chore(release): 1.5.18 — issue-reporting guidance + clear data:-over-relay error

#5: the CLI never pointed users at the issue tracker. `--help` and `--version`
now print the issues URL (https://github.com/leeguooooo/chrome-use/issues), so
agents hitting a rough edge know where to report it (the skill already nudges).

#1: navigating to a `data:` URL over the extension relay fails with a cryptic
`net::ERR_ABORTED` on an about:blank tab (chrome.debugger/chrome.tabs can't drive
a top-frame data: navigation; it works fine under --launch). Detect that case and
explain it — use a real http(s)://file:// URL or --launch — instead of leaking
the raw code. (Surfaced while stress-testing 5 concurrent agents.)

cli/Cargo.lock

@@ -290,7 +290,7 @@ checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
 
 [[package]]
 name = "chrome-use"
-version = "1.5.4"
+version = "1.5.18"
 dependencies = [
  "aes",
  "aes-gcm",

cli/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "chrome-use"
-version = "1.5.4"
+version = "1.5.18"
 edition = "2021"
 description = "Fast browser automation CLI for AI agents"
 license = "Apache-2.0"

cli/src/commands.rs

@@ -34,11 +34,50 @@ pub enum ParseError {
 /// suggestions on an unknown command (issue #29). Not exhaustive — just the
 /// common verbs plus a few known wrong-guesses mapped to the real command.
 const KNOWN_COMMANDS: &[&str] = &[
-    "open", "navigate", "click", "fill", "type", "press", "snapshot", "screenshot", "eval", "get",
-    "text", "html", "frames", "find", "wait", "scroll", "hover", "select", "check", "uncheck",
-    "tab", "tabs", "close", "back", "forward", "reload", "sessions", "status", "daemon", "doctor",
-    "upgrade", "connect", "cookies", "mouse", "keyboard", "stream", "frame", "profiles", "title",
-    "url", "is", "drag", "dialog", "upload",
+    "open",
+    "navigate",
+    "click",
+    "fill",
+    "type",
+    "press",
+    "snapshot",
+    "screenshot",
+    "eval",
+    "get",
+    "text",
+    "html",
+    "frames",
+    "find",
+    "wait",
+    "scroll",
+    "hover",
+    "select",
+    "check",
+    "uncheck",
+    "tab",
+    "tabs",
+    "close",
+    "back",
+    "forward",
+    "reload",
+    "sessions",
+    "status",
+    "daemon",
+    "doctor",
+    "upgrade",
+    "connect",
+    "cookies",
+    "mouse",
+    "keyboard",
+    "stream",
+    "frame",
+    "profiles",
+    "title",
+    "url",
+    "is",
+    "drag",
+    "dialog",
+    "upload",
 ];
 
 /// Levenshtein distance, capped — small inputs only (command names).
@@ -525,20 +564,31 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             Ok(json!({ "id": id, "action": "fill", "selector": sel, "value": rest[1..].join(" ") }))
         }
         "type" => {
+            // `--key-events` (alias `--keys`): send real per-character keystrokes
+            // instead of Input.insertText, so autocomplete/combobox widgets that
+            // only react to key events fire (e.g. Google address postal lookup).
+            let key_events = rest.iter().any(|a| *a == "--key-events" || *a == "--keys");
+            let rest: Vec<&str> = rest
+                .iter()
+                .copied()
+                .filter(|a| *a != "--key-events" && *a != "--keys")
+                .collect();
             // `type --focused <text>` types into whatever element currently has
             // focus (no selector) — for custom widgets that move focus to a hidden
             // input after you open them.
             if rest.first() == Some(&"--focused") {
                 return Ok(json!({
                     "id": id, "action": "type", "focused": true,
-                    "text": rest[1..].join(" "),
+                    "text": rest[1..].join(" "), "keyEvents": key_events,
                 }));
             }
             let sel = rest.first().ok_or_else(|| ParseError::MissingArguments {
                 context: "type".to_string(),
-                usage: "type <selector> <text>   (or: type --focused <text>)",
+                usage: "type <selector> <text>   (or: type --focused <text>) [--key-events]",
             })?;
-            Ok(json!({ "id": id, "action": "type", "selector": sel, "text": rest[1..].join(" ") }))
+            Ok(
+                json!({ "id": id, "action": "type", "selector": sel, "text": rest[1..].join(" "), "keyEvents": key_events }),
+            )
         }
         "pick" => {
             // pick <selector|@ref> --option "<text>"  — atomic combobox select:
@@ -729,10 +779,57 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
                         } else {
                             return Err(ParseError::MissingArguments {
                                 context: "scroll --selector".to_string(),
-                                usage: "scroll [direction] [amount] [--selector <sel>]",
+                                usage: "scroll [direction] [amount] [--selector <sel>] [--at <x,y>] [--frame <n>]",
                             });
                         }
                     }
+                    "--at" => {
+                        // `--at x,y`: dispatch the wheel at this viewport pixel, so it
+                        // scrolls whatever element/iframe is under the pointer — including
+                        // cross-origin iframes that `window.scrollBy` can't reach (#36).
+                        let val = rest.get(i + 1).ok_or(ParseError::MissingArguments {
+                            context: "scroll --at".to_string(),
+                            usage: "scroll [direction] [amount] --at <x,y>",
+                        })?;
+                        let mut parts = val.split(',');
+                        match (
+                            parts.next().and_then(|s| s.trim().parse::<f64>().ok()),
+                            parts.next().and_then(|s| s.trim().parse::<f64>().ok()),
+                        ) {
+                            (Some(x), Some(y)) => {
+                                obj.insert("at".to_string(), json!([x, y]));
+                            }
+                            _ => {
+                                return Err(ParseError::InvalidValue {
+                                    message: format!("scroll --at: invalid coordinate `{}`", val),
+                                    usage:
+                                        "scroll [direction] [amount] --at <x,y> (e.g. --at 640,400)",
+                                })
+                            }
+                        }
+                        i += 1;
+                    }
+                    "--frame" => {
+                        // `--frame n`: scroll the n-th frame from `chrome-use frames` by
+                        // dispatching the wheel at that frame's center — reaches content in
+                        // a cross-origin iframe without needing a selector into it (#36).
+                        let val = rest.get(i + 1).ok_or(ParseError::MissingArguments {
+                            context: "scroll --frame".to_string(),
+                            usage: "scroll [direction] [amount] --frame <n>",
+                        })?;
+                        match val.trim().parse::<usize>() {
+                            Ok(n) => {
+                                obj.insert("frame".to_string(), json!(n));
+                            }
+                            Err(_) => {
+                                return Err(ParseError::InvalidValue {
+                                    message: format!("scroll --frame: invalid index `{}`", val),
+                                    usage: "scroll [direction] [amount] --frame <n> (index from `chrome-use frames`)",
+                                })
+                            }
+                        }
+                        i += 1;
+                    }
                     arg if arg.starts_with('-') => {}
                     _ => {
                         match positional_index {
@@ -907,17 +1004,41 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             // selector: @ref or CSS selector
             // path: file path (contains / or . or ends with known extension)
             let mut full_page = false;
-            let positional: Vec<&str> = rest
-                .iter()
-                .filter(|arg| match **arg {
-                    "--full" | "-f" => {
-                        full_page = true;
-                        false
+            let mut clip: Option<Value> = None;
+            let mut positional: Vec<&str> = Vec::new();
+            let mut i = 0;
+            while i < rest.len() {
+                match rest[i] {
+                    "--full" | "-f" => full_page = true,
+                    // `--clip x,y,w,h` captures a pixel region (issue #34).
+                    "--clip" => {
+                        let raw = rest
+                            .get(i + 1)
+                            .ok_or_else(|| ParseError::MissingArguments {
+                                context: "screenshot --clip".to_string(),
+                                usage: "screenshot --clip <x,y,w,h> [path]",
+                            })?;
+                        let nums: Vec<f64> = raw
+                            .split(',')
+                            .filter_map(|n| n.trim().parse::<f64>().ok())
+                            .collect();
+                        if nums.len() != 4 {
+                            return Err(ParseError::InvalidValue {
+                                message: format!(
+                                    "--clip expects 'x,y,w,h' (4 numbers), got '{raw}'"
+                                ),
+                                usage: "screenshot --clip <x,y,w,h> [path]",
+                            });
+                        }
+                        clip = Some(json!({
+                            "x": nums[0], "y": nums[1], "width": nums[2], "height": nums[3]
+                        }));
+                        i += 1;
                     }
-                    _ => true,
-                })
-                .copied()
-                .collect();
+                    other => positional.push(other),
+                }
+                i += 1;
+            }
             let (selector, path) = match (positional.first(), positional.get(1)) {
                 (Some(first), Some(second)) => {
                     // Two args: first is selector, second is path
@@ -948,6 +1069,9 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
                 "path": path, "selector": selector,
                 "fullPage": full_page, "annotate": flags.annotate
             });
+            if let Some(c) = clip {
+                cmd["clip"] = c;
+            }
             if let Some(ref fmt) = flags.screenshot_format {
                 cmd["format"] = json!(fmt);
             }
@@ -1078,6 +1202,14 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             Ok(json!({ "id": id, "action": "stealth_status" }))
         }
 
