
on --dsw-specific-menu (the surface PopupSelectView already rebinds for),
and the composer input and question composer cards, both on
--dsw-specific-input-major. Each rendered the l1 thumb, which differs
from l2 only in the dark palette and only on that surface, so a
light-palette screenshot and a code read both look correct.
Adds the mechanical check that would have caught them instead of leaving
it to inspection: a sheet that scrolls somewhere and paints a known
elevated surface somewhere must rebind. The elevated set is derived from
the sheets that already rebind, since a rebinding rule paints the surface
whose elevation it declares, so a new elevated surface joins the set by
rebinding rather than by anyone updating a list. Surface-level rather than
element-level because the card and the descendant that scrolls are
separate rules and CSS text does not say which contains which. Verified by
reverting each of the three fixes in turn: the check names the sheet and
the surface every time.
Also commits snapshots/sidebar-scrollbar/geometry.expected.md, the
resolved scrollbar style and geometry in both palettes. The aria goldens
the other web scenarios commit cannot carry a CSS-only change, since it
alters no DOM and no accessible name and leaves their trees
byte-identical. Absolute coordinates stay out: they track font metrics and
the laid-out sidebar width, so committing them would document the platform
and force a per-platform re-record.
The e2e measured the reserved band but never the symptom the change is
named for. Headless chromium defaults to an overlay scrollbar, which is
the configuration where a bar can cover row content at all, so the
scenario already ran in the right mode: against clean master the band is
0 and the bar covers 7px of the relative time.
Adds timeCoveredBy, the overlap between the relative time's right edge
and the range the bar occupies, taking the bar's width from the sheet
where it applies and from the UA's overlay width otherwise. Assuming 0
there would report no occlusion in precisely the state that has it.
Keeps the band assertion rather than replacing it: the two catch
different regressions. Removing only scrollbar-gutter leaves
timeCoveredBy at 0, because the bar is then 8px and the row's right
padding is also 8px, so it abuts the timestamp without covering it.
Removing the pseudo-element width as well is what produces the overlap.
Each was mutation-checked with the other assertions in its test silenced.
Records in the note that the gutter and the ::-webkit-scrollbar width are
jointly necessary against an overlay bar, measured by deleting each from
the live cascade with the other in force: either alone drops the band
from 8 to 0.
Master replaced StateDot's ongoing ring with a pixel-art chase, so the
prompt line's run-state description named an indicator that no longer
exists. Same fix in the note, both READMEs, and the test name.
The terminal card showed no run state: a running command and a settled
command that produced no output rendered the same prompt line, so whether
a command was still running had to be inferred from the absence of output.
Lead the prompt line with a StateDot in three of its states — the spinning
ring while running, red for the same exit status that renders the status
pill, green for a clean settle. That is the same indicator a tool row's
leading icon carries, so a row and its own card cannot disagree about one
command; the row/card agreement is pinned in the ui-conversation spec.
StateDot is aria-hidden, so a visually hidden text label rides beside it,
which is what the refreshed aria goldens now record.
The e2e adds what jsdom cannot compute: the dot's color resolves to the
green success token through the real theme stylesheet, and the dot precedes
the prompt label in document order.
A non-`auto` `scrollbar-width` or `scrollbar-color` makes Chromium and Safari
discard every `::-webkit-scrollbar*` rule for that element, including
`::-webkit-scrollbar-thumb:hover`. Declaring both unconditionally left the
hover tokens rendering nowhere: the engines implementing the hover
pseudo-element are exactly the ones the standard properties silence, and
Firefox has no hover pseudo-element to fall back on. Both hover tokens and all
four elevated surfaces' hover rebinds were therefore dead code.
Measured in chromium on probe elements with `scrollbar-gutter: stable`: an 8px
`::-webkit-scrollbar` alone reserved a 30px band, and adding
`scrollbar-width: thin` dropped it to the 10px `thin` reserves.
The standard properties now sit inside `@supports not
selector(::-webkit-scrollbar)`, so Firefox takes them and WebKit-based engines
take the pseudo-elements. The WebKit rules stay ungated: an engine without
those pseudo-elements drops them as unknown selectors, and gating them would
hide them from an engine that implements them without `selector()` — the
pre-16.4 Safari the ungated form serves correctly.
Three unit assertions pin the split by source offset, which the existing
at-rule-flattening parser cannot see. The web e2e now reads the path chromium
actually takes: the `auto` standard properties as the gate's signature, the
pseudo-element sizing and track, the indirection variables resolved per
throwaway probe, and the hover declaration as cascade rule text — chromium
folds the `:hover` rule into `getComputedStyle(el,
'::-webkit-scrollbar-thumb')`, so no computed query separates the states.
