pi-chrome 0.15.28 → 0.15.30

package/CHANGELOG.md

@@ -2,6 +2,24 @@
 
 All notable user-facing changes to `pi-chrome`.
 
+## 0.15.30 — 2026-05-31
+
+Tab grouping for `chrome_tab`.
+
+- **Pi-opened tabs auto-group.** `action=new` now drops every tab into a shared `Pi` tab group per window by default (created once, then reused), so agent tabs stay visually separated from your own. Opt out per call with `groupTitle:""` or `group:false`.
+- **`chrome_tab` can group/ungroup tabs.** New `action=group` (and `action=ungroup`) plus `groupTitle`/`groupColor` params. Grouping reuses an existing same-title group in the window instead of spawning duplicates. Defaults: title `Pi`, color `blue`; color validated against Chrome's 9 group colors. Target an existing tab with `targetId`/`urlIncludes`/`titleIncludes`.
+- **Tab listings include group info.** `formatTab` now reports `groupId` and a `group` record (`title`, `color`, `collapsed`, `windowId`, `piGroup`), and `chrome_tab list` prefixes grouped tabs with `[Group Title]`.
+- Requires the new `tabGroups` extension permission — reload the companion extension after updating.
+
+## 0.15.29 — 2026-05-31
+
+Strict-CSP support: `chrome_evaluate`, `chrome_snapshot`, `chrome_wait_for`, and `chrome_navigate initScript` now work on pages that block `unsafe-eval`.
+
+- **CDP-based evaluation bypasses page CSP.** `chrome_evaluate`/`chrome_snapshot` (and all snapshot-driven inspection) previously ran user code in the page MAIN world via the **Function constructor**, which is blocked by `script-src 'self'` without `'unsafe-eval'` — so they returned null/empty (or `EvalError`) on github.com and many bank/SaaS apps. They now evaluate through CDP `Runtime.evaluate`, a DevTools protocol command that is not subject to the page's Content-Security-Policy. Rich return values (undefined/function/symbol/bigint/Error markers, DOMRect expansion) and the expression/statement fallback are preserved.
+- **`chrome_wait_for` polls via CDP.** The selector/expression polling loop moved from in-page `new Function()` to service-worker-side CDP evaluation, so waits work under strict CSP too.
+- **`chrome_navigate initScript` injects via CDP.** Document-start init scripts now register with `Page.addScriptToEvaluateOnNewDocument` instead of `new Function()` on `webNavigation.onCommitted`, so seeding localStorage / stubbing `Date.now` works under strict CSP.
+- **Tests.** Added a Node unit harness (`test-suite/unit/csp-eval.test.mjs`, run via `npm test`) validating the evaluate/execute/waitFor refactor, and an in-browser regression challenge (42 `strict-csp-evaluate`) that reads a JS-only secret under strict CSP. Updated challenge 39's notes and docs (FAQ, EXAMPLES, COMPARISON, primer) which previously stated eval/snapshot fail under strict CSP.
+
 ## 0.15.28 — 2026-05-31
 
 Low-risk reliability fixes from a long-session bug report.

package/README.md

@@ -217,7 +217,7 @@ Multiple Pi sessions (planner / worker / audit) can all drive the same Chrome at
 
 ## Built-in benchmark suite
 
-[`test-suite/`](./test-suite) is a benchmark for **any** browser-control agent (not just pi-chrome). It includes **41 primitive challenges** plus **4 hermetic BrowserGym-style long-horizon tasks**.
+[`test-suite/`](./test-suite) is a benchmark for **any** browser-control agent (not just pi-chrome). It includes **42 primitive challenges** plus **4 hermetic BrowserGym-style long-horizon tasks**.
 
 Scoring tracks expected outcomes per challenge rather than raw PASS count, so tools are judged against their declared browser-control capability. Unit challenges are split into gate buckets:
 

package/docs/COMPARISON.md

@@ -134,7 +134,7 @@ If your threat model excludes extensions with broad permissions, neither approac
 
 ## Public benchmarks worth knowing (for axis 2 / axis 3 comparison)
 
-Pi-chrome itself ships a benchmark suite ([`../test-suite/`](../test-suite)) of **41 primitive challenges** plus **4 hermetic BrowserGym-style long-horizon tasks** covering trusted input, pointer humanization, keyboard fidelity, drag/drop, Shadow DOM, iframes, file uploads, strict-CSP screenshot fallback, dynamic waits, tab lifecycle, network observability, fingerprint leaks, and agent-safety honeypots. Scoring tracks expected outcomes per challenge instead of raw PASS count, with `core`, `conditional`, and `quality` gate buckets. That's **driver-level** grading.
+Pi-chrome itself ships a benchmark suite ([`../test-suite/`](../test-suite)) of **42 primitive challenges** plus **4 hermetic BrowserGym-style long-horizon tasks** covering trusted input, pointer humanization, keyboard fidelity, drag/drop, Shadow DOM, iframes, file uploads, strict-CSP screenshot fallback and CDP eval/snapshot bypass, dynamic waits, tab lifecycle, network observability, fingerprint leaks, and agent-safety honeypots. Scoring tracks expected outcomes per challenge instead of raw PASS count, with `core`, `conditional`, and `quality` gate buckets. That's **driver-level** grading.
 
 For **agent-level** comparison (axis 2), the public benchmarks worth citing:
 

package/docs/EXAMPLES.md

@@ -161,6 +161,6 @@ Interactive tools use Chrome's real input layer by default: clicks, typing, fill
 - fullscreen and other user-activation checks
 - pages where DOM injection/evaluate is limited, if the agent can use screenshots + coordinates
 
-Strict CSP note: `chrome_snapshot`/`chrome_evaluate` may be blocked on pages that disallow `unsafe-eval`; `chrome_screenshot`, tab/navigation tools, and real input still work.
+Strict CSP note: `chrome_snapshot`/`chrome_evaluate` work even on pages that disallow `unsafe-eval`, because they run via CDP `Runtime.evaluate` (not page-level `eval`/`new Function`), which is not subject to page CSP. `chrome_screenshot`, tab/navigation tools, and real input also work under any CSP.
 