+        // `cf-status` — Cloudflare challenge/clearance preflight: is the page
+        // currently a CF challenge, and is there a still-valid cf_clearance (the
+        // HttpOnly persistence cookie)? Lets an agent SKIP re-solving when already
+        // cleared, and know when it must solve. Persistence optimization.
+        "cf-status" | "cf" | "cloudflare-status" | "clearance" => {
+            Ok(json!({ "id": id, "action": "cf_status" }))
+        }
+
         // === Close ===
         "close" | "quit" | "exit" => {
             // `close <tab>` closes only that tab (and the output says "Tab
@@ -4035,6 +4167,28 @@ mod tests {
         assert_eq!(cmd["action"], "type");
         assert_eq!(cmd["selector"], "#input");
         assert_eq!(cmd["text"], "some text");
+        assert_eq!(cmd["keyEvents"], false);
+    }
+
+    #[test]
+    fn test_type_key_events() {
+        // --key-events sends real keystrokes (for autocomplete/combobox) and must
+        // not be swallowed into the typed text.
+        let cmd = parse_command(
+            &args("type #postal 201-0001 --key-events"),
+            &default_flags(),
+        )
+        .unwrap();
+        assert_eq!(cmd["action"], "type");
+        assert_eq!(cmd["selector"], "#postal");
+        assert_eq!(cmd["text"], "201-0001");
+        assert_eq!(cmd["keyEvents"], true);
+
+        let focused =
+            parse_command(&args("type --focused 201-0001 --keys"), &default_flags()).unwrap();
+        assert_eq!(focused["focused"], true);
+        assert_eq!(focused["text"], "201-0001");
+        assert_eq!(focused["keyEvents"], true);
     }
 
     #[test]
@@ -4319,6 +4473,24 @@ mod tests {
         assert_eq!(cmd["fullPage"], true);
     }
 
+    #[test]
+    fn test_screenshot_clip() {
+        // `--clip x,y,w,h` captures a pixel region (issue #34); the path still parses.
+        let cmd = parse_command(
+            &args("screenshot --clip 10,20,200,40 out.png"),
+            &default_flags(),
+        )
+        .unwrap();
+        assert_eq!(cmd["action"], "screenshot");
+        assert_eq!(cmd["clip"]["x"], 10.0);
+        assert_eq!(cmd["clip"]["y"], 20.0);
+        assert_eq!(cmd["clip"]["width"], 200.0);
+        assert_eq!(cmd["clip"]["height"], 40.0);
+        assert_eq!(cmd["path"], "out.png");
+        // Bad clip is a clear error, not silent.
+        assert!(parse_command(&args("screenshot --clip 1,2,3"), &default_flags()).is_err());
+    }
+
     #[test]
     fn test_screenshot_with_ref() {
         let cmd = parse_command(&args("screenshot @e1"), &default_flags()).unwrap();
@@ -4886,7 +5058,10 @@ mod tests {
         assert_eq!(nearest_command("sesions").as_deref(), Some("sessions"));
         assert_eq!(nearest_command("session").as_deref(), Some("sessions"));
         assert_eq!(nearest_command("clik").as_deref(), Some("click"));
-        assert_eq!(nearest_command("screenshits").as_deref(), Some("screenshot"));
+        assert_eq!(
+            nearest_command("screenshits").as_deref(),
+            Some("screenshot")
+        );
         // Nonsense with no close match stays silent.
         assert_eq!(nearest_command("xyzzy"), None);
         // The unknown-command error embeds the suggestion.
@@ -5822,6 +5997,35 @@ mod tests {
         assert_eq!(cmd["selector"], ".sidebar");
     }
 
+    #[test]
+    fn test_scroll_at_coordinate() {
+        // `--at x,y` carries a [x, y] array for a wheel dispatched at that pixel
+        // (issue #36: cross-origin iframe scroll).
+        let cmd = parse_command(&args("scroll down 700 --at 640,400"), &default_flags()).unwrap();
+        assert_eq!(cmd["action"], "scroll");
+        assert_eq!(cmd["direction"], "down");
+        assert_eq!(cmd["amount"], 700);
+        assert_eq!(cmd["at"], json!([640.0, 400.0]));
+    }
+
+    #[test]
+    fn test_scroll_at_rejects_garbage() {
+        assert!(parse_command(&args("scroll --at nope"), &default_flags()).is_err());
+        assert!(parse_command(&args("scroll --at 1"), &default_flags()).is_err());
+    }
+
+    #[test]
+    fn test_scroll_frame_index() {
+        let cmd = parse_command(&args("scroll down 700 --frame 2"), &default_flags()).unwrap();
+        assert_eq!(cmd["action"], "scroll");
+        assert_eq!(cmd["frame"], 2);
+    }
+
+    #[test]
+    fn test_scroll_frame_rejects_non_integer() {
+        assert!(parse_command(&args("scroll --frame two"), &default_flags()).is_err());
+    }
+
     #[test]
     fn test_scroll_selector_before_positional() {
         let cmd =

cli/src/connect.rs

@@ -449,6 +449,47 @@ pub fn relay_url() -> Option<String> {
     }
 }
 
+/// Append a one-line record of how a CDP connection was established, to
+/// `~/.chrome-use/connect-mode.log`. This is the smoking-gun detector for the
+/// "Allow remote debugging?" consent modal: that modal ONLY appears on a raw
+/// remote-debugging attach / a browser we launched with a debug port — NEVER on
+/// the extension relay. When the modal reappears, this log says which session
+/// took which path and when, so we can tell a code regression (`raw-port` /
+/// `launched` while the relay was up) from Chrome's own extension-debugger
+/// consent UX. Low volume (one line per connection); best-effort, never fails a
+/// connection.
+pub fn log_connect_mode(ws_url: &str, launched: bool, session: &str) {
+    let relay = relay_url();
+    let relay_up = relay.is_some();
+    let mode = if launched {
+        "launched(debug-port)"
+    } else if relay.as_deref() == Some(ws_url) {
+        "relay"
+    } else if ws_url.contains("127.0.0.1") || ws_url.contains("localhost") {
+        "raw-port-attach"
+    } else {
+        "remote-ws"
+    };
+    // A raw-port attach or a self-launch while the relay was available is the
+    // exact thing that pops the consent modal — flag it loudly in the line.
+    let suspect = (mode == "raw-port-attach" || launched) && relay_up;
+    let line = format!(
+        "session={session} mode={mode} relay_up={relay_up}{} ws={ws_url}\n",
+        if suspect { " CONSENT-MODAL-RISK" } else { "" }
+    );
+    if let Some(home) = dirs::home_dir() {
+        let path = home.join(".chrome-use").join("connect-mode.log");
+        use std::io::Write;
+        if let Ok(mut f) = std::fs::OpenOptions::new()
+            .create(true)
+            .append(true)
+            .open(&path)
+        {
+            let _ = f.write_all(line.as_bytes());
+        }
+    }
+}
+
 /// Sidecar recording the connected extension's version, written by the host when
 /// it receives the extension's `hello` (sibling of `relay-cdp-url`). Lets
 /// `doctor` surface which extension build is live without a CDP round-trip.

cli/src/main.rs

@@ -1369,6 +1369,20 @@ fn main() {
         && flags.provider.is_none()
         && (flags.force_launch || !flags.auto_connect)
     {
+        // Launching a debug-port Chrome pops Chrome's "Allow remote debugging?"
+        // consent modal (Chrome 136+). When the ab-connect relay is already up,
+        // this is almost always unintended — the relay drives the user's real
+        // Chrome with NO modal. Warn so the modal is self-explained and the
+        // caller (often a stray --launch / --no-auto-connect) is fixable (#32).
+        if !flags.json && connect::relay_url().is_some() {
+            eprintln!(
+                "{} launching a new Chrome with a debug port — this pops Chrome's \
+                 \"Allow remote debugging?\" modal.\n  The ab-connect relay is up; \
+                 drop --launch/--new (and don't pass --no-auto-connect) to drive your \
+                 real Chrome with no modal.",
+                color::warning_indicator()
+            );
+        }
         let mut launch_cmd = json!({
             "id": gen_id(),
             "action": "launch",

cli/src/native/actions.rs

@@ -1341,6 +1341,7 @@ pub async fn execute_command(cmd: &Value, state: &mut DaemonState) -> Value {
         "forward" => handle_forward(state).await,
         "reload" => handle_reload(state).await,
         "cookies_get" => handle_cookies_get(cmd, state).await,
+        "cf_status" => handle_cf_status(cmd, state).await,
         "cookies_set" => handle_cookies_set(cmd, state).await,
         "cookies_clear" => handle_cookies_clear(state).await,
         "storage_get" => handle_storage_get(cmd, state).await,
@@ -2999,6 +3000,14 @@ async fn handle_screenshot(cmd: &Value, state: &mut DaemonState) -> Result<Value
             .get("screenshotDir")
             .and_then(|v| v.as_str())
             .map(String::from),
+        clip: cmd.get("clip").and_then(|c| {
+            Some((
+                c.get("x")?.as_f64()?,
+                c.get("y")?.as_f64()?,
+                c.get("width")?.as_f64()?,
+                c.get("height")?.as_f64()?,
+            ))
+        }),
     };
 
     if annotate {
@@ -3216,6 +3225,14 @@ async fn handle_type(cmd: &Value, state: &mut DaemonState) -> Result<Value, Stri
     let mgr = state.browser.as_ref().ok_or("Browser not launched")?;
     let session_id = mgr.active_session_id()?.to_string();
 