The bash tool already declares the `card: 'terminal'` render intent for
both its call and its result, and host/connection/runtime already deliver
it to the browser as callView/resultView. The Web client ignored it:
rows derived from raw args, and the details panel flattened every tool's
content into one soft-wrapping `<pre>`. Column-aligned output folded into
a paragraph and a long listing stretched the panel without bound.
`TerminalBlock` (ui-primitives) renders a command as a terminal surface:
a shortened-cwd prompt line, output at `white-space: pre` in a
horizontally scrolling box, a head/tail height cap with an expand
control, an exit-code/signal status pill, and a copy control for the raw
output. ANSI SGR runs are parsed with `anser` and resolved onto `--dsw-*`
theme tokens, with literal rgb kept for values the design system has no
token for. Geometry and fonts mirror CodeBlock; the clipboard write both
need moved into a package-internal `clipboard.ts`.
Both Web render sites for a bash call consume the intent through one
derivation (`terminal-card-model.ts`), so they cannot disagree about a
command, its cwd, or its exit status: the keyed BashRow carries the card
resident below its summary row, and the render-site fallback row keeps it
behind its existing expand control. Rows cap at 8 lines against the
panel's 16.
Inline output in the chat row reverses this package's stated
no-inline-output convention, on the owner's explicit decision; the Agent
Note records the reversal and its bound.
Tests: TerminalBlock/ansi/clipboard unit specs, ui-conversation wiring
specs at every render site, a built-client-graph snapshot covering both
chat-row shapes, and a real-browser e2e asserting the no-wrap layout and
the page's own Clipboard API.
Verifying browser-visible plugin CSS needs a rebuild build:web does not
perform: WorkspaceBrowser.module.css never reaches apps/web/dist, because
ui-workspace loads as a runtime plugin with its CSS inlined into
lib/client.js by that package's own bundle script. A negative control that
reruns only build:web exercises a stale bundle and passes with the
declaration removed, which reads as a vacuous test rather than an invalid
control. No script in the web lane does this rebuild, so every scroll-region
or plugin-CSS change hits the same trap.
Every scaffold-importing e2e compiles on the host plane, so the new file
goes in tsconfig.host.json's include list and apps/web/tsconfig.json's
exclude list. Without both, tsc -p apps/web/tsconfig.json fails with
TS6059/TS6307.
design-platform.css declared four --dsw-alias-scrollbar-* tokens in both
palettes that no rule read, so every scrolling region rendered the user
agent's own scrollbar and the dark theme showed a light native bar against
dark surfaces.
The symptom that surfaced the gap was in the sidebar: the workspace
browser's session list is its only scrolling region, and each row's
trailing content (the relative timestamp, and the hover action buttons
that replace it) is `flex: none` flush against the row's 8px right
padding, so an overlaid scrollbar painted on top of the timestamp.
ui-theme/styles/scrollbar.css becomes the sole consumer of the four
tokens, imported by the web shell's base.css after design-platform.css
because it reads that sheet's tokens. The rules sit on `body`, not
`html`: the alias tokens are declared on `body`, custom properties
inherit only downward, and from `html` they resolve to the
guaranteed-invalid value with scrollbar-color computing to `auto`.
scrollbar-width and scrollbar-color are declared on `body, body *` rather
than inherited, because inheritance would carry the color already
substituted at `body` and an elevated surface could not retint its own
thumb; scrollbar-width does not inherit at all.
Both the standard properties and the ::-webkit-scrollbar pseudo-elements
read one indirection pair bound to the l1 tokens, so an elevated surface
rebinds that pair to the l2 tokens once and retints both renderings. The
command popup, slash menu, model-select panel, and settings panel do so,
which gives the l2 tokens their first consumers.
WorkspaceBrowser's `.list` declares scrollbar-gutter: stable, keeping the
bar beside the rows. `stable` rather than `auto` so the reservation holds
when the list is short enough not to scroll: expanding a workspace group
would otherwise shift every row sideways at the moment it starts
scrolling.
The card cap is viewport-relative (min(60vh, 520px)), so shrinking the
conversation column clips the card without under-allocating it. Anything
demonstrating or measuring this defect outside the e2e scenario has to
change the viewport.
The Chinese counterpart kept its title and every section heading in
English. The bilingual contract localizes heading text; only the
`# Agent Note:` prefix and the `Status:` token stay verbatim, matching
the renderings already used across the note corpus (问题 / 决策 /
曾考虑的替代方案 / 后果 / 验证). Pairing hashes re-recorded.
The answered golden omitted the `复制` / `在新对话中分支` / `编辑` buttons
that `MessageItem`'s `UserActions` renders under every user bubble, so the
scenario failed against the built client. They are unconditional in the
accessibility tree — the `@media (hover: hover)` rule only sets `opacity`,
which does not remove a node from an aria snapshot — so there is no
message-row state to stabilize before capture; the golden was simply
recorded against a stale `lib/`. A partially failed `pnpm run build` had
left the client bundles from before its failure current and the rest not.