 Chrome may show its debugger banner while pi-chrome is attached.

package/docs/FAQ.md

@@ -55,7 +55,7 @@ pi-chrome controls web pages through Chrome extension APIs, page inspection, scr
 
 ## Does `chrome_evaluate` work on strict-CSP pages?
 
-Not always. `chrome_evaluate` and `chrome_snapshot` run in the page's MAIN world through the Function constructor, so pages whose CSP blocks `'unsafe-eval'` can reject them. `chrome_screenshot`, `chrome_navigate`, tab tools, and real Chrome input still work because they use extension/browser APIs rather than page JavaScript.
+Yes. `chrome_evaluate` and `chrome_snapshot` run in the page's MAIN world through CDP `Runtime.evaluate`, which is a DevTools protocol command and is **not** subject to the page's Content-Security-Policy. They work even on pages that block `'unsafe-eval'` (e.g. github.com and many bank/SaaS apps). `chrome_navigate`'s `initScript` injects at document_start via CDP and likewise bypasses CSP. `chrome_screenshot`, tab tools, and real Chrome input also keep working under any CSP.
 
 ## How do I tell whether a click or type worked?
 

package/extensions/chrome-profile-bridge/browser-extension/manifest.json

@@ -1,10 +1,11 @@
 {
   "manifest_version": 3,
   "name": "Pi Chrome Connector",
-  "version": "0.15.28",
+  "version": "0.15.30",
   "description": "Lets Pi control tabs in Chrome via a local connector at 127.0.0.1.",
   "permissions": [
     "tabs",
+    "tabGroups",
     "scripting",
     "storage",
     "activeTab",

package/extensions/chrome-profile-bridge/browser-extension/service_worker.js

@@ -1,6 +1,9 @@
 const BRIDGE_URL = "http://127.0.0.1:17318";
 const CLIENT_NAME = `Pi Chrome Connector ${chrome.runtime.id}`;
 const POLL_ERROR_BACKOFF_MS = 2000;
+const DEFAULT_GROUP_COLOR = "blue";
+const PI_GROUP_RE = /^Pi(\b|\s*-)/i;
+const VALID_GROUP_COLORS = new Set(["grey", "blue", "red", "yellow", "green", "pink", "purple", "cyan", "orange"]);
 let polling = false;
 
 // =================== Chrome input (CDP) layer ===================
@@ -217,6 +220,38 @@ async function cdp(tabId, method, params) {
   }
 }
 
+// cdpEval: evaluate a JavaScript expression string in the page's MAIN world via CDP
+// Runtime.evaluate. Runtime.evaluate is a DevTools protocol command and is NOT subject to
+// the page's Content-Security-Policy, so it works on pages that ship `script-src 'self'`
+// without `'unsafe-eval'` (which blocks `eval`/`new Function`). Ensures the debugger is
+// attached first. Returns the raw CDP result ({ result, exceptionDetails }).
+async function cdpEval(tabId, expression, opts) {
+  await attachDebugger(tabId);
+  return cdp(tabId, "Runtime.evaluate", {
+    expression,
+    returnByValue: true,
+    awaitPromise: true,
+    userGesture: true,
+    ...(opts || {}),
+  });
+}
+
+function cdpExceptionText(details) {
+  if (!details) return "";
+  return String(
+    details.exception?.description ||
+      details.exception?.value ||
+      details.text ||
+      "",
+  );
+}
+
+function cdpIsSyntaxError(details) {
+  if (!details) return false;
+  const className = String(details.exception?.className || "");
+  return className === "SyntaxError" || /SyntaxError/.test(cdpExceptionText(details));
+}
+
 // Resolve target -> {x, y, rect} in viewport coords by running tiny script in tab.
 async function resolveTargetInTab(tabId, params) {
   const results = await chrome.scripting.executeScript({
@@ -677,6 +712,58 @@ function isVersionOlder(a, b) {
   return false;
 }
 