+    // `--key-events`: dispatch real per-character keyDown/keyUp instead of
+    // Input.insertText, so autocomplete/combobox widgets that only react to key
+    // events fire (e.g. Google's address postal-code lookup) (issue #4/#36).
+    let key_events = cmd
+        .get("keyEvents")
+        .and_then(|v| v.as_bool())
+        .unwrap_or(false);
+
     // `type --focused <text>`: type into the currently-focused element without a
     // selector (custom widgets that move focus to a hidden input on open).
     if cmd
@@ -3227,7 +3244,14 @@ async fn handle_type(cmd: &Value, state: &mut DaemonState) -> Result<Value, Stri
             .get("text")
             .and_then(|v| v.as_str())
             .ok_or("Missing 'text' parameter")?;
-        interaction::type_text_into_active_context(&mgr.client, &session_id, text, None).await?;
+        interaction::type_text_into_active_context(
+            &mgr.client,
+            &session_id,
+            text,
+            None,
+            key_events,
+        )
+        .await?;
         return Ok(json!({ "typed": text, "focused": true }));
     }
 
@@ -3251,6 +3275,7 @@ async fn handle_type(cmd: &Value, state: &mut DaemonState) -> Result<Value, Stri
         clear,
         delay,
         &state.iframe_sessions,
+        key_events,
     )
     .await?;
     Ok(json!({ "typed": text }))
@@ -3458,17 +3483,182 @@ async fn handle_scroll(cmd: &Value, state: &mut DaemonState) -> Result<Value, St
         }
     }
 
+    // An explicit `--selector` keeps the precise element-scroll path (scrollBy on
+    // the resolved node, same-origin only).
+    if let Some(sel) = selector {
+        interaction::scroll(
+            &mgr.client,
+            &session_id,
+            &state.ref_map,
+            Some(sel),
+            dx,
+            dy,
+            &state.iframe_sessions,
+        )
+        .await?;
+        return Ok(json!({ "scrolled": true, "via": "selector" }));
+    }
+
+    // `--at x,y` / `--frame n`: dispatch a real (isTrusted) wheel at a viewport
+    // coordinate. This hits the compositor and scrolls whatever scroll container
+    // is under the pointer — including cross-origin iframes that `window.scrollBy`
+    // on the top document silently no-ops on (issue #36).
+    if cmd.get("at").is_some() || cmd.get("frame").is_some() {
+        let (x, y, via) = if let Some(at) = cmd.get("at").and_then(|v| v.as_array()) {
+            let x = at.first().and_then(|v| v.as_f64()).unwrap_or(0.0);
+            let y = at.get(1).and_then(|v| v.as_f64()).unwrap_or(0.0);
+            (x, y, "at")
+        } else {
+            let n = cmd.get("frame").and_then(|v| v.as_u64()).unwrap_or(0);
+            let (x, y) = frame_center(mgr, &session_id, &state.iframe_sessions, n as usize).await?;
+            (x, y, "frame")
+        };
+        dispatch_wheel(&mgr.client, &session_id, x, y, dx, dy).await?;
+        return Ok(json!({ "scrolled": true, "via": via, "at": [x, y] }));
+    }
+
+    // Default (no selector/at/frame): scroll the page with `window.scrollBy`. This
+    // is the reliable path for ordinary page scrolling; a coordinate wheel at the
+    // viewport centre is NOT a dependable substitute (it no-ops on some pages,
+    // e.g. headless), so the wheel stays opt-in via `--at`/`--frame` for the
+    // cross-origin-iframe case (issue #36).
     interaction::scroll(
         &mgr.client,
         &session_id,
         &state.ref_map,
-        selector,
+        None,
         dx,
         dy,
         &state.iframe_sessions,
     )
     .await?;
-    Ok(json!({ "scrolled": true }))
+    Ok(json!({ "scrolled": true, "via": "page" }))
+}
+
+/// Viewport center in CSS pixels, used as the default wheel landing point for
+/// `scroll` (issue #36). Falls back to a sane 640×400 center if the page can't
+/// be evaluated (e.g. a restricted document).
+async fn viewport_center(mgr: &BrowserManager, session_id: &str) -> Result<(f64, f64), String> {
+    let dims = mgr
+        .client
+        .send_command_typed::<_, Value>(
+            "Runtime.evaluate",
+            &super::cdp::types::EvaluateParams {
+                expression: "[window.innerWidth, window.innerHeight]".to_string(),
+                return_by_value: Some(true),
+                await_promise: Some(false),
+            },
+            Some(session_id),
+        )
+        .await
+        .ok();
+    let arr = dims
+        .as_ref()
+        .and_then(|v| v.get("result"))
+        .and_then(|v| v.get("value"))
+        .and_then(|v| v.as_array());
+    let w = arr
+        .and_then(|a| a.first())
+        .and_then(|v| v.as_f64())
+        .filter(|w| *w > 0.0)
+        .unwrap_or(1280.0);
+    let h = arr
+        .and_then(|a| a.get(1))
+        .and_then(|v| v.as_f64())
+        .filter(|h| *h > 0.0)
+        .unwrap_or(800.0);
+    Ok((w / 2.0, h / 2.0))
+}
+
+/// Center of the `n`-th frame (as listed by `chrome-use frames`) in top-viewport
+/// CSS pixels, so `scroll --frame n` lands its wheel inside a cross-origin iframe
+/// without needing a selector into it (issue #36). Resolves the frame's owning
+/// `<iframe>` element box via `DOM.getFrameOwner` + `DOM.getBoxModel` — exact for
+/// a frame nested directly under the top document; for a deeper nesting the box is
+/// relative to the intermediate frame, so prefer `--at x,y` from a screenshot.
+async fn frame_center(
+    mgr: &BrowserManager,
+    session_id: &str,
+    iframe_sessions: &HashMap<String, String>,
+    n: usize,
+) -> Result<(f64, f64), String> {
+    let frames =
+        super::element::collect_all_frames_text(&mgr.client, session_id, iframe_sessions).await?;
+    let frame = frames.get(n).ok_or_else(|| {
+        format!(
+            "frame index {} out of range (run `chrome-use frames`: {} frame(s))",
+            n,
+            frames.len()
+        )
+    })?;
+    if n == 0 {
+        // Frame 0 is the top document — there's no owner element; scroll its center.
+        return viewport_center(mgr, session_id).await;
+    }
+    let owner = mgr
+        .client
+        .send_command_typed::<_, Value>(
+            "DOM.getFrameOwner",
+            &json!({ "frameId": frame.frame_id }),
+            Some(session_id),
+        )
+        .await
+        .map_err(|e| format!("can't locate frame {}'s owner element: {}", n, e))?;
+    let backend_node_id = owner
+        .get("backendNodeId")
+        .and_then(|v| v.as_i64())
+        .ok_or_else(|| format!("frame {} has no owner <iframe> element", n))?;
+    let box_model = mgr
+        .client
+        .send_command_typed::<_, Value>(
+            "DOM.getBoxModel",
+            &json!({ "backendNodeId": backend_node_id }),
+            Some(session_id),
+        )
+        .await
+        .map_err(|e| format!("can't measure frame {}'s box: {}", n, e))?;
+    let content = box_model
+        .get("model")
+        .and_then(|m| m.get("content"))
+        .and_then(|c| c.as_array())
+        .ok_or_else(|| format!("frame {} box model has no content quad", n))?;
+    let coord = |i: usize| content.get(i).and_then(|v| v.as_f64()).unwrap_or(0.0);
+    // content quad is [x1,y1, x2,y2, x3,y3, x4,y4]; opposite corners are 0 and 2.
+    let cx = (coord(0) + coord(4)) / 2.0;
+    let cy = (coord(1) + coord(5)) / 2.0;
+    Ok((cx, cy))
+}
+
+/// Dispatch a trusted mouse wheel at `(x, y)`, humanized like `handle_wheel`.
+async fn dispatch_wheel(
+    client: &super::cdp::client::CdpClient,
+    session_id: &str,
+    x: f64,
+    y: f64,
+    delta_x: f64,
+    delta_y: f64,
+) -> Result<(), String> {
+    let level = humanize::active_level();
+    let seed = humanize::next_seed();
+    for (dx, dy, delay) in humanize::scroll_segments(delta_x, delta_y, level, seed) {
+        client
+            .send_command(
+                "Input.dispatchMouseEvent",
+                Some(json!({
+                    "type": "mouseWheel",
+                    "x": x,
+                    "y": y,
+                    "deltaX": dx,
+                    "deltaY": dy,
+                })),
+                Some(session_id),
+            )
+            .await?;
+        if !delay.is_zero() {
+            tokio::time::sleep(delay).await;
+        }
+    }
+    Ok(())
 }
 
 async fn handle_select(cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {
@@ -3665,12 +3855,9 @@ async fn handle_gettext(cmd: &Value, state: &mut DaemonState) -> Result<Value, S
 async fn handle_frames(_cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {
     let mgr = state.browser.as_ref().ok_or("Browser not launched")?;
     let session_id = mgr.active_session_id()?.to_string();
-    let frames = super::element::collect_all_frames_text(
-        &mgr.client,
-        &session_id,
-        &state.iframe_sessions,
-    )
-    .await?;
+    let frames =
+        super::element::collect_all_frames_text(&mgr.client, &session_id, &state.iframe_sessions)
+            .await?;
     let list: Vec<Value> = frames
         .iter()
         .enumerate()
@@ -4122,6 +4309,111 @@ async fn handle_cookies_clear(state: &DaemonState) -> Result<Value, String> {
     Ok(json!({ "cleared": true }))
 }
 