The Agent Note records that trap: confirm the build exited zero before
refreshing a golden, and note that untracked directories under `packages/`
are compiled too, so a leftover from another branch fails the build for
reasons the diff does not explain.
The same omission is present in the other 15 committed web goldens on
master, where 9 tests across 7 files fail for this reason on a pristine
tree. That is pre-existing and belongs to whoever landed `UserActions`;
this commit fixes only the golden this PR already touches.
The load-time range check rejects a maxWallMs above Node's maximum
setTimeout delay, but the constraint appeared only in the README design
section. Deployments reading the Config field JSDoc, the generated config
catalog, or the README config summary saw maxWallMs described as a
positive finite wall-clock ceiling, so an out-of-range value looked valid
until plugin load failed.
The question composer card is capped against the viewport and scrolls its
option list. `.options` is a flex column whose children defaulted to
`flex-shrink: 1`, so a short seat shrank the rows before overflowing the
scroll container: a row collapsed to its 42px minimum while `.optionCopy`
kept the taller height its wrapped copy needs, and `align-items: center`
then painted that copy outside the row's border box — over the question
title above and the next row below. Measured 6.5px of spill at 900x440 on
the shipped client, 10px at 380px tall, with `.options` reporting
scrollHeight === clientHeight and therefore offering no scrollbar.
`.option` and `.custom` now declare `flex-shrink: 0`, so the shortfall
reaches the scroll container that already owns `overflow-y: auto` — the
behavior the cap was designed for. Only rows whose copy wraps could
reproduce this, which is why the recorded scenario now asks a question
with long option descriptions; the web e2e asserts at three squeezed seat
heights that every row's children stay inside its border box, guarded
against holding vacuously by requiring a wrapped row and a scrolling list.
The paragraph claimed later child output reaches only the scrollback. Output
after a settle still reaches the scrollback and whichever send is active when
it arrives — appendOutput writes both (session.ts:302-307), and settleActive
clears `active` rather than suppressing delivery. Only the settled operation
stops receiving it, which is what the timing bound protects.
master already stabilized the raw-mode SIGINT scenario in 65e0ffe03 and
3be090f1d with wider scenario-owned bounds and deliberately delayed child
readiness, so the merge takes master's test implementation. What this branch
keeps is the contract that explains why those bounds are required: the note
paragraph on the settle boundary in PtySendOperation.append, and the
waitForOutput comment recording that raising its deadline alone cannot
recover output the send never collected.
- The clone exclusion is now '.runner*' (catches .runner_migrated /
.credentials_migrated, which GitHub writes on migrated runners and
which equally trip the already-configured refusal — hit live during
the 2026-07-27 org-group migration).
- The problem statement now states the verdict runs on ubuntu-latest,
scopes the switch to enterprise Linux-pool outages, and names the
residual dependency: node-compat/python-sdk/windows stay standard-
hosted, so a broader hosted-capacity failure still blocks the
verdict, with Windows the only leg lacking any in-house substitute.
Static gate green locally: 32 passed, 0 failed.
The Agent Note paragraph recorded test-specific timing advice, which
docs/AGENTS.md excludes from implemented notes. The durable half of it —
LocalSendOperation.append drops output after settle, so it survives only in
the scrollback — now documents append itself.
Two sessions resolved the same all-checks-passed conflict differently. The
remote's version is the one kept: the selector with '|| ubuntu-latest'
covers a standard-hosted outage too, which a bare ubuntu-latest pin does
not, and its runbook wording is internally consistent at four failover
jobs. Reverted this side's three-job doc downgrade.
One semantic conflict on the `all-checks-passed` verdict job. Master moved
it to `ubuntu-latest`; this branch had routed it through the failover
selector so a hosted-pool outage could not leave the required verdict
queued. Master's resolution satisfies that requirement more directly:
standard hosted capacity is independent of both custom pools, so the
verdict is reachable whichever pool is degraded, and the selector is no
longer needed. Kept `ubuntu-latest` and folded the failover reasoning
into its comment.
The failover runbook accordingly documents three failover jobs (the three
required Linux workers), not four, and states why the verdict stays on
standard hosted capacity in both states.
config.sh only registers; the runner stays offline until svc.sh
install/start. Both language sides updated so emergency capacity
actually comes online.
- serial-linux-selfhosted checks out fetch-depth 0: depth 2 misses
github.event.before on multi-commit or force pushes, failing the
archive verifier on a valid tree. Full fetch is cheap against the
VM's local mirror.
- Runbook (both languages): every remaining admin phrasing (problem
statement, switch heading, alternatives, consequences) now says
writer; and the 'composes with this mechanism' claim about a
master-ref-pinned runner group is replaced with the truth observed
live on 2026-07-27 — master-ref pinning blocks PR failover, and the
shipped posture is repository-scoped all-workflow group access.
Static gate green locally: 32 passed, 0 failed.
Chinesezjc