desktop/kubuntu: working screenshots, no screen lock, Chrome for agents

desktop/packaging/kubuntu/Dockerfile

@@ -35,8 +35,17 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
       kwin-wayland plasma-desktop plasma-workspace konsole breeze breeze-cursor-theme libkscreen-bin \
       xdg-desktop-portal xdg-desktop-portal-kde pipewire wireplumber dbus-daemon flatpak \
       libegl-mesa0 libgl1-mesa-dri libpipewire-0.3-0t64 libjpeg-turbo8 libpixman-1-0 libpam0g \
-      fonts-noto-core ca-certificates curl tini procps \
+      fonts-noto-core fonts-noto-cjk fonts-noto-color-emoji ca-certificates curl gnupg tini procps \
     && rm -rf /var/lib/apt/lists/*
+# A browser for agents: Google Chrome from Google's signed repository (Ubuntu's Chromium is a snap,
+# and snaps do not run in a container).
+RUN install -d -m 0755 /etc/apt/keyrings \
+    && curl -fsSL https://dl.google.com/linux/linux_signing_key.pub | gpg --dearmor -o /etc/apt/keyrings/google-chrome.gpg \
+    && echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/google-chrome.gpg] https://dl.google.com/linux/chrome/deb/ stable main" \
+       > /etc/apt/sources.list.d/google-chrome.list \
+    && apt-get update && apt-get install -y --no-install-recommends google-chrome-stable \
+    && rm -rf /var/lib/apt/lists/*
+COPY browser.sh /usr/local/bin/w3bs-browser
 # KWin asks for real-time scheduling, which a container's capability set does not grant.
 RUN setcap -r /usr/bin/kwin_wayland 2>/dev/null || true
 COPY --from=w0vncserver /usr/local/bin/w0vncserver /usr/local/bin/w0vncserver
@@ -48,6 +57,6 @@ RUN arch="$(uname -m)" \
     && tar xzf "$file" ./bin/w3bs-desktop && install -m 0755 bin/w3bs-desktop /usr/local/bin/ \
     && rm -rf /tmp/*
 COPY session.sh /usr/local/bin/w3bs-kubuntu-session
-RUN chmod 0755 /usr/local/bin/w3bs-kubuntu-session && mkdir -p /vnc /home/desktop && chmod 1777 /vnc /home/desktop
+RUN chmod 0755 /usr/local/bin/w3bs-kubuntu-session /usr/local/bin/w3bs-browser && mkdir -p /vnc /home/desktop && chmod 1777 /vnc /home/desktop
 ENV HOME=/home/desktop
 ENTRYPOINT ["/usr/bin/tini", "--", "/usr/local/bin/w3bs-kubuntu-session"]

desktop/packaging/kubuntu/browser.sh

@@ -0,0 +1,7 @@
+#!/bin/sh
+# Open Google Chrome in the agent desktop: `w3bs-browser [URL]`.
+# --no-sandbox: Chrome's sandbox needs user namespaces, which this container is not given; the
+# container is the boundary here. A separate profile keeps agent sign-ins out of anything else.
+exec google-chrome --no-sandbox --ozone-platform=wayland --no-first-run --no-default-browser-check \
+  --password-store=basic --hide-crash-restore-bubble --start-maximized \
+  --user-data-dir="${W3BS_BROWSER_PROFILE:-$HOME/.config/w3bs-browser}" "$@"

desktop/packaging/kubuntu/session.sh

@@ -15,6 +15,12 @@ export SHELL=/bin/bash
 LOGS=$HOME/.local/state/w3bs-desktop/logs
 mkdir -p -m 0700 "$XDG_RUNTIME_DIR" && mkdir -p "$LOGS" "$XDG_CONFIG_HOME/xdg-desktop-portal"
 printf '[preferred]\ndefault=kde;gtk\n' > "$XDG_CONFIG_HOME/xdg-desktop-portal/kde-portals.conf"
+# The KDE screenshot portal writes the image under Pictures and answers "response 2" when the
+# folder does not exist; a fresh home has none.
+mkdir -p "$HOME/Pictures"
+# Never lock: nobody is at this screen to unlock it, and a locked session makes the desktop
+# layer disable agent access ("session locked").
+printf '[Daemon]\nAutolock=false\nLockOnResume=false\nTimeout=0\n' > "$XDG_CONFIG_HOME/kscreenlockerrc"
 rm -f /vnc/vnc.sock
 