+// Detect whether the active page is *currently* a Cloudflare challenge
+// (the full-page "Just a moment…" / "正在进行安全验证" interstitial), so an agent
+// knows whether it must solve or can proceed. Runs in the top frame main world.
+const CF_CHALLENGE_JS: &str = r#"(function(){
+  var t = document.title || '';
+  var challenged =
+    /just a moment|attention required|checking (your|if)|verify you are human|正在进行安全验证|安全验证|请稍候|请完成|人机验证/i.test(t) ||
+    !!document.querySelector('#challenge-form, #challenge-running, #cf-challenge-running, [id^="cf-chl"], script[src*="/cdn-cgi/challenge-platform/"]');
+  var turnstile = !!document.querySelector('.cf-turnstile, [data-sitekey]');
+  return JSON.stringify({ title: t, challenged: challenged, turnstile: turnstile, readyState: document.readyState });
+})()"#;
+
+/// Recommendation for a Cloudflare-gated page, from the current challenge state
+/// and whether a still-valid `cf_clearance` exists. Pure so it's unit-testable.
+/// - not challenged            → "proceed" (the page is cleared/loaded)
+/// - challenged, valid cookie  → "reissue" (clearance present but page still
+///   blocks → it's stale or the IP/UA no longer matches what it was issued for)
+/// - challenged, no cookie     → "solve"
+fn cf_recommendation(challenged: bool, clearance_valid: bool) -> &'static str {
+    if !challenged {
+        "proceed"
+    } else if clearance_valid {
+        "reissue"
+    } else {
+        "solve"
+    }
+}
+
+/// `cf_clearance` validity for a cookie's expiry (epoch seconds; <=0 = session
+/// cookie, treated as non-expiring). Returns (present, expired). Pure.
+fn clearance_state(expires: Option<f64>, now: f64) -> (bool, bool) {
+    match expires {
+        None => (false, false),
+        Some(e) if e <= 0.0 => (true, false), // session cookie: no expiry
+        Some(e) => (true, e < now),
+    }
+}
+
+async fn handle_cf_status(_cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {
+    let mgr = state.browser.as_ref().ok_or("Browser not launched")?;
+    let session_id = mgr.active_session_id()?.to_string();
+    let url = mgr.get_url().await.unwrap_or_default();
+
+    // 1. Is the page a Cloudflare challenge right now?
+    let probe_raw = mgr
+        .evaluate(CF_CHALLENGE_JS, None)
+        .await
+        .unwrap_or(Value::Null);
+    let probe = parse_json_string(probe_raw, "cf challenge probe").unwrap_or(Value::Null);
+    let challenged = probe
+        .get("challenged")
+        .and_then(|v| v.as_bool())
+        .unwrap_or(false);
+    let turnstile = probe
+        .get("turnstile")
+        .and_then(|v| v.as_bool())
+        .unwrap_or(false);
+    let title = probe
+        .get("title")
+        .and_then(|v| v.as_str())
+        .unwrap_or("")
+        .to_string();
+
+    // 2. Persistence artifacts: cf_clearance (HttpOnly → must read via CDP, not
+    //    document.cookie) + CF_VERIFIED_DEVICE. Scope to the current URL.
+    let urls = if url.is_empty() {
+        None
+    } else {
+        Some(vec![url.clone()])
+    };
+    let cookies = super::cookies::get_cookies(&mgr.client, &session_id, urls)
+        .await
+        .unwrap_or_default();
+    let clearance = cookies.iter().find(|c| c.name == "cf_clearance");
+    let now = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .map(|d| d.as_secs_f64())
+        .unwrap_or(0.0);
+    let (present, expired) = clearance_state(clearance.map(|c| c.expires), now);
+    let expires_in = clearance
+        .filter(|_| present && !expired)
+        .map(|c| (c.expires - now).max(0.0) as i64);
+    let device_verified = cookies
+        .iter()
+        .any(|c| c.name.starts_with("CF_VERIFIED_DEVICE"));
+
+    let clearance_valid = present && !expired;
+    let recommendation = cf_recommendation(challenged, clearance_valid);
+
+    Ok(json!({
+        "url": url,
+        "title": title,
+        "challenged": challenged,
+        "turnstile": turnstile,
+        "clearance": {
+            "present": present,
+            "expired": expired,
+            "expiresIn": expires_in,
+            "httpOnly": clearance.map(|c| c.http_only).unwrap_or(false),
+        },
+        "deviceVerified": device_verified,
+        "recommendation": recommendation,
+    }))
+}
+
 async fn handle_storage_get(cmd: &Value, state: &DaemonState) -> Result<Value, String> {
     let mgr = state.browser.as_ref().ok_or("Browser not launched")?;
     let session_id = mgr.active_session_id()?.to_string();
@@ -4459,8 +4751,18 @@ async fn handle_keyboard(cmd: &Value, state: &DaemonState) -> Result<Value, Stri
                 .get("text")
                 .and_then(|v| v.as_str())
                 .ok_or("Missing 'text' parameter")?;
-            interaction::type_text_into_active_context(&mgr.client, &session_id, text, None)
-                .await?;
+            let key_events = cmd
+                .get("keyEvents")
+                .and_then(|v| v.as_bool())
+                .unwrap_or(false);
+            interaction::type_text_into_active_context(
+                &mgr.client,
+                &session_id,
+                text,
+                None,
+                key_events,
+            )
+            .await?;
             return Ok(json!({ "typed": text }));
         }
         Some("insertText") => {
@@ -6989,6 +7291,28 @@ async fn handle_drag(cmd: &Value, state: &mut DaemonState) -> Result<Value, Stri
         .and_then(|v| v.as_str())
         .ok_or("Missing 'target' parameter")?;
 
+    // Over the relay (or into an iframe) a coordinate drag drifts to the
+    // foreground tab and can't reach an OOPIF — DOM-dispatch an HTML5 drag in the
+    // element's own session instead (issues #31/#36). `coord` mode forces the
+    // coordinate path for pointer-driven drags (canvas/sliders) on a launched
+    // browser.
+    if std::env::var("AGENT_BROWSER_CLICK_MODE").as_deref() != Ok("coord")
+        && (crate::connect::relay_url().is_some()
+            || state.ref_map.ref_is_in_iframe(source)
+            || state.ref_map.ref_is_in_iframe(target))
+    {
+        super::interaction::dom_drag(
+            &mgr.client,
+            &session_id,
+            &state.ref_map,
+            source,
+            target,
+            &state.iframe_sessions,
+        )
+        .await?;
+        return Ok(json!({ "dragged": { "source": source, "target": target }, "via": "dom" }));
+    }
+
     let (sx, sy, _, _, source_session_id) = super::element::resolve_element_center(
         &mgr.client,
         &session_id,
@@ -7343,6 +7667,7 @@ async fn handle_diff_screenshot(cmd: &Value, state: &DaemonState) -> Result<Valu
         quality: None,
         annotate: false,
         output_dir: None,
+        clip: None,
     };
 
     let result = screenshot::take_screenshot(
@@ -9069,6 +9394,26 @@ mod tests {
     use crate::test_utils::EnvGuard;
     use std::fs;
 
+    #[test]
+    fn test_cf_recommendation() {
+        assert_eq!(cf_recommendation(false, false), "proceed");
+        assert_eq!(cf_recommendation(false, true), "proceed");
+        assert_eq!(cf_recommendation(true, false), "solve");
+        assert_eq!(cf_recommendation(true, true), "reissue");
+    }
+
+    #[test]
+    fn test_clearance_state() {
+        // no cookie
+        assert_eq!(clearance_state(None, 1000.0), (false, false));
+        // session cookie (expires <= 0) → present, never expired
+        assert_eq!(clearance_state(Some(-1.0), 1000.0), (true, false));
+        // valid: expiry in the future
+        assert_eq!(clearance_state(Some(2000.0), 1000.0), (true, false));
+        // expired: expiry in the past
+        assert_eq!(clearance_state(Some(500.0), 1000.0), (true, true));
+    }
+
     #[test]
     fn test_url_glob_to_regex() {
         assert_eq!(url_glob_to_regex("**/dashboard"), "^.*/dashboard$");

cli/src/native/browser.rs

@@ -121,7 +121,7 @@ fn normalize_url_for_match(url: &str) -> String {
 fn update_page_target_info_in_pages(pages: &mut [PageInfo], target: &TargetInfo) -> bool {
     if let Some(page) = pages.iter_mut().find(|p| p.target_id == target.target_id) {
         page.url = target.url.clone();
-        page.title = target.title.clone();
+        page.title = sanitize_title(&target.title);
         page.target_type = target.target_type.clone();
         return true;
     }
@@ -166,6 +166,112 @@ fn resolve_active_index(
     active_page_index
 }
 
