* Smooth scrolling: mouse-wheel notches glide like Chrome's, with a setting
A mouse-wheel notch moved content in one jump: GPUI hands non-precise
wheels to containers as `ScrollDelta::Lines`, applied at once. The window
root now takes each notch at the end of the capture phase and replays it
over the following frames as fractional `Lines` slices, on Chrome's wheel
curve (`cc::ScrollOffsetAnimationCurve` for `kMouseWheel`): ease-in-out,
cubic-bezier(0.42, 0, 0.58, 1), over 200ms, shortening to 100ms for long
moves. A notch in flight retargets at the current speed, so a spin is one
continuous glide. Slices stay in lines, so every container keeps its own
line height and a notch travels exactly as far as before.
Precise input, notches with Ctrl, Alt, Cmd or Fn held, and content that
eases notches itself (WebView2) stay raw; a press ends a glide.
The in-app browser on macOS glides too. AppKit hands a page its wheel
directly and WKWebView applies a tick at once, so a local event monitor
holds the page's ticks and replays the same glide as precise pixel deltas,
which WebKit routes like a trackpad's: to the scroller under the pointer,
clamped at its edges.
Settings → Appearance → Motion → Smooth scrolling, on by default. Reduced
motion turns it off.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Transcript: upward input breaks the bottom pin as it arrives
The pin broke only when one wheel event moved the view more than 1px from
the bottom, decided in a deferred check. A glided notch starts and ends
with sub-pixel slices, and its slices arrive on frames, where the spring's
step runs before that check: it pulled the view back and recorded the
distance the check compared against. A notch up from the bottom of a chat
drifted back down. Upward input over a scrollable transcript now breaks
the pin on the spot, as use-stick-to-bottom's wheel handler does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: smooth scrolling media
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Transcript: only a wheel the list took breaks the bottom pin
`on_scroll_wheel` broke the pin for any upward wheel that bubbled to the
transcript root, including one an overlay above the list let through
(an image viewer at its pan limit, say) while the list never moved. The
list's scroll handler now marks the event it took, and only that event
can escape the pin.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Smooth scrolling: the AppKit glide flag is macOS-only
Only `browser::macos` reads it; elsewhere `gliding` was dead code.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Smooth scrolling: leave touchpads alone on X11 and Windows
X11 reports touchpads (and high-resolution wheels) as fractional
`ScrollDelta::Lines`, so every touchpad event was held back and eased
over 200ms. Outside macOS, where trackpads are flagged precise, only a
whole-line delta is a wheel notch; fractional lines now pass through.
The transcript's arrival-time pin escape is for wheel notches only:
precise input keeps the distance test it had before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Smooth scrolling: drop the browser's unused wheel fixture hook
No fixture calls fixture_wheel; it served a throwaway capture that was never
committed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Smooth scrolling: findable in settings search
The Motion row joins the settings index under Appearance → Motion, with the
words people reach for: mouse, wheel, glide, ease.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Settings search: the Appshots pick test runs where Appshots shows
Appshots is hidden off macOS and Linux, so search can't find its rows there
and the test failed on Windows since #1018.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Wing <44917547+wingleeio@users.noreply.github.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Claude subagents: fill a typed fork's tab while the CLI holds it
A `/skill` typed as the prompt and run in a fork has a chip and a tab,
but the tab held only the skill's prompt until the fork was over: the
CLI sends the fork's messages "together and in order after the forked
skill finishes"
(https://code.claude.com/docs/en/headless.md#follow-subagent-messages).
The fork has a transcript of its own on disk, though:
"`agent_transcript_path` is the subagent's own transcript stored in a
nested `subagents/` folder"
(https://code.claude.com/docs/en/hooks.md#subagentstop-input). In
2.1.310 that's `<project>/<session>/subagents/agent-<task>.jsonl`,
created as the fork starts and appended as it works (live-verified).
While a typed fork runs, the driver now reads that file every 250 ms.
It's looked for under every project directory, the way Grok's driver
looks for a subagent's history, not by encoding the cwd, and the reader
is Grok's, moved to `jsonl_tail` for both. The fork's messages (text,
thinking, tool calls) and its tool results go to its tab. Its prompt,
user text and the CLI's bookkeeping records don't.
The replay repeats each record under the uuid it has on disk, so a
record already shown is dropped, and the first replayed frame ends the
reading: the replay brings the rest. If the file hasn't appeared after
10 s (`CLAUDE_CODE_SKIP_PROMPT_HISTORY`, `--no-session-persistence`, a
wrapper's own config dir), the tab fills from the replay, as before. If
a replay repeats none of the records read, they showed twice: a warning
says so, and the run stops reading typed forks' transcripts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Typed-fork-live demo media for the PR (live before/after stills + GIF)
Settings → Providers → Claude Code gets a Native subagents switch (per device, on by default). With it off, a top-level chat launches Claude Code with --disallowedTools=Agent,Task,Workflow, and Zeron's MCP server (ZERON_SIDE_CHATS_ONLY=1) tells the agent to delegate through side chats. Side chats, and chats whose parent is not known yet, keep their subagents. A parked session is replaced before its next idle turn.