 say() { echo "w3bs-kubuntu: $*"; }
@@ -40,6 +46,9 @@ wireplumber >"$LOGS/wireplumber.log" 2>&1 &
 # screen. Everything in this compositor is part of one agent desktop that the desktop layer
 # already lets agents see and drive, so turn the check off here; it never touches the host.
 export KWIN_WAYLAND_NO_PERMISSION_CHECKS=1
+# Screenshots are a separate KWin plugin with its own check (ScreenShot2 answers
+# NoAuthorized to the KDE portal here), so the desktop layer's screen capture needs it too.
+export KWIN_SCREENSHOT_NO_PERMISSION_CHECKS=1
 kbuildsycoca6 >"$LOGS/kbuildsycoca.log" 2>&1 || true
 
 # KWin on the virtual display; with no GPU, Mesa renders OpenGL on the CPU.

package-manager/crates/w3bs-cli/src/desktop.rs

@@ -23,6 +23,7 @@ pub const IMAGE: &str = "w3bs-kubuntu-desktop";
 pub const MCP: &str = "127.0.0.1:7802";
 const DOCKERFILE: &str = include_str!("../../../../desktop/packaging/kubuntu/Dockerfile");
 const SESSION: &str = include_str!("../../../../desktop/packaging/kubuntu/session.sh");
+const BROWSER: &str = include_str!("../../../../desktop/packaging/kubuntu/browser.sh");
 const START_TIMEOUT: Duration = Duration::from_secs(120);
 
 fn fail(code: &'static str, msg: impl Into<String>) -> Error {
@@ -50,6 +51,8 @@ pub fn image_tag() -> String {
     h.update(DOCKERFILE);
     h.update([0]);
     h.update(SESSION);
+    h.update([0]);
+    h.update(BROWSER);
     let hex: String = h.finalize().iter().take(6).map(|b| format!("{b:02x}")).collect();
     format!("{IMAGE}:{hex}")
 }
@@ -139,7 +142,9 @@ fn gid_of(_: &Path) -> Option<u32> {
 fn build_image(tag: &str) -> Result<()> {
     let ctx = data_dir()?.join("build");
     std::fs::create_dir_all(&ctx).map_err(|e| fail("DESKTOP_BUILD", format!("{}: {e}", ctx.display())))?;
-    std::fs::write(ctx.join("Dockerfile"), DOCKERFILE).and_then(|_| std::fs::write(ctx.join("session.sh"), SESSION)).map_err(|e| fail("DESKTOP_BUILD", e.to_string()))?;
+    std::fs::write(ctx.join("Dockerfile"), DOCKERFILE)
+        .and_then(|_| std::fs::write(ctx.join("session.sh"), SESSION))
+        .and_then(|_| std::fs::write(ctx.join("browser.sh"), BROWSER)).map_err(|e| fail("DESKTOP_BUILD", e.to_string()))?;
     eprintln!("w3bs: building the Kubuntu desktop image {tag} (first time only, a few minutes)");
     let ok = Command::new("docker")
         .args(["build", "--quiet", "-t", tag, "-t", &format!("{IMAGE}:latest")])
@@ -310,5 +315,8 @@ mod tests {
         assert_eq!(t.len(), "w3bs-kubuntu-desktop:".len() + 12);
         assert_eq!(t, image_tag());
         assert!(DOCKERFILE.contains("w0vncserver") && SESSION.contains("/vnc/vnc.sock"));
+        // Screen capture needs KWin's screenshot check off and a Pictures folder; agents need a browser.
+        assert!(SESSION.contains("KWIN_SCREENSHOT_NO_PERMISSION_CHECKS=1") && SESSION.contains("Pictures"));
+        assert!(DOCKERFILE.contains("google-chrome-stable") && DOCKERFILE.contains("w3bs-browser") && BROWSER.contains("google-chrome"));
     }
 }