+/// Strip zero-width / invisible / bidi-format Unicode from a page title before
+/// we store it. Some sites prepend runs of ZWJ / word-joiner / invisible-times /
+/// BOM to `document.title` (badging, watermarking, anti-scrape); left in, they
+/// pollute `tab list`, break text matching, and wreck column alignment (#33).
+fn sanitize_title(s: &str) -> String {
+    s.chars()
+        .filter(|&c| {
+            !matches!(c as u32,
+                0x00AD            // soft hyphen
+                | 0x200B..=0x200F // ZWSP, ZWNJ, ZWJ, LRM, RLM
+                | 0x2028 | 0x2029 // line / paragraph separators
+                | 0x202A..=0x202E // bidi embedding/override
+                | 0x2060..=0x2064 // word joiner, invisible operators
+                | 0x2066..=0x2069 // bidi isolates
+                | 0x180E          // Mongolian vowel separator
+                | 0xFEFF          // BOM / ZW no-break space
+            )
+        })
+        .collect::<String>()
+        .trim()
+        .to_string()
+}
+
+/// Best-effort MIME type from a filename extension, for the relay file-upload
+/// fallback (the page-constructed `File` needs a sensible `type`). Covers the
+/// common upload kinds; anything unknown falls back to a generic binary type.
+fn mime_for_path(name: &str) -> &'static str {
+    let ext = name.rsplit('.').next().unwrap_or("").to_lowercase();
+    match ext.as_str() {
+        "png" => "image/png",
+        "jpg" | "jpeg" => "image/jpeg",
+        "gif" => "image/gif",
+        "webp" => "image/webp",
+        "svg" => "image/svg+xml",
+        "bmp" => "image/bmp",
+        "pdf" => "application/pdf",
+        "txt" => "text/plain",
+        "csv" => "text/csv",
+        "json" => "application/json",
+        "mp4" => "video/mp4",
+        "webm" => "video/webm",
+        "mov" => "video/quicktime",
+        "mp3" => "audio/mpeg",
+        "zip" => "application/zip",
+        _ => "application/octet-stream",
+    }
+}
+
+/// Target ids to prune after a `Target.getTargets` resync: tracked pages whose
+/// target is no longer in the live set — EXCEPT the explicitly-pinned active
+/// target, which is protected. The relay against a busy real Chrome occasionally
+/// returns a different window's tabs for a single `getTargets` call ("tab list
+/// hops windows", issue #31); pruning on that transient snapshot would drop the
+/// agent's adopted tab and drift subsequent eval/click onto a foreign tab. A
+/// genuine close still arrives as `Target.targetDestroyed` (handled in the event
+/// drain), which removes the pin properly — so protecting it here only guards
+/// against flaky snapshots, not real closures.
+fn prunable_target_ids(
+    pages: &[PageInfo],
+    live_ids: &HashSet<String>,
+    pinned: Option<&str>,
+) -> Vec<String> {
+    pages
+        .iter()
+        .map(|p| p.target_id.clone())
+        .filter(|tid| !live_ids.contains(tid) && pinned != Some(tid.as_str()))
+        .collect()
+}
+
+/// Consecutive missing `getTargets` snapshots before an owned relay tab is
+/// pruned. >1 so a single churning/partial snapshot (other agents opening/closing
+/// tabs) or a brief cross-process-nav gap can't drop the tab the agent is driving.
+const RELAY_PRUNE_MISSES: u32 = 3;
+
+/// Debounced prune for the relay: target ids to drop, mutating per-target miss
+/// counters. A tab in `live_ids` resets to 0; an absent (non-pinned) tab
+/// increments and is pruned only at `RELAY_PRUNE_MISSES`. Counters for
+/// no-longer-tracked targets are forgotten. Pure, so the multi-agent churn
+/// tolerance is unit-testable without a live browser.
+fn debounced_prune_ids(
+    pages: &[PageInfo],
+    live_ids: &HashSet<String>,
+    pinned: Option<&str>,
+    misses: &mut HashMap<String, u32>,
+) -> Vec<String> {
+    let tracked: HashSet<&str> = pages.iter().map(|p| p.target_id.as_str()).collect();
+    misses.retain(|tid, _| tracked.contains(tid.as_str()));
+    let mut prune = Vec::new();
+    for p in pages {
+        let tid = p.target_id.as_str();
+        if live_ids.contains(tid) {
+            misses.remove(tid);
+            continue;
+        }
+        if pinned == Some(tid) {
+            continue;
+        }
+        let c = misses.entry(p.target_id.clone()).or_insert(0);
+        *c += 1;
+        if *c >= RELAY_PRUNE_MISSES {
+            prune.push(p.target_id.clone());
+        }
+    }
+    prune
+}
+
 /// Whether the resolved active page is a tab the session created (its target_id
 /// is in `created_targets`). Pure core of [`BrowserManager::active_is_session_owned`]
 /// so the relay no-hijack rule is unit-testable without a live browser.
@@ -185,6 +291,21 @@ fn active_index_is_owned(
         .unwrap_or(false)
 }
 
+/// Whether a CDP error means the bound relay target is gone — the tab was
+/// closed, navigated across processes (renderer swap), or lost after an
+/// extension/service-worker restart, and the relay could not re-attach. The
+/// ab-connect relay surfaces these as `stale sessionId … its tab is gone`,
+/// `unknown sessionId …`, or `no attached tab …`. `navigate` keys its
+/// auto-reattach recovery off this (issue #35) so a dead session rebinds to a
+/// fresh tab instead of erroring on every command until the user runs `tab new`.
+fn is_stale_target_error(error: &str) -> bool {
+    let lower = error.to_lowercase();
+    lower.contains("its tab is gone")
+        || lower.contains("stale sessionid")
+        || lower.contains("unknown sessionid")
+        || lower.contains("no attached tab")
+}
+
 /// Converts common error messages into AI-friendly, actionable descriptions.
 pub fn to_ai_friendly_error(error: &str) -> String {
     let lower = error.to_lowercase();
@@ -380,6 +501,14 @@ pub struct BrowserManager {
     /// the session's commands onto the wrong page — the wrong-origin-fetch hazard
     /// in the dogfood reports. Falls back to the index if the pinned tab is gone.
     active_target_id: Option<String>,
+    /// Per-target count of CONSECUTIVE `resync_targets` snapshots in which an
+    /// owned tab was missing from `Target.getTargets`. Over the relay a single
+    /// snapshot routinely omits live tabs (multi-agent churn, a cross-process nav
+    /// briefly dropping the target), so we must not prune on one miss — that lost
+    /// the tab the agent was driving. A tab is removed only after it's been absent
+    /// for `RELAY_PRUNE_MISSES` consecutive snapshots; any snapshot that includes
+    /// it resets the counter. Keyed by stable target_id.
+    relay_target_misses: HashMap<String, u32>,
     next_tab_id: u32,
     /// Whether to enable the CDP `Runtime` domain (console / error / exception capture).
     /// OFF by default for stealth: a live `Runtime.enable` is a detectable CDP signal
@@ -490,6 +619,16 @@ impl BrowserManager {
             }
         };
 
+        // A launched browser carries a debug port → it's the other path that can
+        // pop Chrome's consent modal; record it for #31 diagnosis.
+        crate::connect::log_connect_mode(
+            &ws_url,
+            true,
+            DAEMON_SESSION
+                .get()
+                .map(String::as_str)
+                .unwrap_or("default"),
+        );
         let manager = if engine == "lightpanda" {
             initialize_lightpanda_manager(ws_url, process).await?
         } else {
@@ -506,6 +645,7 @@ impl BrowserManager {
                 visited_origins: HashSet::new(),
                 created_targets: HashSet::new(),
                 active_target_id: None,
+                relay_target_misses: HashMap::new(),
                 next_tab_id: 1,
                 capture_console: console_capture_enabled(),
             };
@@ -585,6 +725,16 @@ impl BrowserManager {
         headers: Option<Vec<(String, String)>>,
     ) -> Result<Self, String> {
         let ws_url = resolve_cdp_url(url).await?;
+        // Record the transport so a reappearing "Allow remote debugging?" modal
+        // can be traced to a raw-port attach vs the consent-free relay (#31).
+        crate::connect::log_connect_mode(
+            &ws_url,
+            false,
+            DAEMON_SESSION
+                .get()
+                .map(String::as_str)
+                .unwrap_or("default"),
+        );
         let client = Arc::new(CdpClient::connect_with_headers(&ws_url, headers).await?);
         let mut manager = Self {
             client,
@@ -598,6 +748,7 @@ impl BrowserManager {
             visited_origins: HashSet::new(),
             created_targets: HashSet::new(),
             active_target_id: None,
+            relay_target_misses: HashMap::new(),
             next_tab_id: 1,
             capture_console: console_capture_enabled(),
         };
@@ -627,6 +778,43 @@ impl BrowserManager {
         Self::connect_cdp(&ws_url).await
     }
 