+function cleanGroupTitle(value) {
+  const text = String(value || "Pi").replace(/\s+/g, " ").trim().slice(0, 80);
+  return text || "Pi";
+}
+
+function cleanGroupColor(value) {
+  const color = String(value || DEFAULT_GROUP_COLOR).toLowerCase();
+  return VALID_GROUP_COLORS.has(color) ? color : DEFAULT_GROUP_COLOR;
+}
+
+async function groupRecord(groupId) {
+  if (typeof groupId !== "number" || groupId < 0 || !chrome.tabGroups) return null;
+  const group = await chrome.tabGroups.get(groupId).catch(() => null);
+  if (!group) return null;
+  return {
+    id: group.id,
+    title: group.title || "",
+    color: group.color || "",
+    collapsed: Boolean(group.collapsed),
+    windowId: group.windowId,
+    piGroup: Boolean(group.title && PI_GROUP_RE.test(group.title)),
+  };
+}
+
+// Find an existing tab group in `windowId` whose title matches `title` (case-insensitive).
+// Used so all Pi-opened tabs collect into one group per window instead of spawning new ones.
+async function findGroupByTitle(windowId, title) {
+  if (!chrome.tabGroups) return null;
+  const wanted = cleanGroupTitle(title).toLowerCase();
+  const groups = await chrome.tabGroups.query({ windowId }).catch(() => []);
+  const match = groups.find((g) => (g.title || "").trim().toLowerCase() === wanted);
+  return match ? match.id : null;
+}
+
+// Add `tab` to a tab group, then set title/color. If the tab is ungrouped, reuse an
+// existing same-title group in its window when present, otherwise create a new group.
+async function groupTab(tab, title, color) {
+  if (!chrome.tabGroups) throw new Error("chrome.tabGroups API unavailable; reload the extension after granting the tabGroups permission");
+  if (!tab || typeof tab.id !== "number") throw new Error("No tab to group");
+  const groupTitle = cleanGroupTitle(title);
+  let groupId = tab.groupId;
+  if (typeof groupId !== "number" || groupId < 0) {
+    const existing = await findGroupByTitle(tab.windowId, groupTitle);
+    groupId = existing !== null
+      ? await chrome.tabs.group({ groupId: existing, tabIds: [tab.id] })
+      : await chrome.tabs.group({ tabIds: [tab.id] });
+  }
+  await chrome.tabGroups.update(groupId, { title: groupTitle, color: cleanGroupColor(color), collapsed: false });
+  const grouped = await chrome.tabs.get(tab.id);
+  return { tab: await formatTab(grouped), group: await groupRecord(groupId) };
+}
+
 async function dispatch(action, params) {
   switch (action) {
     case "tab.version":
@@ -686,17 +773,31 @@ async function dispatch(action, params) {
         bridgeUrl: BRIDGE_URL,
         userAgent: navigator.userAgent,
       };
-    case "tab.list":
-      return (await chrome.tabs.query({})).map(formatTab);
+    case "tab.list": {
+      const tabs = await chrome.tabs.query({});
+      return Promise.all(tabs.map(formatTab));
+    }
     case "tab.new": {
       const tab = await chrome.tabs.create({ url: params.url || "about:blank", active: true });
-      return formatTab(tab);
+      // Every Pi-opened tab joins a group by default. Pass groupTitle:"" (or group:false) to opt out.
+      const optOut = params.groupTitle === "" || params.group === false;
+      if (optOut && !params.groupColor) return formatTab(tab);
+      return groupTab(tab, params.groupTitle || "Pi", params.groupColor);
     }
     case "tab.activate": {
       const tab = await getTabByParams(params);
       await chrome.windows.update(tab.windowId, { focused: true });
       return formatTab(await chrome.tabs.update(tab.id, { active: true }));
     }
+    case "tab.group": {
+      const tab = await getTabByParams(params);
+      return groupTab(tab, params.groupTitle || "Pi", params.groupColor);
+    }
+    case "tab.ungroup": {
+      const tab = await getTabByParams(params);
+      if (typeof tab.groupId === "number" && tab.groupId >= 0) await chrome.tabs.ungroup(tab.id);
+      return formatTab(await chrome.tabs.get(tab.id));
+    }
     case "tab.close": {
       const tab = await getTabByParams(params);
       await chrome.tabs.remove(tab.id);
@@ -739,8 +840,29 @@ async function dispatch(action, params) {
       return executeInTab(params, listNetworkRequests, [params.includePreservedRequests === true, params.clear === true]);
     case "page.network.get":
       return executeInTab(params, getNetworkRequest, [params.requestId]);
-    case "page.waitFor":
-      return executeInTab(params, waitForPage, [params.kind, params.value, params.timeoutMs || 10000, params.intervalMs || 250]);
+    case "page.waitFor": {
+      // Poll from the service worker via CDP (bypasses CSP). The old approach ran the polling
+      // loop in-page with new Function() for expression checks, which fails under strict CSP.
+      const tab = await getTabByParams(params);
+      if (params.foreground) await bringToFront(tab);
+      const timeoutMs = params.timeoutMs || 10000;
+      const intervalMs = params.intervalMs || 250;
+      const started = Date.now();
+      while (Date.now() - started < timeoutMs) {
+        let ok = false;
+        try {
+          const expr = params.kind === "selector"
+            ? `!!document.querySelector(${JSON.stringify(params.value)})`
+            : params.value;
+          ok = Boolean(await evaluateInTab({ ...params, expression: expr, foreground: false }));
+        } catch {
+          ok = false;
+        }
+        if (ok) return { elapsedMs: Date.now() - started };
+        await sleep(intervalMs);
+      }
+      throw new Error(`Timed out after ${timeoutMs}ms waiting for ${params.kind}: ${params.value}`);
+    }
     case "page.probe":
       // Lightweight capability probe for /chrome-doctor. Runs in MAIN world.
       return executeInTab(params, probePage, []);
@@ -758,7 +880,7 @@ async function dispatch(action, params) {
       } finally {
         if (params.initScript) await unregisterInitScript(tab.id).catch(() => undefined);
       }
-      return formatTab(await chrome.tabs.get(updated.id));
+      return await formatTab(await chrome.tabs.get(updated.id));
     }
     case "page.screenshot":
       return takeScreenshot(params);
@@ -767,7 +889,7 @@ async function dispatch(action, params) {
   }
 }
 
-function formatTab(tab) {
+async function formatTab(tab) {
   return {
     id: tab.id,
     windowId: tab.windowId,
@@ -778,6 +900,8 @@ function formatTab(tab) {
     status: tab.status,
     pinned: tab.pinned,
     incognito: tab.incognito,
+    groupId: typeof tab.groupId === "number" ? tab.groupId : -1,
+    group: await groupRecord(tab.groupId),
   };
 }
 
@@ -847,25 +971,33 @@ const HELPER_FUNCS = [
 async function executeInTab(params, func, args) {
   const tab = await getTabByParams(params);
   if (params.foreground) await bringToFront(tab);
-  const helperSource = HELPER_FUNCS.map((helper) => helper.toString()).join("\n");
+
+  // Phase 1: define the helpers and the action function as page globals via CDP
+  // Runtime.evaluate. This bypasses page CSP (no `eval`/`new Function`), which is the
+  // root cause of snapshot/click/etc silently failing on `script-src 'self'` sites.
+  // Each helper is a named function declaration, assigned to window.<name> so the action
+  // (which references helpers by bare name) resolves them as globals at call time.
+  const assignments = HELPER_FUNCS.map((helper) => `window.${helper.name}=${helper.toString()}`).join(";\n");
+  const actionAssign = `window.__piAction=(${func.toString()})`;
+  const defineRes = await cdpEval(tab.id, `(()=>{${assignments};\n${actionAssign};})()`);
+  if (defineRes.exceptionDetails) {
+    throw new Error(`Failed to inject Chrome page helpers: ${cdpExceptionText(defineRes.exceptionDetails) || "unknown error"}`);
+  }
+
+  // Phase 2: run the action via chrome.scripting.executeScript. The `func:` form is
+  // injected by Chrome itself (not `new Function`), so it is CSP-safe, and it lets Chrome
+  // serialize the invocation args. The wrapper references window.__piAction defined above.
   const results = await chrome.scripting.executeScript({
     target: { tabId: tab.id },
     world: "MAIN",
-    func: async (helperSource, source, invocationArgs) => {
+    func: async (invocationArgs) => {
       try {
-        // Helpers are plain function declarations; injecting them via Function constructor avoids
-        // running through `eval` (which is restricted under strict CSP) and keeps them isolated.
-        new Function(helperSource).call(globalThis);
-        // The action itself is reconstructed from its source text. We use `new Function` rather
-        // than `eval` because the latter is blocked by `script-src 'self'` (no `'unsafe-eval'`)
-        // CSPs that are common on production sites.
-        const injected = new Function(helperSource + "\nreturn (" + source + ");").call(globalThis);
-        return { ok: true, value: await injected(...invocationArgs) };
+        return { ok: true, value: await window.__piAction(...invocationArgs) };
       } catch (error) {
         return { ok: false, error: error?.stack || error?.message || String(error) };
       }
     },
-    args: [helperSource, func.toString(), args],
+    args: [args || []],
   });
   const first = results?.[0];
   if (first?.error) {
@@ -879,72 +1011,54 @@ async function executeInTab(params, func, args) {
   return envelope?.value;
 }
 
