Astro 7.3.2 lets crafted input inject CSS through View Transitions; the
site doesn't use them, but 7.3.3 is a patch release. `npm audit fix`
takes zod and fast-uri past their advisories too, inside their ranges.
undici 7.29.1 under wrangler's miniflare stays: npm's only fix moves
wrangler back to 4.35.0, and it's deploy tooling that nothing shipped
contains.
The comet above the composer sweeps in plan's hue while the session
plans, from a second trail shaded onto the same rail. It reads the mode
the engine reports, never the keypress: a mid-turn switch lands in the
engine at once and each tool call is judged against it as it starts, so
the color changes when what the turn can do changes.
derived_activity_shades moves to styles/shades.rs beside the ramps it
builds, which keeps styles.rs under 500 lines. CHANGELOG entries for
this and the git checks -R fix.
`checks` read only its first argument as a pull request number, so
`["19", "-R", "owner/name"]` dropped the flag and answered with this
checkout's #19. It now reads `-R`/`--repo`/`--repo=` as `gh` does
(owner/name, or host/owner/name), reads that repository's pull request
and its checks, and names the repository in the answer. `-R` without a
number is refused, since the branch lookup only knows this checkout, and
so is any flag or second number rather than being skipped.
The phone app's WebKit e2e also fetched its system libraries with `playwright install
--with-deps`, and a slow mirror ran that job out of time too. It now runs in Playwright's image
through `docker run`, with the workspace mounted at its own path so the test finds the `luxe` the
job built, after removing the runner's unused Android SDK to make room for the image. Both images
take their tag from the suite's lockfile, read in `what changed`, so a Dependabot bump of
Playwright moves its image with it.
In the image, site-cross's dev server came up on localhost and refused every probe: the server
and the client resolved localhost to different loopbacks. Both servers now bind 127.0.0.1 and
everything dials it. site-cross also reports why a dev server never answered: what the last probe
got and the server's last lines of output.
`playwright install --with-deps` fetched about 120 MB of Ubuntu libraries from the runner's mirror
on every run, and a slow mirror held the web build past its 15-minute timeout four runs in a row.
Playwright's own image already carries the three browsers and those libraries, so the cross-browser
steps move to a job inside it. The render check stays on the runner, since it drives the system
Chrome the image doesn't have. A step fails with the fix spelled out if web/'s Playwright and the
image's tag drift apart.
cargo-deny fails on a yanked crate, and 5.4.1 was yanked after 0.1.6 merged. 5.4.2 fixes an
out-of-bounds read in get_html when an HTML fragment's start offset is bad. luxe reads only text
and images from the clipboard, so it never reached that code, but the changelog says so.
On macOS, screenshot's window mode captures one window, even one on another desktop. The window
list comes from CoreGraphics through osascript, and the chosen id goes to `screencapture -l`, so
nothing needs installing. A title matches a window's title first and its app's name second.
On Windows, a paste no longer submits at its first line break: the console types a paste out as
key presses, so a burst of text keys is folded back into one paste. Screenshots on a scaled display
capture the whole screen, because the capture script now asks to be DPI-aware first.
On a Mac with fontconfig, the font probe ignores .LastResort, so the OpenAI and Claude marks fall
back to the robot and cloud when no font has them.
The dock's provider row opens a list of providers, and the model picker opens on the provider in use, with tab to widen it; signing in opens it on the provider just signed in to. Models go by their id unless config names them, and the model row wears its maker's mark.
The guided review's guide prompt is shorter, the pull request brief shows the start of the description with the rest behind p, its commits drop the Luxe-review and Luxe-revision lines, and the weakened-gates signal is gone.
luxe update reruns the installer it was installed with, in place of luxe-cli-update. The preview's FILE tab reads files the session changed outside the project, and only those. Smaller fixes: bash headers are highlighted, the preview loses a blank row, the launcher divider sits inside its frame, and the phone app's top bar is clear of iOS's blur.
rustls-webpki 0.103.16 and webpki-roots 1.0.10 fix RUSTSEC-2026-0381 and RUSTSEC-2026-0382, and vite 8.3.4 fixes GHSA-587h-7wmj-h5pm, GHSA-698v-4r3x-3vcv and GHSA-r825-w8mm-5pxw.
- luxe starts on the model you last chose in this project, then the last one anywhere. Signing in
opens the model picker instead of choosing for you.
- Every list in a modal takes the mouse wheel and a draggable scrollbar.
- The phone app shows reasoning in full, stays put while you read back through a long session, and
shows a question's answer on its card.