+    /// Page targets to adopt, merging several `Target.getTargets` snapshots over
+    /// the extension relay. A single relay snapshot is flaky on a busy real Chrome
+    /// — it can omit live tabs (a different window's set, or a partial list; issue
+    /// #31) — so a tab the daemon should adopt would silently vanish (e.g. after a
+    /// daemon restart the page being driven disappeared from the tab list). Taking
+    /// the union of a few snapshots makes adoption resilient to a transient miss.
+    /// Off the relay (a browser we launched) one snapshot is authoritative.
+    async fn collect_page_targets(&self) -> Result<Vec<TargetInfo>, String> {
+        let rounds = if crate::connect::relay_url().is_some() {
+            3
+        } else {
+            1
+        };
+        let mut by_id: HashMap<String, TargetInfo> = HashMap::new();
+        let mut any_ok = false;
+        for i in 0..rounds {
+            if i > 0 {
+                tokio::time::sleep(Duration::from_millis(150)).await;
+            }
+            match self
+                .client
+                .send_command_typed::<_, GetTargetsResult>("Target.getTargets", &json!({}), None)
+                .await
+            {
+                Ok(result) => {
+                    any_ok = true;
+                    for t in result.target_infos.into_iter().filter(should_track_target) {
+                        by_id.entry(t.target_id.clone()).or_insert(t);
+                    }
+                }
+                Err(e) if i == rounds - 1 && !any_ok => return Err(e),
+                Err(_) => {}
+            }
+        }
+        Ok(by_id.into_values().collect())
+    }
+
     async fn discover_and_attach_targets(&mut self) -> Result<(), String> {
         self.client
             .send_command_typed::<_, Value>(
@@ -636,16 +824,7 @@ impl BrowserManager {
             )
             .await?;
 
-        let result: GetTargetsResult = self
-            .client
-            .send_command_typed("Target.getTargets", &json!({}), None)
-            .await?;
-
-        let page_targets: Vec<TargetInfo> = result
-            .target_infos
-            .into_iter()
-            .filter(should_track_target)
-            .collect();
+        let page_targets: Vec<TargetInfo> = self.collect_page_targets().await?;
 
         if page_targets.is_empty() {
             // Create a new tab
@@ -691,7 +870,19 @@ impl BrowserManager {
             self.active_page_index = 0;
             self.pin_active_target();
             self.enable_domains(&attach_result.session_id).await?;
+        } else if self.agent_group().is_some() {
+            // STRICT MULTI-AGENT ISOLATION (relay / the user's real Chrome).
+            // `page_targets` here are the USER's and OTHER agents' tabs. A tab
+            // group belongs to exactly ONE agent, so this session must NOT adopt
+            // any of them — it tracks ONLY tabs it creates (its own colored group)
+            // plus popups it opens. Adopting foreign tabs is precisely what let
+            // another concurrent agent's tab churn drop the tab we were driving and
+            // drift eval/click onto the wrong page (multi-agent failure). Open our
+            // own dedicated background tab in the session's group and pin it; the
+            // user's / other agents' tabs stay invisible to us.
+            self.tab_new(None, None).await?;
         } else {
+            // A browser WE launched: every tab is ours, so adopt them all.
             for target in &page_targets {
                 let attach_result: AttachToTargetResult = self
                     .client
@@ -713,11 +904,10 @@ impl BrowserManager {
                     target_id: target.target_id.clone(),
                     session_id: attach_result.session_id.clone(),
                     url: target.url.clone(),
-                    title: target.title.clone(),
+                    title: sanitize_title(&target.title),
                     target_type: target.target_type.clone(),
                 });
             }
-
             self.active_page_index = 0;
             self.pin_active_target();
             let session_id = self.pages[0].session_id.clone();
@@ -832,6 +1022,24 @@ impl BrowserManager {
         )
     }
 
+    /// Drop the page bound to `session_id` from the tracked list — used when the
+    /// relay reports its tab is gone (issue #35) so the stale entry can't keep
+    /// resolving as active. Forgets ownership, unpins it if it was pinned, and
+    /// keeps `active_page_index` in range.
+    fn drop_page_by_session(&mut self, session_id: &str) {
+        let Some(pos) = self.pages.iter().position(|p| p.session_id == session_id) else {
+            return;
+        };
+        let target_id = self.pages[pos].target_id.clone();
+        self.pages.remove(pos);
+        self.created_targets.remove(&target_id);
+        if self.active_target_id.as_deref() == Some(target_id.as_str()) {
+            self.active_target_id = None;
+        }
+        self.active_page_index =
+            active_page_index_after_removal(self.active_page_index, pos, self.pages.len());
+    }
+
     /// Pin the current active page by target_id so later commands stick to it.
     /// Call after any explicit open / tab new / tab switch.
     fn pin_active_target(&mut self) {
@@ -861,10 +1069,10 @@ impl BrowserManager {
         if self.agent_group().is_some() && !self.active_is_session_owned() {
             self.tab_new(None, None).await?;
         }
-        let session_id = self.active_session_id()?.to_string();
+        let mut session_id = self.active_session_id()?.to_string();
         let mut lifecycle_rx = self.client.subscribe();
 
-        let nav_result: PageNavigateResult = self
+        let nav_result: PageNavigateResult = match self
             .client
             .send_command_typed(
                 "Page.navigate",
@@ -874,9 +1082,51 @@ impl BrowserManager {
                 },
                 Some(&session_id),
             )
-            .await?;
+            .await
+        {
+            Ok(r) => r,
+            // Auto-reattach when the bound tab is gone (issue #35). On the shared
+            // real browser the human can close/swap the agent's tab, and a
+            // cross-process nav can destroy the target without a re-attachable
+            // tabId — both leave the cached `cb-tab-<id>` session stale, so every
+            // command (including `open`) failed on it and only `tab new`
+            // recovered. The relay error literally says "re-open your target URL
+            // to re-attach"; fulfil that here: drop the dead page, open a fresh
+            // owned tab in this session's group, and navigate THAT. Gated on the
+            // relay (`agent_group`) and on the explicit navigation intent — read
+            // commands deliberately still fail loudly rather than silently
+            // recover onto a blank tab and return wrong data (issue #8.1).
+            Err(e) if self.agent_group().is_some() && is_stale_target_error(&e) => {
+                self.drop_page_by_session(&session_id);
+                self.tab_new(None, None).await?;
+                session_id = self.active_session_id()?.to_string();
+                lifecycle_rx = self.client.subscribe();
+                self.client
+                    .send_command_typed(
+                        "Page.navigate",
+                        &PageNavigateParams {
+                            url: url.to_string(),
+                            referrer: None,
+                        },
+                        Some(&session_id),
+                    )
+                    .await?
+            }
+            Err(e) => return Err(e),
+        };
 
         if let Some(ref error_text) = nav_result.error_text {
+            // `data:` URLs abort over the extension relay: chrome.debugger /
+            // chrome.tabs can't drive a top-frame data: navigation, so it comes
+            // back net::ERR_ABORTED on an about:blank tab. Explain it instead of
+            // leaking the cryptic code (data: works fine under `--launch`).
+            if url.starts_with("data:") && error_text.contains("ERR_ABORTED") {
+                return Err(format!(
+                    "Navigation failed: {error_text}. Chrome blocks top-frame `data:` URLs over \
+                     the extension relay — use a real http(s):// or file:// URL, or run with \
+                     `--launch` (where data: URLs work)."
+                ));
+            }
             return Err(format!("Navigation failed: {}", error_text));
         }
 
@@ -939,7 +1189,7 @@ impl BrowserManager {
         self.active_page_index = self.resolved_active_index();
         if let Some(page) = self.pages.get_mut(self.active_page_index) {
             page.url = page_url.clone();
-            page.title = title.clone();
+            page.title = sanitize_title(&title);
         }
         self.pin_active_target();
 
@@ -1001,7 +1251,7 @@ impl BrowserManager {
 
     pub async fn get_title(&self) -> Result<String, String> {
         let result = self.evaluate_simple("document.title").await?;
-        Ok(result.as_str().unwrap_or("").to_string())
+        Ok(sanitize_title(result.as_str().unwrap_or("")))
     }
 
     pub async fn get_content(&self) -> Result<String, String> {
@@ -1014,15 +1264,31 @@ impl BrowserManager {
     pub async fn evaluate(&self, script: &str, _args: Option<Value>) -> Result<Value, String> {
         let session_id = self.active_session_id()?.to_string();
 
