Baixar wmux 4.9.0 per a Windows

### Added

• A bot's profile shows the commands it ran or was refused, a switch to turn its commands off, and an Always allow button that asks you to confirm before a bot may run a program; Settings › Bots can turn the command reviewer off (#2541)

• **New workspace setup.** Choose which git-ignored config files every new
worktree of a project gets, and its setup command, from the app："New
workspace setup…" in the command palette, or Edit setup in the wmux.json
dialog. The `.env` family is ticked by default, the setup command is
suggested from the lockfile, and copying into bot and fan-out workspaces
stays off unless you turn it on. Saving writes `wmux.json` and approves
exactly what was written; a file with commands you have not approved yet is
sent to review first. (#2544)

• **A Preparing card for new worktrees.** A worktree you open (start from an
issue, Quick Launch, the Git page) is now prepared under a card at the top
of the sidebar that shows the copied files and how long the setup has run
against its 5-minute limit. If the setup fails, the card shows its last
output lines and offers Retry, Start anyway, or Discard. Discard removes the
worktree and the branch it made, and refuses if you changed anything in it.
Nothing starts and nothing is deleted without your answer. A failure, a
skipped setup or a missing file also shows a toast on any page. The first
worktree of a project with `.env` files and no copy settings offers to
choose files to copy. (#2545)

• **Add a whole bot team in one step.** The Bots page's "+" menu now has Add team, which asks only for the workspace and its repository. It adds the lead, research, developer, issues and reviewer bots the workspace doesn't have yet. Before, a workspace that already had a lead couldn't use Add team at all, and a default name that was already taken stopped the whole team. Roles that are already there are named instead of added, a taken name gets a number ("Developer 2"), and only a lead made here greets you. (#2551)

• **Archive for deleted bots.** A deleted bot's kept chat now goes under a folded Archive at the bottom of the Bots list instead of staying among your conversations. It still opens read-only, and Remove from Archive (asked in place) takes it off the list for good. Its messages stay saved, and what it wrote in group chats keeps its name. (#2551)

• **Bots can read from the services you signed in to.** A bot's research worker can now use a connection you signed in to in Plugins › Connectors (Notion, Google, Slack and the rest). Before this, a signed-in connection reached only your own terminals. The worker reaches the service through wmux, which keeps your sign-in to itself and lets the worker use only the tools you picked for that connection, and only when the service marks each of them read-only. Such a task carries no other connection and has no web search, no network for its commands and no repository, and it works with files only through its commands, which reach only its task folder. A bot starts, messages or resumes it only from a turn, in a conversation, that has taken in no outside text (your own message, for example); you can always resume it yourself. If the sign-in changes or wmux restarts, the bot tells you, and the worker starts again in a new pane once you resume it. macOS and Linux only for now (#2552)

• An HTML file a bot attaches now opens rendered in the Bots side panel, with Preview / Code to see its source, where before it showed only as code. An ```html block in a bot's message gets the same toggle (Code first). The page is drawn in a sandboxed frame：its scripts don't run, and it can't load anything from the network or reach the app (#2553)

• **Bot recipes.** Add bot now has a Recipe select with fifteen ready-made
jobs. Each one reads a service you connect (Notion, Figma, Canva, Linear,
Atlassian, Asana, GitHub, Gmail, Google Calendar, Sheets, Drive or Slack)
and reports in the chat. Anything to send or change comes to you as a
draft. A recipe connects nothing by itself. The bot's profile lists each
service the recipe needs, what is still left to set up, and a link to
Plugins › Connectors. Before this, you wrote each bot's job from scratch
and worked out its connections yourself.

• **Apply a bot's code changes to your own checkout.** A dev worker's
changes used to stay in its own worktree on its task branch, and getting
them into your folder meant a pull request and a merge, or typing git
commands by hand. A code task that reported or finished now has *Apply to
my checkout* on its card. It applies the task's changes to the bot's
repository folder as unstaged changes, including the files the task added
and removed, and keeps anything you pulled since the task started. A line
then names the files and tells you how to undo them with `git restore`.
Your branch, HEAD and staged changes are left alone. Nothing is written
when a file the task touched has uncommitted changes (they are listed),
when your copy of those files changed so the task's changes no longer
apply cleanly, when your checkout runs those files through a content
filter such as Git LFS, or when a merge or rebase is in progress.