-// Dedicated executor for page.evaluate. Doesn't go through the helper-source injection chain;
-// that chain was the root cause of `chrome_evaluate` silently returning null on pages with strict
-// CSP. We build a single Function in MAIN world and invoke it directly.
+// Serializer for page.evaluate results. Embedded (via .toString()) into the CDP-evaluated
+// expression so we can return rich markers for values that don't survive returnByValue
+// (undefined/function/symbol/bigint/Error), plus expand DOMRect-like objects whose fields
+// are non-enumerable. Kept as a standalone function so it stays editable/lintable.
+function piEvalStringify(v) {
+  if (v === undefined) return { kind: "undefined" };
+  if (typeof v === "function") return { kind: "function", source: v.toString().slice(0, 500) };
+  if (typeof v === "symbol") return { kind: "symbol", description: v.description };
+  if (typeof v === "bigint") return { kind: "bigint", value: v.toString() };
+  if (v instanceof Error) return { kind: "error", name: v.name, message: v.message, stack: v.stack };
+  // DOMRect/DOMRectReadOnly (and getBoundingClientRect results) have non-enumerable
+  // properties, so JSON.stringify yields `{}`. Expand the fields explicitly.
+  if ((typeof DOMRectReadOnly !== "undefined" && v instanceof DOMRectReadOnly) ||
+      (typeof DOMRect !== "undefined" && v instanceof DOMRect) ||
+      (v && typeof v === "object" && typeof v.toJSON === "function" &&
+       typeof v.width === "number" && typeof v.height === "number" && typeof v.top === "number")) {
+    return { x: v.x, y: v.y, width: v.width, height: v.height, top: v.top, right: v.right, bottom: v.bottom, left: v.left };
+  }
+  return v;
+}
+
+// Dedicated executor for page.evaluate. Uses CDP Runtime.evaluate (via cdpEval) which is not
+// subject to the page's CSP, fixing `chrome_evaluate` silently returning null / failing on
+// pages that ship `script-src 'self'` without `'unsafe-eval'` (which blocks `eval`/`new Function`).
 async function evaluateInTab(params) {
   const tab = await getTabByParams(params);
   if (params.foreground) await bringToFront(tab);
   const expression = String(params.expression ?? "");
-  const awaitPromise = params.awaitPromise !== false;
-  const results = await chrome.scripting.executeScript({
-    target: { tabId: tab.id },
-    world: "MAIN",
-    func: async (expression, awaitPromise) => {
-      const stringify = (v) => {
-        if (v === undefined) return { kind: "undefined" };
-        if (typeof v === "function") return { kind: "function", source: v.toString().slice(0, 500) };
-        if (typeof v === "symbol") return { kind: "symbol", description: v.description };
-        if (typeof v === "bigint") return { kind: "bigint", value: v.toString() };
-        if (v instanceof Error) return { kind: "error", name: v.name, message: v.message, stack: v.stack };
-        // DOMRect/DOMRectReadOnly (and getBoundingClientRect results) have non-enumerable
-        // properties, so JSON.stringify yields `{}`. Expand the fields explicitly.
-        if ((typeof DOMRectReadOnly !== "undefined" && v instanceof DOMRectReadOnly) ||
-            (typeof DOMRect !== "undefined" && v instanceof DOMRect) ||
-            (v && typeof v === "object" && typeof v.toJSON === "function" &&
-             typeof v.width === "number" && typeof v.height === "number" && typeof v.top === "number")) {
-          return { x: v.x, y: v.y, width: v.width, height: v.height, top: v.top, right: v.right, bottom: v.bottom, left: v.left };
-        }
-        return v;
-      };
-      // Compile via the Function constructor. We try expression form first so callers can pass
-      // `1+1` or `document.title` without a `return`; if that's a SyntaxError we retry with the
-      // statement form so callers can use multi-statement bodies (loops, var decls, etc).
-      const compile = (src) => {
-        try {
-          return { fn: new Function(`return (async () => (${src}))();`), mode: "expression" };
-        } catch (e1) {
-          if (e1 && e1.name === "SyntaxError") {
-            try {
-              return { fn: new Function(`return (async () => { ${src} })();`), mode: "statement" };
-            } catch (e2) {
-              throw e2;
-            }
-          }
-          throw e1;
-        }
-      };
-      try {
-        const { fn } = compile(expression);
-        const value = await fn.call(globalThis);
-        const resolved = awaitPromise && value && typeof value.then === "function" ? await value : value;
-        return { ok: true, value: stringify(resolved) };
-      } catch (error) {
-        return { ok: false, error: error?.stack || error?.message || String(error) };
-      }
-    },
-    args: [expression, awaitPromise],
-  });
-  const first = results?.[0];
-  if (first?.error) {
-    const message = typeof first.error === "string" ? first.error : (first.error.message || JSON.stringify(first.error));
-    throw new Error(`chrome_evaluate failed: ${message}`);
-  }
-  const envelope = first?.result;
-  if (!envelope) throw new Error("chrome_evaluate returned no envelope from MAIN world");
-  if (envelope.ok === false) throw new Error(envelope.error || "chrome_evaluate failed");
-  const v = envelope.value;
-  // Unwrap special markers from MAIN world
+  const stringifySrc = `(${piEvalStringify.toString()})`;
+  // Wrap the user expression so the result is run through piEvalStringify in-page before it
+  // crosses the returnByValue boundary. Try expression form first (so `1+1` / `document.title`
+  // work without `return`); on a SyntaxError fall back to statement form for multi-statement
+  // bodies (loops, var decls, etc), matching the previous new Function() two-form behavior.
+  const buildWrapper = (form) => `(async () => { const __s=${stringifySrc}; const __v = await ${form}; return __s(__v); })()`;
+  const exprForm = `(async () => (${expression}))()`;
+  const stmtForm = `(async () => { ${expression} })()`;
+
+  let res = await cdpEval(tab.id, buildWrapper(exprForm));
+  if (res.exceptionDetails && cdpIsSyntaxError(res.exceptionDetails)) {
+    res = await cdpEval(tab.id, buildWrapper(stmtForm));
+  }
+  if (res.exceptionDetails) {
+    throw new Error(`chrome_evaluate failed: ${cdpExceptionText(res.exceptionDetails) || "evaluation failed"}`);
+  }
+  const result = res.result;
+  if (!result || result.type === "undefined") return undefined;
+  const v = result.value;
+  // Unwrap special markers produced by piEvalStringify.
   if (v && typeof v === "object" && !Array.isArray(v)) {
     if (v.kind === "undefined") return undefined;
     if (v.kind === "function") return `[Function: ${v.source}]`;
@@ -964,29 +1078,23 @@ async function withOptionalSnapshot(params, actionFn) {
   return result;
 }
 