- On Windows, luxe update replaces the running luxe, and a signed-in install no longer sees setup
again.
- The file tree no longer shows "engine busy" in a large project, loop output is easier to follow,
and the name is written lowercase.
The CHANGELOG has the full list.
- luxe remote puts the sessions on your computer in reach of your own phone or tablet over
Tailscale: write, steer, answer prompts, send photos, browse files, diffs and GitHub, with a
paired device and your own Tailscale login.
- luxe join lets another program on the same machine join a running session over the RPC protocol.
- An AGENTS.md in a subfolder reaches the model the first time it works there.
- Long sessions stay fast: no stall before each model call, lighter memory, cheaper drawing.
- A session is open in one Luxe at a time, web_search shares results across its phrasings, and a
grep glob starting with ! excludes.
The CHANGELOG has the full list.
- grep groups its lines under each file, the layout of rg --heading, rather than repeating the
path on every line: 9–28% fewer bytes than rg -n on the searches measured. scopes: true labels
each run of matches with the function or type around it.
- The changes pane's header shortens the path rather than cut its position, and keeps a gap
before the FILE / DIFF switch.
- The bash teardown tests count a loop still running as a survivor, not one write already under
way when the kill landed.
- The README's wordmark is an animated GIF with a transparent background, rendered from the
dashboard's own art and lights, in light and dark variants for GitHub's two themes.
- The README says Luxe is built with AI assistance, much of it Luxe working on itself, and
CONTRIBUTING has an AI-assistance section: say what helped, understand the change, write to
people in your own words.
- The redaction table's Google-key example is joined at compile time, so secret scanning stops
reading the invented value as a leaked key.
- The MCP trust tests read Luxe's own report of what it started, so a slow runner can neither fail
them nor pass them falsely.
- luxe.dev/install.sh redirects to the latest release's installer, and luxe.dev/install.ps1 runs
the release's PowerShell installer under Windows' default policy, so the commands are
`curl -fsSL https://luxe.dev/install.sh | sh` and `powershell -c "irm https://luxe.dev/install.ps1 | iex"`.
The install tabs run shell, brew, cargo, then windows.
- The edit and write tools and format-on-save keep a file's mode, so editing a script no longer
leaves it unrunnable.
- The tools page gives Claude Code and Codex their own columns and bars, draws Luxe's own bar in
each group, and its terminal's window controls take macOS's colors.
- scripts/aur-pkgbuild.sh writes a luxe-bin PKGBUILD from a release, for the AUR once it reopens.
- On Windows a lock directory being removed answers access denied, not already exists, so a
second writer, such as two MCP servers signing in at once, failed instead of waiting.
- The header is Docs, Tools and the GitHub mark, one row down to 320px; /themes folds into the
homepage's gallery and redirects there.
- The tools page has a terminal of measured rows beside its intro and a chart of whole-task cost
against Luxe's 1x beside its table; Pi leaves the comparison.
- The README opens on a generated wordmark banner that carries its own colors.
- release.yml, generated by dist from dist-workspace.toml, builds macOS, Linux and Windows,
attests every artifact, and publishes shell and PowerShell installers with their updater,
a luxe formula in evanwhittaker/homebrew-tap, and the crates through Trusted Publishing
(publish-crates.yml, which skips versions crates.io already has).
- luxe update recognizes an installer copy by luxe-cli-update beside it and runs that.
- The README, the install guide and the site lead with the installer, Homebrew and Windows.
- SECURITY.md: report privately through a GitHub advisory or contact@luxe.dev.
- luxe update and /update update Luxe with the command for however it was installed (cargo or
Homebrew), run one program at a time with the terminal attached, then check that the luxe on
PATH is the new version. Luxe never downloads or replaces its own binary. The start screen says
"Update available · /update".
- A question's typed answer edits at its caret: arrows, word jumps, Home, End and Delete work,
and up and down move through a wrapped answer before leaving it.
- The version check owes a bump once per release: the version on main is the next release, and
only a released one (a v-tag) has to move.
- Dependency updates arrive through the sync, so Dependabot and the weekly jobs no longer run here.
A path filter on the trigger meant a docs-only pull request never got the checks a ruleset would require, so it could never merge. scripts/ci-changes.py decides from the diff whether the jobs have work, and CI passed sums them up, since a skipped matrix job reports under no name a ruleset can hold. The old filter also skipped crates/luxe-cli/guide.md, which the CLI compiles in, and docs/guide/, which the suite checks against the code; both now run the jobs.