tabs: agent sessions share one window instead of opening one each

A session's first tab used chrome.windows.create, so every new session name (and
every Claude Code process) got its own window. It now goes into the window that
already holds any 'Pi Session:' group; a window is created only when none exists.
chrome_tab new for a session with no group also uses that window, not the user's
focused one. The primer and the chrome_tab description say so.

2 files changed, 34 insertions, 5 deletions

diff --git a/extensions/chrome-profile-bridge/browser-extension/service_worker.js b/extensions/chrome-profile-bridge/browser-extension/service_worker.js
index c62aff9..7fbbbba 100644
--- a/extensions/chrome-profile-bridge/browser-extension/service_worker.js
+++ b/extensions/chrome-profile-bridge/browser-extension/service_worker.js
@@ -226,11 +226,26 @@ function isPiChromeOwnedTarget(tabId, sessionKey) {
   return false;
 }
 
+// The window agents share. Every session has its own tab group and a group cannot span windows,
+// but sessions do not need their own windows: new tabs go to the window that already holds any
+// "Pi Session:" group, so agents stay together instead of opening one window each. Returns
+// undefined when no such window exists (the first session then creates one).
+async function sharedPiWindowId() {
+  if (!chrome.tabGroups || typeof chrome.windows?.get !== "function") return undefined;
+  const groups = await chrome.tabGroups.query({}).catch(() => []);
+  for (const group of groups) {
+    if (!/^pi session:/i.test((group.title || "").trim()) || typeof group.windowId !== "number") continue;
+    const win = await chrome.windows.get(group.windowId).catch(() => null);
+    if (win && (win.type === undefined || win.type === "normal")) return win.id ?? group.windowId;
+  }
+  return undefined;
+}
+
 // Create a fresh automation target for `sessionKey`. If this session already has a tab group,
 // create the tab inside that group's window so one Pi session keeps one Chrome tab group (Chrome
-// groups cannot span windows). If no group exists yet, prefer an isolated window; fall back to a
-// tab. When the tab is created in a pre-existing group window, leave windowId unset so cleanup only
-// closes our tab, never that whole window.
+// groups cannot span windows). If it has none, use the shared Pi window; only when there is no
+// Pi window yet, create one; fall back to a tab. When the tab is created in a pre-existing window,
+// leave windowId unset so cleanup only closes our tab, never that whole window.
 async function createAutomationTarget(sessionKey, groupTitle) {
   const existingGroup = groupTitle ? await findGroupRecordByTitle(groupTitle) : null;
   if (existingGroup && typeof existingGroup.windowId === "number") {
@@ -239,6 +254,14 @@ async function createAutomationTarget(sessionKey, groupTitle) {
     await persistAutomationTargets();
     return tab;
   }
+  const sharedWindowId = await sharedPiWindowId();
+  if (sharedWindowId !== undefined) {
+    // Someone else's window: leave windowId unset so cleanup only ever closes our tab.
+    const tab = await chrome.tabs.create({ url: "about:blank", active: false, windowId: sharedWindowId });
+    automationTargets.set(sessionKey, { windowId: undefined, tabId: typeof tab.id === "number" ? tab.id : undefined });
+    await persistAutomationTargets();
+    return tab;
+  }
   if (chrome.windows && typeof chrome.windows.create === "function") {
     try {
       const win = await chrome.windows.create({ url: "about:blank", focused: false });
@@ -1825,6 +1848,11 @@ async function dispatch(action, params) {
       const existingGroup = await findGroupRecordByTitle(groupTitle);
       const createParams = { url: params.url || "about:blank", active: foregroundRequested(params) };
       if (existingGroup && typeof existingGroup.windowId === "number") createParams.windowId = existingGroup.windowId;
+      else {
+        // No group of this session yet: use the shared Pi window, never the user's focused one.
+        const sharedWindowId = await sharedPiWindowId();
+        if (sharedWindowId !== undefined) createParams.windowId = sharedWindowId;
+      }
       // Register the creation as in flight until the tab is tracked, so early capture (which only
       // runs in tracked tabs) waits for it instead of skipping the new tab's first page load.
       const creation = (async () => {
diff --git a/extensions/chrome-profile-bridge/index.ts b/extensions/chrome-profile-bridge/index.ts
index 90f080e..32ae79f 100644
--- a/extensions/chrome-profile-bridge/index.ts
+++ b/extensions/chrome-profile-bridge/index.ts
@@ -1276,7 +1276,8 @@ Usage rules:
 6. /chrome background on (default) is a hard policy: per-call \`background=false\` cannot override it, new tabs stay inactive, and \`chrome_tab activate\` is blocked. Ask the user to run /chrome background off when they want foreground/watch mode. With background off, per-call \`background=true\` still avoids explicit focus/tab activation. Screenshots use CDP without activating background tabs; debugger failures never fall back to tab activation. Page scripts, trusted input, native prompts, and Chrome/OS behavior can still affect focus.
 7. If you hit a native file-picker or privileged browser prompt gate, tell the user; generic clicks/typing/CSP gates are handled by Chrome input.
 8. Run /chrome doctor when in doubt about connectivity or capabilities.
-9. A page can open dialogs you cannot see: alert, confirm, prompt, and the \"Leave site? Changes you made may not be saved\" prompt of a page with unsaved changes. They are dismissed for you (stay, cancel). \`chrome_navigate\` reports one by failing, and \`leaveIfAsked: true\` leaves anyway; use that only when you know nothing unsaved matters, and ask the user when unsure. If a click or key press that should have navigated did nothing, a Leave-site prompt may have stopped it: check the page, then ask the user before discarding anything.`;
+9. A page can open dialogs you cannot see: alert, confirm, prompt, and the \"Leave site? Changes you made may not be saved\" prompt of a page with unsaved changes. They are dismissed for you (stay, cancel). \`chrome_navigate\` reports one by failing, and \`leaveIfAsked: true\` leaves anyway; use that only when you know nothing unsaved matters, and ask the user when unsure. If a click or key press that should have navigated did nothing, a Leave-site prompt may have stopped it: check the page, then ask the user before discarding anything.
+10. All agent sessions share one browser window; your tabs sit in your own colored tab group inside it. Reuse a tab you already have (navigate it) rather than opening new ones, and open a new tab only when you need two pages at once. Do not open new windows unless the user asks.`;
 