-// One-shot init script registry, scoped per tab. The script source is injected at
-// document_start of the next committed navigation in that tab, in MAIN world, then cleared.
-const initScriptIds = new Map();
+// One-shot init script registry, scoped per tab. The source is registered with CDP
+// Page.addScriptToEvaluateOnNewDocument, which runs it at document_start in the page's MAIN
+// world and is NOT subject to page CSP (the old func:(code)=>new Function(code) path was
+// blocked by `script-src 'self'`). page.navigate registers before the nav and unregisters
+// after load, so only the intended navigation receives the script.
+const initScriptIds = new Map(); // tabId -> CDP script identifier
 async function registerInitScript(tabId, source) {
-  initScriptIds.set(tabId, source);
+  await attachDebugger(tabId);
+  await cdp(tabId, "Page.enable", {}).catch(() => undefined);
+  const result = await cdp(tabId, "Page.addScriptToEvaluateOnNewDocument", { source });
+  if (result && result.identifier !== undefined) initScriptIds.set(tabId, result.identifier);
 }
 async function unregisterInitScript(tabId) {
+  const identifier = initScriptIds.get(tabId);
+  if (identifier === undefined) return;
   initScriptIds.delete(tabId);
-}
-
-if (chrome.webNavigation && chrome.webNavigation.onCommitted) {
-  chrome.webNavigation.onCommitted.addListener((details) => {
-    if (details.frameId !== 0) return;
-    const source = initScriptIds.get(details.tabId);
-    if (!source) return;
-    chrome.scripting.executeScript({
-      target: { tabId: details.tabId, frameIds: [0] },
-      world: "MAIN",
-      injectImmediately: true,
-      func: (code) => { try { new Function(code).call(globalThis); } catch (e) { console.error("[pi-chrome init script]", e); } },
-      args: [source],
-    }).catch(() => undefined);
-  });
+  await cdp(tabId, "Page.removeScriptToEvaluateOnNewDocument", { identifier }).catch(() => undefined);
 }
 
 // Always inject early console/network capture at document_start on every navigation.
@@ -1058,7 +1166,7 @@ async function takeScreenshot(params) {
       await executeInTab({ ...params, foreground: false }, scrollToY, [tiles.originalScrollY]);
       return {
         fullPage: true,
-        tab: formatTab(tab),
+        tab: await formatTab(tab),
         dimensions: { width: tiles.width, height: tiles.height, viewportHeight: tiles.viewportHeight, dpr: tiles.dpr },
         tiles: captured,
       };
@@ -1067,7 +1175,7 @@ async function takeScreenshot(params) {
       format: params.format || "png",
       quality: params.format === "jpeg" ? params.quality : undefined,
     });
-    return { dataUrl, tab: formatTab(tab) };
+    return { dataUrl, tab: await formatTab(tab) };
   } finally {
     if (previousActiveId !== undefined && previousActiveId !== tab.id) {
       await chrome.tabs.update(previousActiveId, { active: true }).catch(() => undefined);
@@ -2076,20 +2184,6 @@ function getNetworkRequest(requestId) {
   return request;
 }
 
-async function waitForPage(kind, value, timeoutMs, intervalMs) {
-  const started = Date.now();
-  while (Date.now() - started < timeoutMs) {
-    let ok = false;
-    if (kind === "selector") ok = Boolean(document.querySelector(value));
-    else {
-      try { ok = Boolean(new Function("return (" + value + ");").call(globalThis)); } catch { ok = false; }
-    }
-    if (ok) return { elapsedMs: Date.now() - started };
-    await new Promise((resolve) => setTimeout(resolve, intervalMs));
-  }
-  throw new Error(`Timed out after ${timeoutMs}ms waiting for ${kind}: ${value}`);
-}
-
 function normalizeKey(key) {
   const table = {
     enter: "Enter",

package/extensions/chrome-profile-bridge/index.ts

@@ -500,7 +500,7 @@ class ChromeProfileBridge {
 	}
 }
 
-const tabActionValues = ["list", "new", "activate", "close", "version"] as const;
+const tabActionValues = ["list", "new", "activate", "close", "group", "ungroup", "version"] as const;
 const imageFormatValues = ["png", "jpeg"] as const;
 const waitForValues = ["selector", "expression"] as const;
 const CHROME_TOOL_NAMES = [
@@ -665,7 +665,7 @@ Chrome control is available through the chrome_* tools via a companion Chrome ex
 Capability model (important):
 - Interactive controls (click/type/fill/key/hover/drag/scroll/tap) use Chrome's real input layer via chrome.debugger / CDP. Events satisfy normal user-activation gates.
 - Input bypasses page CSP because it is injected at browser input layer, not page JavaScript. Chrome may show the “Pi Chrome Connector started debugging this browser” banner while attached.
-- \`chrome_evaluate\` and \`chrome_snapshot\` run in MAIN world via the **Function constructor**, which requires \`'unsafe-eval'\` in the page CSP. Pages with strict CSP (e.g. github.com, many bank/SaaS apps) will throw \`EvalError: ... 'unsafe-eval' is not an allowed source of script\` and chrome_snapshot will return empty. On those pages, drive the page with \`chrome_screenshot\` + viewport-coordinate \`chrome_click\`/\`chrome_type\`/\`chrome_key\`. \`chrome_navigate\`, \`chrome_screenshot\`, \`chrome_tab\`, and Chrome input all keep working under any CSP.
+- \`chrome_evaluate\` and \`chrome_snapshot\` run in MAIN world via **CDP \`Runtime.evaluate\`**, which is not subject to the page's Content-Security-Policy. They work even on strict-CSP pages (e.g. github.com, many bank/SaaS apps) that block \`'unsafe-eval'\`. \`chrome_navigate initScript\` likewise injects at document_start via CDP and bypasses CSP. \`chrome_screenshot\`, \`chrome_tab\`, and Chrome input also work under any CSP.
 - Input tools return structured details and support \`includeSnapshot=true\` on click/type/fill/key. Use the fresh snapshot to verify state instead of repeating blindly.
 