+        // `replMode: true` lets successive `eval`s re-declare top-level
+        // `let`/`const` instead of throwing "Identifier 'x' has already been
+        // declared" (issue #38 — independent `eval` steps in a test suite collided
+        // in the page's shared lexical scope). BUT replMode and `awaitPromise` are
+        // mutually exclusive in Chrome: under replMode a returned promise is NOT
+        // awaited (it serialises to `{}`), which breaks `fetch(...).then(...)` and
+        // every other async eval. So enable replMode ONLY for synchronous scripts
+        // that declare a top-level `let`/`const`; promise-returning scripts keep
+        // `awaitPromise` (no replMode) — exactly the pre-#38 behaviour.
+        let mentions_async = script.contains("await")
+            || script.contains(".then(")
+            || script.contains("fetch(")
+            || script.contains("Promise");
+        let declares = script.contains("let ") || script.contains("const ");
+        let repl_mode = declares && !mentions_async;
         let result: EvaluateResult = self
             .client
             .send_command_typed(
                 "Runtime.evaluate",
-                &EvaluateParams {
-                    expression: script.to_string(),
-                    return_by_value: Some(true),
-                    await_promise: Some(true),
-                },
+                &json!({
+                    "expression": script,
+                    "returnByValue": true,
+                    "awaitPromise": !repl_mode,
+                    "replMode": repl_mode,
+                }),
                 Some(&session_id),
             )
             .await?;
@@ -1305,6 +1571,18 @@ impl BrowserManager {
     /// the active tab, per #7/#8.1); the caller surfaces it so the agent knows a
     /// tab opened instead of seeing the old page (issue #24-A).
     pub async fn adopt_newly_opened(&mut self, before: &HashSet<String>) -> Option<PageInfo> {
+        // STRICT MULTI-AGENT ISOLATION: on the relay this session's `before` set is
+        // only its OWN tabs, so EVERY foreign tab (the user's, other agents') looks
+        // "new" relative to it and would be adopted here — exactly the leak where a
+        // concurrent agent's tabs (github/Lark/iphone-use) showed up in this
+        // session mid-flow. A tab the agent itself opened (a pop-up) can't be
+        // distinguished from a foreign tab over the relay (no opener/window/group
+        // in the synthesized targetInfo), so don't adopt anything: the agent drives
+        // only tabs it explicitly created, and pop-ups (e.g. an OAuth/login window)
+        // are the user's. A launched browser (every tab ours) still follows pop-ups.
+        if self.agent_group().is_some() {
+            return None;
+        }
         let result: GetTargetsResult = self
             .client
             .send_command_typed("Target.getTargets", &json!({}), None)
@@ -1344,9 +1622,14 @@ impl BrowserManager {
                 target_id: target.target_id.clone(),
                 session_id: attach.session_id.clone(),
                 url: target.url.clone(),
-                title: target.title.clone(),
+                title: sanitize_title(&target.title),
                 target_type: target.target_type.clone(),
             };
+            // A tab that appeared right after THIS session's action (a click that
+            // opened a popup/new tab) is ours — record it as owned so it's tracked,
+            // protected from churn-pruning, and cleaned up on close, consistent with
+            // strict multi-agent isolation (we only ever own tabs we created/opened).
+            self.created_targets.insert(target.target_id.clone());
             self.add_background_page(page.clone());
             let _ = self.enable_domains(&attach.session_id).await;
             if opened.is_none() {
@@ -1374,13 +1657,19 @@ impl BrowserManager {
             .filter(should_track_target)
             .collect();
         let live_ids: HashSet<String> = live.iter().map(|t| t.target_id.clone()).collect();
+        let on_relay = self.agent_group().is_some();
 
         for target in &live {
             if self.update_page_target_info(target) {
                 continue;
             }
-            // A target this session hasn't tracked yet — attach and add it in the
-            // background so it's listable/adoptable without stealing the active tab.
+            // STRICT MULTI-AGENT ISOLATION: on the relay (the user's real Chrome,
+            // shared with other agents), NEVER adopt a tab this session didn't
+            // create — it belongs to the user or another agent's group. Only a
+            // browser we launched (every tab ours) adopts unknown targets.
+            if on_relay {
+                continue;
+            }
             let attach_result: AttachToTargetResult = match self
                 .client
                 .send_command_typed(
@@ -1405,21 +1694,32 @@ impl BrowserManager {
                 target_id: target.target_id.clone(),
                 session_id: attach_result.session_id.clone(),
                 url: target.url.clone(),
-                title: target.title.clone(),
+                title: sanitize_title(&target.title),
                 target_type: target.target_type.clone(),
             });
             let _ = self.enable_domains(&attach_result.session_id).await;
         }
 
-        // Drop tabs that no longer exist so `tab list` doesn't show phantom rows.
-        let gone: Vec<String> = self
-            .pages
-            .iter()
-            .map(|p| p.target_id.clone())
-            .filter(|tid| !live_ids.contains(tid))
-            .collect();
-        for tid in gone {
-            self.remove_page_by_target_id(&tid);
+        // Prune tabs that are gone. On a LAUNCHED browser a missing target really
+        // is closed, so prune immediately. On the RELAY a single `getTargets`
+        // snapshot routinely omits live tabs (multi-agent churn, a brief
+        // cross-process-nav gap) — dropping the tab we're driving on one bad
+        // snapshot is the failure we're fixing — so prune only after the tab has
+        // been absent for several CONSECUTIVE snapshots (debounced). The pinned
+        // active target is protected either way (issue #31).
+        let gone = if on_relay {
+            debounced_prune_ids(
+                &self.pages,
+                &live_ids,
+                self.active_target_id.as_deref(),
+                &mut self.relay_target_misses,
+            )
+        } else {
+            prunable_target_ids(&self.pages, &live_ids, self.active_target_id.as_deref())
+        };
+        for tid in &gone {
+            self.relay_target_misses.remove(tid);
+            self.remove_page_by_target_id(tid);
         }
 
         // Refresh url/title from each live tab. The relay only stamps target_info
@@ -1451,7 +1751,7 @@ impl BrowserManager {
                             }
                         }
                         if let Some(t) = ti.get("title").and_then(|v| v.as_str()) {
-                            page.title = t.to_string();
+                            page.title = sanitize_title(t);
                         }
                     }
                 }
@@ -1639,7 +1939,7 @@ impl BrowserManager {
 
         if let Some(page) = self.pages.get_mut(index) {
             page.url = url.clone();
-            page.title = title.clone();
+            page.title = sanitize_title(&title);
         }
 
         let page = &self.pages[index];
@@ -1894,7 +2194,8 @@ impl BrowserManager {
             .and_then(|v| v.as_i64())
             .ok_or("Could not get backendNodeId for file input")?;
 
-        self.client
+        let set_files = self
+            .client
             .send_command(
                 "DOM.setFileInputFiles",
                 Some(json!({
@@ -1903,26 +2204,153 @@ impl BrowserManager {
                 })),
                 Some(&effective_session_id),
             )
-            .await
-            .map_err(|e| {
-                // Chrome's chrome.debugger API (the extension-relay transport)
-                // forbids DOM.setFileInputFiles for security, surfacing as an
-                // opaque `-32000 "Not allowed"`. Translate it into an actionable
-                // message rather than leaking the raw CDP error (issue #13).
-                if e.contains("Not allowed") || e.contains("-32000") {
-                    "file upload isn't supported over the extension relay — \
-                     Chrome's chrome.debugger API forbids DOM.setFileInputFiles. \
-                     Use a direct-CDP session instead: \
-                     `chrome-use --session up --launch open <url>` (carry your \
-                     login over with `cookies export` | `cookies set --curl`), \
-                     then run `upload` in that session. \
-                     See https://github.com/leeguooooo/chrome-use/issues/13"
-                        .to_string()
-                } else {
-                    e
-                }
-            })?;
+            .await;
 