Co-authored-by: Alexandre VARGAS <alexandre.vargas.lopez@gmail.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Claude subagents: give a typed /skill's fork a chip and a tab
A `/skill` typed as the prompt, when the skill runs in a fork
(`context: fork`), showed nothing while it ran: no chip, no tab, and
the badge and card didn't count it. No tool call starts it. Its
`task_started` carries no tool id, and its messages carry
`forked-command-<name>` and arrive "together and in order after the
forked skill finishes"
(https://code.claude.com/docs/en/headless.md#follow-subagent-messages).
The spawn gate kept them in the journal.
That task start now makes the fork a spawn chip, `Agent: /<name>`, and
the skill's prompt opens its tab. The chip's id adds the task to the
CLI's, which repeats on every run. Until the turn ends, any
`forked-command-*` traffic feeds the chip, and the prompt's echo is
dropped once. The fork's notification comes before that replay, so it
only records the status. The chip settles just before the turn's Done,
or when the driver ends a stopped turn itself. After that the CLI's id
stays in the journal again, and a late frame can't reopen the chip.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Typed-fork-chip demo media for the PR (live stills)
* Claude subagents: a second typed fork settles the first
A turn runs one typed fork, but nothing enforced it: a second task start
before the first settled replaced it, and the first's chip ran on with
nothing left to settle it, counted in the badge for good.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Wing <44917547+wingleeio@users.noreply.github.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Fork: settle the running subagent chips a fork copies
A fork copies the source's entries as they are, so a subagent still
running there kept a `running` chip in the fork. No run in the fork ever
settles it: the chip and its lanes-card row spun forever (#1001).
The copy now settles it `failed`, like run end and the restart sweep
do. The chip keeps its ref, so it still opens the subagent's transcript.
The source keeps its own.
CI now runs the engine's `side_chats` tests, where the fork tests live.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Fork-settles-subagents demo media for the PR (live before/after stills)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: a fork settles copied questions and running spawn chips
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* docs: reflow the ForkSideChat paragraph
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
---------
Co-authored-by: Wing <44917547+wingleeio@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Claude subagents: count and open only this chat's spawns
The ● N badge counted every tagged stream the engine opened a sink for,
while the lanes card lists only the chat's spawn chips. Claude Code tags
a forked skill's traffic with its Skill call's id, the agents a subagent
starts with their own Agent call ids, and a skill passed as `/<name>`
with `forked-command-<name>`. None of them is a spawn chip of the chat,
so the badge could read 3 over an empty card (#999).
A sink (its own doc, one unit on the badge) now opens only for a spawn
chip of the chat (`is_chat_spawn`). Traffic tagged with any other id
stays in the journal.
A forked skill is the chat's own subagent, though. The Claude adapter
sets each Skill call aside until its task starts, which comes ahead of
all the fork's traffic, and re-decodes it then as a spawn named after
the skill (`Agent: /review`). Its traffic then has a card and a tab to
feed. An inline skill starts no task, and its result drops it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Claude subagents: forward a foreground subagent's text and thinking
Claude Code leaves the text and thinking blocks of a foreground subagent
(or a forked skill) out of the stream unless asked, so its tab showed
tool calls only (#1000). Runs now set CLAUDE_CODE_FORWARD_SUBAGENT_TEXT=1,
which "enables the same behavior" as `--forward-subagent-text`
(https://code.claude.com/docs/en/cli-reference). The variable rather than
the flag: a CLI older than 2.1.211 ignores it, but refuses an unknown flag.
The CLI then forwards subagents "at every nesting depth"
(https://code.claude.com/docs/en/headless). The spawn gate keeps the
agents a subagent starts in the journal, instead of badging the chat
with docs no card lists (#999).
The adapter already emitted tagged text blocks. Tagged thinking blocks
now become the subagent's reasoning, in block order. A background
subagent sends them unasked, so its tab gains its reasoning too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Forward-subagent-text demo media for the PR (live before/after stills)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* New project: type a folder's path, with ⇥ completion
Adding a project meant walking to its folder one crumb at a time, even
when its path was known. The trail can now be typed into, like a file
manager's address bar:
- ⌘L, the pen at the end of the trail, or a click on the trail's empty
end turns it into a path field: the open folder's path, spelled from ~
under home, selected.