 Usage rules:
@@ -1029,20 +1029,25 @@ Usage rules:
 	pi.registerTool({
 		name: "chrome_tab",
 		label: "Chrome Tab",
-		description: "List, create, activate, close, or inspect tabs in the user's existing Chrome profile via the companion extension.",
-		promptSnippet: "List/open/activate/close existing Chrome tabs through the companion extension.",
+		description: "List, create, activate, close, group, ungroup, or inspect tabs in the user's existing Chrome profile via the companion extension.",
+		promptSnippet: "List/open/activate/close/group existing Chrome tabs through the companion extension.",
 		parameters: Type.Object({
 			action: StringEnum(tabActionValues),
 			url: Type.Optional(Type.String({ description: "URL for action=new." })),
-			targetId: Type.Optional(Type.String({ description: "Chrome tab id for activate/close." })),
+			targetId: Type.Optional(Type.String({ description: "Chrome tab id for activate/close/group/ungroup." })),
+			urlIncludes: Type.Optional(Type.String({ description: "Match the target tab by URL substring for activate/close/group/ungroup." })),
+			titleIncludes: Type.Optional(Type.String({ description: "Match the target tab by title substring for activate/close/group/ungroup." })),
+			group: Type.Optional(Type.Boolean({ description: "action=new only: pass false to open an ungrouped tab. By default every Pi-opened tab joins the window's 'Pi' tab group." })),
+			groupTitle: Type.Optional(Type.String({ description: "Tab group title for action=group (or action=new to open into a named group). Defaults to 'Pi'. Pass an empty string on action=new to opt out of grouping." })),
+			groupColor: Type.Optional(Type.String({ description: "Tab group color for action=group/new: grey, blue, red, yellow, green, pink, purple, cyan, or orange. Defaults to blue." })),
 			host: Type.Optional(Type.String()),
 			port: Type.Optional(Type.Number()),
 		}),
 		async execute(_id, params, signal): Promise<ToolTextResult> {
 			const result = await authorizedBridgeSend(`tab.${params.action}`, params, DEFAULT_TIMEOUT_MS, signal);
 			if (params.action === "list") {
-				const tabs = result as Array<{ id: number; title: string; url: string; active: boolean; windowId: number }>;
-				const text = tabs.map((tab) => `${tab.id}\t${tab.active ? "*" : " "}\t${tab.title || "(untitled)"}\t${tab.url}`).join("\n") || "No tabs.";
+				const tabs = result as Array<{ id: number; title: string; url: string; active: boolean; windowId: number; group?: { title?: string } | null }>;
+				const text = tabs.map((tab) => `${tab.id}\t${tab.active ? "*" : " "}\t${tab.group?.title ? `[${tab.group.title}] ` : ""}${tab.title || "(untitled)"}\t${tab.url}`).join("\n") || "No tabs.";
 				return { content: [{ type: "text", text }], details: { tabs } };
 			}
 			return { content: [{ type: "text", text: safeJson(result) }], details: { result: result as Json } };

package/package.json

@@ -1,7 +1,8 @@
 {
 	"name": "pi-chrome",
-	"version": "0.15.28",
+	"version": "0.15.30",
 	"scripts": {
+		"test": "node test-suite/unit/csp-eval.test.mjs",
 		"version": "node scripts/sync-manifest-version.js",
 		"prepublishOnly": "node scripts/sync-manifest-version.js"
 	},

package/test-suite/README.md

@@ -125,7 +125,7 @@ Each unit challenge has a `gate` field:
 - `dom-complexity` / `frames` — Shadow DOM and iframe targeting.
 - `files` — file attachment to `<input type=file>`.
 - `observability` — console/network capture tools.
-- `csp` — strict Content Security Policy fallback where eval/snapshot may fail.
+- `csp` — strict Content Security Policy: screenshot/coordinate fallback (39) and the CDP eval/snapshot bypass that works under `script-src 'self'` without `unsafe-eval` (42).
 - `lazy-loading` — dynamic DOM readiness and wait behavior.
 - `fingerprint` — environment and stack fingerprint probes.
 - `agent-safety` — hidden honeypots and safe target selection.
@@ -175,6 +175,7 @@ The dashboard renders this from `manifest.json`. In brief:
 39. strict CSP screenshot/coordinate fallback
 40. dynamic wait/readiness
 41. explicit tab lifecycle
+42. strict CSP eval/snapshot via CDP (regression guard for the CSP bypass)
 
 ## Design notes
 

package/test-suite/challenges/42-strict-csp-evaluate.html

@@ -0,0 +1,16 @@
+<!doctype html>
+<meta charset="utf-8">
+<meta http-equiv="Content-Security-Policy" content="default-src 'self'; script-src 'self'; style-src 'self' 'unsafe-inline'; object-src 'none'; base-uri 'none'">
+<title>42 strict CSP evaluate/snapshot</title>
+<link rel="stylesheet" href="../_style.css">
+<script src="../_lib.js"></script>
+<body>
+<main>
+  <p>Goal: this page ships a strict CSP (<code>script-src 'self'</code>, no <code>unsafe-eval</code>), which blocks <code>eval</code>/<code>new Function</code>. <code>chrome_evaluate</code> and <code>chrome_snapshot</code> must still work because they run through CDP, which is not subject to page CSP.</p>
+  <p id="hint">A secret token is exposed only at <code>window.__cspToken</code> — it is never written into the DOM. Use <code>chrome_evaluate</code> to read it, type it into the field (snapshot/uid to find the field), then click Verify.</p>
+  <label for="tokenInput">Token:</label>
+  <input id="tokenInput" type="text" autocomplete="off" aria-label="csp token">
+  <button id="verify" aria-label="verify token">Verify</button>
+</main>
+<script src="42-strict-csp-evaluate.js"></script>
+</body>