 	// Shared handlers, dispatched by the unified /chrome command below.
 	const doctorHandler = async (ctx: ExtensionContext) => {
@@ -1651,7 +1652,7 @@ Usage rules:
 	pi.registerTool({
 		name: "chrome_tab",
 		label: "Chrome Tab",
-		description: "List, create, activate, close, group, ungroup, or inspect tabs in the user's existing Chrome profile via the companion extension. New/grouped tabs always use this session's Pi tab group. Background mode keeps new tabs inactive and blocks activate; ask the user to run /chrome background off for foreground/watch mode. activate/close/group/ungroup require a target (targetId/urlIncludes/titleIncludes); with no target they act on this session's pi-chrome automation tab if one exists, and otherwise error rather than touching the user's active tab.",
+		description: "List, create, activate, close, group, ungroup, or inspect tabs in the user's existing Chrome profile via the companion extension. New/grouped tabs always use this session's Pi tab group, inside the one window all agent sessions share; prefer navigating a tab you already have to opening another. Background mode keeps new tabs inactive and blocks activate; ask the user to run /chrome background off for foreground/watch mode. activate/close/group/ungroup require a target (targetId/urlIncludes/titleIncludes); with no target they act on this session's pi-chrome automation tab if one exists, and otherwise error rather than touching the user's active tab.",
 		promptSnippet: "List/open/activate/close/group existing Chrome tabs through the companion extension.",
 		parameters: Type.Object({
 			action: StringEnum(tabActionValues),