- ⇥ completes the folder under the caret from the device's listing,
previewed as a ghost. ↵ opens the path and gives the trail back; ⌘↵
opens it and adds it as a project. Esc gives the trail back and keeps
New project open.
- A folder pasted from the file manager becomes the path.
- A path that doesn't open keeps the field and says why ("No folder at
~/…", "~/x isn't a folder"); the list below it doesn't change.
- Relative paths start from the open folder. On Windows devices \dir
and /dir start from the drive's root, D:dir from D:\, //nas/share is
a share; elsewhere a backslash escapes, as in a shell.
Tab and Shift+Tab now stay in the palette while it's open: the search's
⇥ used to move focus behind it as well.
The new-project-path-fixture example runs the production Shell against
an engine served over IPC, drives the flow with window events, checks
the palette and the engine's project registry, and writes the stills
and recording in docs/media/new-project-path.
* New project typed-path demo media for the PR (screenshots + recording)
The new-project-path-fixture example's output: eight stills (full window
and palette crop each), the MP4 recording, a GIF encoded from it, and
the run's 16 passing checks. README.md says how they were made.
Lead with the desktop downloads per platform and present the curl
installer as the headless (Linux server) option.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The dmg now opens as the classic drag-to-install window, drawn over the
landing page's ascii hands: scripts/dmg-background.py re-renders the two
hand density maps through the same glyph ramp, noise field and purple on
near-black, frozen at one instant and composed for a 660x400 window
(committed as 1x/2x pngs so CI never needs Pillow). package-macos.sh
builds the layout with dmgbuild — Applications symlink, icon positions,
hidpi background tiff, volume icon — writing the .DS_Store directly, so
no Finder scripting and it stays headless-CI safe. Notarization and the
auto-updater tarball are untouched.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The trailing titlebar strip (pane-header controls + expand/close) is
right-pinned by a flex_1 spacer in a row with 8px child gaps, and its
width is capped to the space left of row_left. The gaps sit outside the
strip's width, so whenever the cap binds — exactly the takeover state,
where the strip spans everything right of the sidebar — the row overflows
by one gap and the whole button cluster slides 8px right (user report:
buttons shift going into expanded mode).
Cap the strip at avail minus the row's gaps (one in takeover, two with
the title row present). Rig-verified at 2x: pre-fix the three icons sat
16 physical px right of their collapsed columns in takeover; post-fix
the columns are identical in both states.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Harness settings page, model prefetch, picker highlight fix
- Settings → Agents: enable/disable harnesses (t3code models-page rows);
only Claude Code and Codex ship enabled, a harness whose CLI is missing
on the device renders dimmed with an install hint and can't be enabled,
and the last enabled harness can't be switched off. The set persists in
ui-settings.json and filters the composer's agents rail.
- HarnessDescriptor gains an installed flag: a filesystem probe for the
agent's own CLI (claude/codex/grok/hermes/pi) — the npx adapter
fallback deliberately doesn't count. Lazy registry slots re-probe per
ListHarnesses call so a mid-session install shows up on reopen.
- Model prefetch: the catalog load now kicks LIST_MODELS for every
offered harness in parallel (plus the committed chat's), so switching
agents in the picker never shows an on-demand "Loading models…".
- Picker fix: the keyboard-nav highlight opens ON the selected model row
instead of always row 0, which read as a second active row.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Per-device harness enablement + device switcher on the Agents page
The enabled set moves off the viewer's ui-settings.json onto the ENGINE
(harness-prefs.json in its data dir) — per-device, like the CLI installs
it gates. HarnessDescriptor now carries enabled (Option, None from older
engines falls back to default-set membership), and a new forwardable
SetHarnessEnabled RPC flips it on the target device, replying with the
fresh catalog. The engine enforces the gates where the state lives: no
enabling a missing CLI, no disabling the last enabled harness.