package/test-suite/challenges/42-strict-csp-evaluate.js

@@ -0,0 +1,21 @@
+Challenge.init({
+  id: "strict-csp-evaluate",
+  instructions: "under strict CSP: read window.__cspToken via chrome_evaluate, type it into the field, click Verify",
+});
+
+// Secret available only via JS evaluation. It is intentionally NOT rendered into the DOM and
+// is defined non-enumerable, so the only way to obtain it is to evaluate window.__cspToken in
+// the page (which proves chrome_evaluate works despite script-src 'self' blocking eval).
+const token = "csp-" + Math.random().toString(36).slice(2, 10);
+Object.defineProperty(window, "__cspToken", { value: token, enumerable: false, configurable: false, writable: false });
+
+document.getElementById("verify").addEventListener("click", (e) => {
+  const bad = [];
+  if (!e.isTrusted) bad.push("verify click isTrusted=false (use trusted/CDP input)");
+  const val = (document.getElementById("tokenInput").value || "").trim();
+  if (val !== token) {
+    bad.push(`token mismatch: got "${val}" expected "${token}" — chrome_evaluate must read window.__cspToken under strict CSP`);
+  }
+  if (bad.length) Challenge.fail(...bad);
+  else Challenge.pass("strict CSP: chrome_evaluate read the hidden token via CDP and trusted input submitted it");
+});

package/test-suite/manifest.json

@@ -1510,7 +1510,7 @@
       "strict-csp"
     ],
     "notes": [
-      "chrome_snapshot/chrome_evaluate may fail on this page because script-src omits unsafe-eval; this is intentional. Use screenshot plus viewport coordinates, then read verdict from dashboard/localStorage after leaving CSP page."
+      "This challenge exercises the pure screenshot + viewport-coordinate path (no snapshot/evaluate needed). Note: as of the CDP CSP bypass, chrome_snapshot/chrome_evaluate DO work here even though script-src omits unsafe-eval (see challenge 42 strict-csp-evaluate). Read verdict from dashboard/localStorage after leaving the CSP page."
     ],
     "manualBaseline": "unverified",
     "gradeSource": "page"
@@ -1626,5 +1626,56 @@
     ],
     "manualBaseline": "unverified",
     "gradeSource": "page"
+  },
+  {
+    "id": "strict-csp-evaluate",
+    "file": "challenges/42-strict-csp-evaluate.html",
+    "category": "csp",
+    "difficulty": "L2",
+    "gate": "core",
+    "goal": "On a page with strict CSP (script-src 'self', no unsafe-eval), use chrome_evaluate + chrome_snapshot via CDP to read a JS-only secret (window.__cspToken), then submit it with trusted input.",
+    "expected": {
+      "synthetic": "FAIL",
+      "trusted": "PASS",
+      "manual": "PASS"
+    },
+    "recipe": [
+      {
+        "tool": "chrome_evaluate",
+        "params": {
+          "expression": "window.__cspToken"
+        }
+      },
+      {
+        "tool": "chrome_fill",
+        "params": {
+          "selector": "#tokenInput",
+          "value": "$RESULT_OF_chrome_evaluate",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#verify",
+          "trusted": true
+        }
+      }
+    ],
+    "requires": {
+      "cdp": true
+    },
+    "tags": [
+      "csp",
+      "evaluate",
+      "snapshot",
+      "strict-csp",
+      "cdp-bypass"
+    ],
+    "notes": [
+      "Regression guard for the CDP CSP bypass: chrome_evaluate and chrome_snapshot must work even though script-src omits unsafe-eval (eval/new Function are blocked). The token is non-enumerable and never in the DOM, so it can only be obtained via chrome_evaluate. Runner must substitute the evaluate result into the fill value. synthetic FAIL is due to the trusted-click gate, not eval availability."
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page"
   }
 ]