• **A quiet way to star wmux on GitHub.** After you have used wmux for a while (at least three days, and 20 agent sessions started in panes), one muted row at the foot of the sidebar reads "Enjoying wmux? · Star on GitHub". If the GitHub CLI (`gh`) is signed in, Star on GitHub stars the repository in one click and says thank you; otherwise it opens the repository page. Either way the row never comes back. Not now hides it for at least three days and doubles the number of sessions before it asks again. It never shows while "Agent not set up" does, and if `gh` already shows wmux as starred it never shows at all. Settings › About keeps a Star wmux on GitHub row for any time. Everything is counted on your machine; nothing is reported anywhere (#2565)

• **Bots work on GitLab repositories.** A bot in a GitLab repository can now
read merge requests, issues and CI, give a verdict, and merge an approved
merge request through `glab`, the way it already did on GitHub through `gh`.
The merge is refused if the head moved since approval, the merge request is
not open, or its pipeline is not green on that head; a merge request that
changes CI configuration is never merged on its own pipeline's word. When
`glab` is missing or not signed in, you get one "Connect GitLab" card
instead of bots failing on every read. Before, bots in a GitLab repository
could not read or merge anything there.

• **Bots can read commits in their repository.** A bot with a repository can now read a file at any commit (`git_show`) and see what a branch changed (`git_diff`). A reviewer can read a worker's commit as soon as it exists instead of waiting for a report. Both are read-only, protected files such as `.env` are never shown, and neither is available in a turn started by an outside guest's message.

• **Agents can send images with their messages, to a pane on this PC or on another one.** `send_message` takes up to four PNG, JPEG or WebP files (5 MB each, 10 MB per message) from the pane's project folder or `~/.wmux/uploads/`. The receiving PC keeps its own copy in `~/.wmux/uploads/a2a/` for 7 days. The receiving agent's message ends with the path of each image, so it can open them. Metadata such as a photo's location is removed before an image leaves, and only real, still image files are accepted, whatever their name says (an animated image is refused). Cross-PC messages carried text only before. A PC whose wmux is older gets the text with a note that images were left out. (#2569, #2571, #2572)

• **Bot triggers fire on GitLab repositories.** Bots watching a GitLab
repository now wake on merge requests opened, updated, labelled, merged or
closed, on pipelines turning green or red, and on new or labelled issues,
and an approved merge lane merges a GitLab merge request once its pipeline
is green. Before, the watch only read GitHub, so nothing fired for GitLab.
Trusted authors on GitLab are written `host/username`; a GitHub login in
the list no longer counts as a GitLab user of the same name.

• **Bots open GitLab merge requests.** When a worker's task is done in a
GitLab repository, `task_pr` now pushes the branch and opens a merge
request with `glab`, returning the existing one instead of a duplicate.
The approval dialog names the host and project it will push to. Before,
`task_pr` only knew `gh` and you had to open the merge request yourself.

• **Settings → Browser → Allow private network addresses.** Browser navigation can now reach intranet sites on private ranges (10.x, 172.16–31.x, 192.168.x and IPv6 fc00::/7) once you turn this on, with no restart and no environment variable. It is off by default. Link-local and cloud metadata addresses (169.254.x, fe80::), 0.0.0.0 and loopback other than 127.0.0.1 stay blocked either way. `WMUX_ALLOW_PRIVATE_NETWORK=1` still works as before.

### Changed

• **After a crash, programs started from a terminal close with it (Windows).** With the default Node terminal engine, if the background service crashes or is force-closed, the shells it ran and anything started from them now end with it instead of staying behind unseen and holding memory. The same goes for the helpers the service itself starts, such as chat agents and git commands. A normal quit or an update leaves programs you started on purpose (for example with `start`) running, as before, and the shared Codex background server, which detaches itself from wmux (as Codex 0.162 does), keeps running either way. (#2398)

• **The Git page no longer waits for a worktree's setup.** The worktree is
created at once and its setup runs on its Preparing card. (#2545)

• **Starting an issue whose branch is held by a kept worktree** now names that
worktree and how to remove it, in