The Agents page gains the Accounts-style page-header device switcher —
retargeting both the ListHarnesses probe and the toggles at any
registered device over the relay. The picker's offered_harnesses now
reads the catalog's own enabled flags, so a space on another device
follows THAT device's toggles automatically. The short-lived
crate::harness_prefs global and ui-settings field are gone.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Autolink bare URLs in transcript markdown
Agents paste naked PR/issue URLs constantly and they rendered as plain
text (user report: the PR link wasn't clickable). pulldown-cmark has no
option for GFM's autolink extension, so promote http(s):// spans into
link runs ourselves — after run-merging, since pulldown splits Text
events at would-be emphasis chars and a per-event scan truncates URLs
at every underscore. GFM-ish boundaries: no glued schemes, trailing
punctuation stays text, closing parens only join when balanced inside
the URL. Existing InteractiveText plumbing makes them clickable.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: wing-anara <wing@anara.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* Add Hermes and Pi agents over ACP
Two new AcpAgentSpec entries on the shared harness: Hermes Agent (Nous
Research, native `hermes acp` server) and the pi coding agent (pi.dev,
community pi-acp adapter pinned 0.0.33 with npx fallback). Registry slots
mirror the specs (turn-boundary steering; pi's minimal->max thinking
ladder, no effort surface on Hermes yet), brand marks land in the picker
rail/tabs/sidebar/settings, and COMET_HARNESS gains grok/hermes/pi arms
(grok was a latent gap). Wire coverage rides the existing table-driven
fake-ACP tests plus descriptor and login-shell-resolution mirrors.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ACP as source of truth for models; separate Traits dropdown
models() now runs a short-lived probe (the discover_commands pattern) and
reads the session/new response's first-class SessionModelState, falling
back to the model config option's choices — paseo's preference order. The
static catalogs survive as enrichment for matched ids and as the answer
when an agent advertises nothing (per-model effort ladders don't exist on
the ACP wire; discovered models inherit the probe session's thought_level
ladder unless a curated entry supplies one). Successful discovery caches
per harness instance; failures retry on the next picker open.
The traits picker un-merges from the harness/model popover into its own
composer chip + dropdown (t3code TraitsPicker arrangement): traits are
properties of the selected model, so they follow the pick instead of
sharing the browse surface. The chip hides entirely when a model has
neither a ladder nor options. Media re-captured: discovered Hermes list
(provider-tagged entries), the traits dropdown, both runs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Traits dropdown: t3code menu rows with Default badges
Replace the segmented-chip grid with t3code's TraitsPicker arrangement:
headed sections of menu rows (reasoning ladder + each model option),
hairline separators between sections, a Default badge on the section's
default choice, and the trailing check on the selected row — all reusing
the existing popover menu primitives. Menu stays open for multi-adjust.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Drop selected checkmarks; right-align the Default badge
Selected menu rows across the app (models, branches, spaces, traits,
account switcher) now carry the active wash alone — the trailing check
duplicated what the highlight already said. The traits Default badge
moves to the row end.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Ghost Default badge: bare muted text, no border or fill
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: wing-anara <wing@anara.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* ACP harness: Grok Build, slash commands, tool output + inline diffs
Adds an Agent Client Protocol (v1) harness over the shared JSON-RPC stdio
client (promoted from codex/rpc.rs), with xAI's Grok Build as the first
registered agent. Slash commands advertised by ACP agents surface in a
composer popup (file-mention pattern, new ListCommands RPC), and tool
results now carry capped output text and inline diffs end-to-end
(AgentEvent → doc parts → TS mirror → expandable transcript chips).
Decision record in docs/research/acp.md; claude/codex keep their custom
adapters (settings passthrough, installed-CLI coupling, steering
semantics). Demo media under docs/media/acp/ (fake-acp.sh fixture rig).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ACP follow-up: wire session config options, restyle tool detail blocks
Model + reasoning now apply through session/set_config_option against the
session's advertised configOptions (category model / thought_level, matched
to advertised value ids, skipped when current, never fatal). Grok gets its
real Low/Medium/High effort ladder in the picker and registry descriptor.
Detail blocks read like the changes pane now: full-width add/del washes in
the diff hues, brighter mono output text, and a chevron tile affordance on
expandable chips matching the group header. Demo media re-captured.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tool details: render diffs with the changes-pane component, outputs as code blocks
The chip detail blocks are no longer hand-rolled capped line rows. Diffs go
through changes::render_file_body — hunk headers, dual line-number gutters,
accent bars, row washes, and synchronous syntax tokens (payloads are
doc-capped, so no background pass needed) — built from the inline ToolDiff
via similar's grouped_ops with 3 context lines. Outputs render as a code
block: verbatim mono lines with indentation intact, 24-line counted-tail
cap. Fold heights stay analytic via changes::body_height. Fixture demo
content upgraded to a real rust hunk + cargo output; media re-captured.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tool details: fold the detail into the chip's card, unbroken guide rail
An expandable chip is now one card: the chip row is the card's header and
the output/diff body sits under a hairline inside the same border, instead
of a separate card floating below. The guide rail is drawn by the chip
column with no fixed height, so it stretches past open details — expanding
a tool no longer breaks the rail. Heights stay analytic (chip + separator
+ body). Media re-captured.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: wing-anara <wing@anara.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Two-device rig (local wrangler edge, mock-harness host device, app as
second device), identical scripted takes on main and on the fix: run
streaming from the remote host, edge sync paused mid-run, user sends
while stalled, sync resumes as a burst. Recorded with the send timeline
aligned across takes to ~10ms.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>