package/test-suite/unit/csp-eval.test.mjs

@@ -0,0 +1,171 @@
+// Unit harness for the CSP-bypass layer in service_worker.js.
+//
+// The real CSP bypass (CDP Runtime.evaluate not being subject to page CSP) can only be
+// proven in a browser — see challenge 39-strict-csp-fallback. These tests instead validate
+// the JS *logic* of the refactor that the bypass depends on:
+//   - evaluateInTab: wrapper-string construction, expression/statement fallback, value
+//     marker round-trip (undefined/function/symbol/bigint/Error/DOMRect), error propagation.
+//   - executeInTab: 2-phase define-then-invoke, envelope unwrap, error propagation, and that
+//     all real HELPER_FUNCS serialize+assign without a parse error.
+//   - page.waitFor: service-worker-side polling via evaluateInTab (selector + expression).
+//
+// We load the worker into a vm sandbox with mocked chrome.* APIs, then replace `cdp` with a
+// shim that evaluates the expression in a separate "page world" vm context (simulating CDP
+// Runtime.evaluate returnByValue). No browser, no network, no deps.
+
+import vm from "node:vm";
+import fs from "node:fs";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+const workerPath = path.resolve(__dirname, "../../extensions/chrome-profile-bridge/browser-extension/service_worker.js");
+const src = fs.readFileSync(workerPath, "utf8");
+
+let failures = 0;
+let passes = 0;
+function ok(cond, msg) {
+  if (cond) { passes++; }
+  else { failures++; console.error(`  ✗ ${msg}`); }
+}
+async function throwsWith(fn, re, msg) {
+  try { await fn(); ok(false, `${msg} (expected throw)`); }
+  catch (e) { ok(re.test(String(e.message || e)), `${msg} (got: ${e.message})`); }
+}
+
+// ---- page world: simulates the page's MAIN world for Runtime.evaluate ----
+const pageGlobals = {
+  console, JSON, Date, Math, Promise, Object, Array, String, Number, Boolean,
+  Error, TypeError, SyntaxError, RangeError, BigInt, Symbol, structuredClone,
+  setTimeout, parseInt, parseFloat, isNaN,
+  document: {
+    title: "page title",
+    _present: new Set(),
+    querySelector(sel) { return this._present.has(sel) ? { sel } : null; },
+  },
+};
+pageGlobals.window = pageGlobals;
+pageGlobals.globalThis = pageGlobals;
+const pageWorld = vm.createContext(pageGlobals);
+
+// Simulate CDP Runtime.evaluate returnByValue serialization.
+function toCdpResult(v) {
+  if (v === undefined) return { result: { type: "undefined" } };
+  if (v === null) return { result: { type: "object", subtype: "null", value: null } };
+  const t = typeof v;
+  if (t === "number" || t === "string" || t === "boolean")
+    return { result: { type: t, value: v } };
+  // object/array: returnByValue deep-clones JSON-able structures
+  return { result: { type: "object", value: JSON.parse(JSON.stringify(v)) } };
+}
+
+// ---- worker sandbox ----
+const noop = () => {};
+const listener = { addListener: noop, removeListener: noop };
+const sandbox = {
+  console, JSON, Date, Math, Promise, Array, Object, String, Number, Boolean,
+  Error, TypeError, Map, Set, BigInt, Symbol, structuredClone,
+  setTimeout, clearTimeout,
+  setInterval: () => 0,
+  clearInterval: noop,
+  fetch: async () => { throw new Error("no network in unit test"); },
+  navigator: { userAgent: "unit-test" },
+  WebSocket: function () {},
+  chrome: {
+    runtime: { id: "unittestextension", getManifest: () => ({ version: "0.0.0" }), onInstalled: listener, onStartup: listener, lastError: null },
+    alarms: { onAlarm: listener, create: noop, clear: noop, clearAll: noop },
+    action: { onClicked: listener },
+    debugger: { sendCommand: noop, attach: async () => {}, detach: async () => {}, getTargets: (cb) => cb([]) },
+    scripting: { executeScript: async () => [{ result: undefined }] },
+    tabs: { query: async () => [], get: async () => ({}), create: async () => ({}), update: async () => ({}), remove: async () => {} },
+    windows: { update: async () => {} },
+    webNavigation: { onCommitted: listener },
+  },
+};
+sandbox.globalThis = sandbox;
+sandbox.self = sandbox;
+vm.createContext(sandbox);
+vm.runInContext(src, sandbox);
+
+// ---- override the page-touching primitives with the page-world shim ----
+sandbox.attachDebugger = async () => ({});
+sandbox.bringToFront = async () => {};
+sandbox.getTabByParams = async (p) => ({ id: (p && p.targetId) || 1, windowId: 1 });
+sandbox.cdp = async (_tabId, method, params) => {
+  if (method !== "Runtime.evaluate") return {};
+  try {
+    const value = await vm.runInContext(params.expression, pageWorld);
+    return toCdpResult(value);
+  } catch (e) {
+    return { exceptionDetails: { exception: { className: e.name, description: String(e.stack || e.message) }, text: "Uncaught " + String(e) } };
+  }
+};
+// Phase-2 of executeInTab: run the injected wrapper func against the page world,
+// where Phase-1 (via cdp shim above) already defined window.__piAction + helpers.
+sandbox.chrome.scripting.executeScript = async ({ func, args }) => {
+  const fn = vm.runInContext("(" + func.toString() + ")", pageWorld);
+  const result = await fn(...(args || []));
+  return [{ result }];
+};
+
+const { evaluateInTab, executeInTab, dispatch } = sandbox;
+
+async function run() {
+  // ===== evaluateInTab: primitives & objects =====
+  ok((await evaluateInTab({ expression: "2 + 2" })) === 4, "evaluate: arithmetic expression");
+  ok((await evaluateInTab({ expression: "document.title" })) === "page title", "evaluate: expression without return");
+  ok((await evaluateInTab({ expression: "'a' + 'b'" })) === "ab", "evaluate: string concat");
+  const obj = await evaluateInTab({ expression: "({a:1, b:[2,3]})" });
+  ok(obj && obj.a === 1 && obj.b[1] === 3, "evaluate: object literal round-trips");
+
+  // ===== value markers =====
+  ok((await evaluateInTab({ expression: "void 0" })) === undefined, "evaluate: undefined marker -> undefined");
+  ok((await evaluateInTab({ expression: "10n" })) === "10", "evaluate: bigint marker -> string");
+  ok(/^\[Function:/.test(await evaluateInTab({ expression: "(function foo(){})" })), "evaluate: function marker");
+  ok((await evaluateInTab({ expression: "Promise.resolve(42)" })) === 42, "evaluate: promise is awaited");
+
+  // DOMRect-like (toJSON + width/height/top) is expanded, not flattened to {}
+  const rect = await evaluateInTab({ expression: "({ x:1,y:2,width:3,height:4,top:2,right:4,bottom:6,left:1, toJSON(){return {}} })" });
+  ok(rect && rect.width === 3 && rect.bottom === 6, "evaluate: DOMRect-like expanded");
+
+  // ===== statement-form fallback (expression form is a SyntaxError) =====
+  // `let x=...; x` is not a valid expression, so the wrapper must retry as a statement body.
+  ok((await evaluateInTab({ expression: "let x = 5; x" })) === undefined, "evaluate: statement form falls back (no return -> undefined)");
+  ok((await evaluateInTab({ expression: "let y = 7; return y" })) === 7, "evaluate: statement form with explicit return");
+
+  // ===== error propagation =====
+  await throwsWith(() => evaluateInTab({ expression: "throw new Error('boom')" }), /chrome_evaluate failed[\s\S]*boom/, "evaluate: runtime error propagates");
+
+  // ===== executeInTab: 2-phase define + invoke =====
+  // Real HELPER_FUNCS get serialized + assigned in Phase 1; a parse error there would throw here.
+  const sum = await executeInTab({ targetId: 1 }, function add(a, b) { return a + b; }, [3, 4]);
+  ok(sum === 7, "executeInTab: action runs with args after helper injection");
+
+  const asyncResult = await executeInTab({ targetId: 1 }, async function asyncEcho(v) { return v * 2; }, [21]);
+  ok(asyncResult === 42, "executeInTab: async action awaited");
+
+  await throwsWith(
+    () => executeInTab({ targetId: 1 }, function boom() { throw new Error("action failed"); }, []),
+    /action failed/,
+    "executeInTab: thrown action error propagates via envelope",
+  );
+
+  // ===== page.waitFor (service-worker-side polling) =====
+  pageGlobals.document._present.add("#ready");
+  const wf = await dispatch("page.waitFor", { targetId: 1, kind: "selector", value: "#ready", timeoutMs: 1000, intervalMs: 20 });
+  ok(wf && typeof wf.elapsedMs === "number", "waitFor: selector present resolves");
+
+  const wfExpr = await dispatch("page.waitFor", { targetId: 1, kind: "expression", value: "1 === 1", timeoutMs: 1000, intervalMs: 20 });
+  ok(wfExpr && typeof wfExpr.elapsedMs === "number", "waitFor: truthy expression resolves");
+
+  await throwsWith(
+    () => dispatch("page.waitFor", { targetId: 1, kind: "selector", value: "#never", timeoutMs: 120, intervalMs: 30 }),
+    /Timed out after 120ms/,
+    "waitFor: missing selector times out",
+  );
+
+  console.log(`\n${passes} passed, ${failures} failed`);
+  if (failures) process.exit(1);
+}
+
+run().catch((e) => { console.error(e); process.exit(1); });