+        if let Err(e) = set_files {
+            // Chrome's chrome.debugger API (the extension-relay transport) forbids
+            // DOM.setFileInputFiles for security, surfacing as an opaque
+            // `-32000 "Not allowed"`. Fall back to constructing the File entirely
+            // IN THE PAGE and assigning it to the input — the standard
+            // Playwright/Cypress trick, which needs no privileged CDP and so works
+            // over the relay (issue #13).
+            if e.contains("Not allowed") || e.contains("-32000") {
+                return self
+                    .upload_files_via_page(object_id, files, &effective_session_id)
+                    .await;
+            }
+            return Err(e);
+        }
+
+        Ok(())
+    }
+
+    /// Relay-safe file upload: read each file locally, hand its bytes to the page
+    /// as base64, and rebuild a `File` there — then either assign it to a file
+    /// `<input>` (Chrome allows `input.files = dataTransfer.files`) or, for a
+    /// dropzone/composer, dispatch synthetic `paste`/`drop` events carrying the
+    /// `DataTransfer`. No `DOM.setFileInputFiles`, so chrome.debugger permits it.
+    async fn upload_files_via_page(
+        &self,
+        object_id: String,
+        files: &[String],
+        session_id: &str,
+    ) -> Result<(), String> {
+        use base64::Engine;
+        // The relay tunnels every CDP message through Chrome native messaging,
+        // which caps a single message at ~1 MiB. A whole image's base64 blows
+        // past that ("CDP response channel closed"), so we STREAM the bytes into
+        // a page-side buffer in sub-limit chunks, then assemble the File from it.
+        const CHUNK: usize = 96 * 1024; // base64 chars per message; safe under 1 MiB
+
+        // Reset the staging buffer.
+        self.client
+            .send_command(
+                "Runtime.evaluate",
+                Some(json!({ "expression": "window.__cuUpload = [];", "returnByValue": true })),
+                Some(session_id),
+            )
+            .await
+            .map_err(|e| format!("relay upload (reset) failed: {}", e))?;
+
+        for path in files {
+            let bytes = std::fs::read(path).map_err(|e| format!("cannot read {}: {}", path, e))?;
+            let name = std::path::Path::new(path)
+                .file_name()
+                .and_then(|n| n.to_str())
+                .unwrap_or("upload.bin")
+                .to_string();
+            let mime = mime_for_path(&name);
+            let b64 = base64::engine::general_purpose::STANDARD.encode(&bytes);
+
+            // Push the file's metadata with an empty buffer.
+            let init = format!(
+                "window.__cuUpload.push({{ name: {}, type: {}, b64: '' }});",
+                serde_json::to_string(&name).unwrap_or_default(),
+                serde_json::to_string(mime).unwrap_or_default(),
+            );
+            self.client
+                .send_command(
+                    "Runtime.evaluate",
+                    Some(json!({ "expression": init, "returnByValue": true })),
+                    Some(session_id),
+                )
+                .await
+                .map_err(|e| format!("relay upload (init) failed: {}", e))?;
+
+            // Stream the base64 in chunks. base64's alphabet (A–Za–z0–9+/=) needs
+            // no escaping inside a single-quoted JS string, so concatenation is safe.
+            let idx = "window.__cuUpload[window.__cuUpload.length-1].b64";
+            let mut start = 0;
+            while start < b64.len() {
+                let end = (start + CHUNK).min(b64.len());
+                let chunk = &b64[start..end];
+                let expr = format!("{idx} += '{chunk}';");
+                self.client
+                    .send_command(
+                        "Runtime.evaluate",
+                        Some(json!({ "expression": expr, "returnByValue": true })),
+                        Some(session_id),
+                    )
+                    .await
+                    .map_err(|e| format!("relay upload (chunk) failed: {}", e))?;
+                start = end;
+            }
+        }
+
+        // Assemble the Files from the buffer and attach to the element, then clean up.
+        let func = r#"function() {
+            const filesData = window.__cuUpload || [];
+            const dt = new DataTransfer();
+            for (const f of filesData) {
+                const bin = atob(f.b64);
+                const arr = new Uint8Array(bin.length);
+                for (let i = 0; i < bin.length; i++) arr[i] = bin.charCodeAt(i);
+                dt.items.add(new File([arr], f.name, { type: f.type }));
+            }
+            try { delete window.__cuUpload; } catch (e) { window.__cuUpload = undefined; }
+            const el = this;
+            if (el.tagName === 'INPUT' && el.type === 'file') {
+                el.files = dt.files;
+                el.dispatchEvent(new Event('input', { bubbles: true }));
+                el.dispatchEvent(new Event('change', { bubbles: true }));
+                return 'input:' + dt.files.length;
+            }
+            // Dropzone / rich composer: replay paste then drop with the files.
+            try { el.dispatchEvent(new ClipboardEvent('paste', { bubbles: true, clipboardData: dt })); } catch (e) {}
+            try {
+                const ev = new DragEvent('drop', { bubbles: true, cancelable: true });
+                Object.defineProperty(ev, 'dataTransfer', { value: dt });
+                el.dispatchEvent(ev);
+            } catch (e) {}
+            return 'event:' + dt.files.length;
+        }"#;
+
+        let result: EvaluateResult = self
+            .client
+            .send_command_typed(
+                "Runtime.callFunctionOn",
+                &CallFunctionOnParams {
+                    function_declaration: func.to_string(),
+                    object_id: Some(object_id),
+                    arguments: None,
+                    return_by_value: Some(true),
+                    await_promise: Some(false),
+                },
+                Some(session_id),
+            )
+            .await
+            .map_err(|e| format!("relay file-injection failed: {}", e))?;
+
+        if let Some(ref details) = result.exception_details {
+            return Err(format!(
+                "relay file-injection threw: {}",
+                details
+                    .exception
+                    .as_ref()
+                    .and_then(|ex| ex.description.as_deref())
+                    .unwrap_or(&details.text)
+            ));
+        }
         Ok(())
     }
 
@@ -2201,6 +2629,7 @@ async fn initialize_lightpanda_manager(
             visited_origins: HashSet::new(),
             created_targets: HashSet::new(),
             active_target_id: None,
+            relay_target_misses: HashMap::new(),
             next_tab_id: 1,
             capture_console: console_capture_enabled(),
         };
@@ -2485,6 +2914,33 @@ mod tests {
         assert_eq!(active_page_index_after_removal(0, 0, 0), 0);
     }
 
+    #[test]
+    fn stale_target_error_matches_relay_signatures() {
+        // The exact relay error `open` must recover from (issue #35), as wrapped
+        // by send_command's `CDP error (Page.navigate): …` prefix.
+        assert!(is_stale_target_error(
+            "CDP error (Page.navigate): stale sessionId cb-tab-1655244623 for Page.navigate: \
+             its tab is gone (closed, navigated across processes, or lost after an extension \
+             restart). Re-attach by re-opening your target URL before retrying."
+        ));
+        assert!(is_stale_target_error(
+            "unknown sessionId cb-tab-7 for Page.navigate"
+        ));
+        assert!(is_stale_target_error("no attached tab for Page.navigate"));
+    }
+
+    #[test]
+    fn stale_target_error_ignores_unrelated_failures() {
+        // A genuine navigation failure (bad URL, DNS, blocked) must NOT trigger
+        // the open-a-fresh-tab recovery — that would mask the real error.
+        assert!(!is_stale_target_error(
+            "Navigation failed: net::ERR_NAME_NOT_RESOLVED"
+        ));
+        assert!(!is_stale_target_error(
+            "CDP command timed out: Page.navigate"
+        ));
+    }
+
     fn page(target_id: &str) -> PageInfo {
         PageInfo {
             tab_id: 1,
@@ -2582,6 +3038,91 @@ mod tests {
         assert!(!active_index_is_owned(&[], None, 0, &created));
     }
 
+    #[test]
+    fn test_sanitize_title() {
+        // The exact pollution from #33: ZWJ / word-joiner / invisible-times / BOM
+        // prepended to "GitHub".
+        let dirty = "\u{200d}\u{2061}\u{200d}\u{2063}\u{200b}\u{2062}\u{feff}GitHub";
+        assert_eq!(sanitize_title(dirty), "GitHub");
+        // Clean titles (incl. CJK + normal punctuation) pass through untouched.
+        assert_eq!(
+            sanitize_title("購入手続きへ - メルカリ"),
+            "購入手続きへ - メルカリ"
+        );
+        assert_eq!(sanitize_title("  Hello World  "), "Hello World");
+        // Emoji and real content survive; only the invisibles are dropped.
+        assert_eq!(sanitize_title("✓ Done\u{200b}"), "✓ Done");
+    }
+
+    #[test]
+    fn test_mime_for_path() {
+        assert_eq!(mime_for_path("a.png"), "image/png");
+        assert_eq!(mime_for_path("PHOTO.JPG"), "image/jpeg");
+        assert_eq!(mime_for_path("clip.webp"), "image/webp");
+        assert_eq!(mime_for_path("doc.pdf"), "application/pdf");
+        assert_eq!(mime_for_path("noext"), "application/octet-stream");
+        assert_eq!(mime_for_path("weird.xyz"), "application/octet-stream");
+    }
+
+    #[test]
+    fn prune_protects_pinned_target_on_transient_snapshot() {
+        // The relay returned a getTargets snapshot missing the pinned tab "A"
+        // (it hopped to another window). "B" is also absent. Without protection
+        // both would be pruned and the next command would drift; with the pin
+        // protected, only the genuinely-unpinned "B" is dropped (issue #31).
+        let pages = vec![page("A"), page("B")];
+        let live: HashSet<String> = HashSet::new(); // snapshot returned neither
+        let gone = prunable_target_ids(&pages, &live, Some("A"));
+        assert_eq!(gone, vec!["B".to_string()]);
+        // With no pin, both are prunable (unchanged behavior).
+        let gone_unpinned = prunable_target_ids(&pages, &live, None);
+        assert_eq!(gone_unpinned.len(), 2);
+        // A pinned target that IS in the live set is simply not prunable anyway.
+        let mut live2 = HashSet::new();
+        live2.insert("A".to_string());
+        assert_eq!(
+            prunable_target_ids(&pages, &live2, Some("A")),
+            vec!["B".to_string()]
+        );
+    }
+
+    #[test]
+    fn debounced_prune_tolerates_transient_churn() {
+        // Multi-agent churn: a single getTargets snapshot omits our owned tab "B"
+        // (another agent opened/closed tabs). It must NOT be pruned on one miss.
+        let pages = vec![page("A"), page("B")];
+        let mut misses = HashMap::new();
+        let empty: HashSet<String> = HashSet::new();
+        // Misses 1 and 2: B absent but under threshold → not pruned.
+        assert!(debounced_prune_ids(&pages, &empty, Some("A"), &mut misses).is_empty());
+        assert!(debounced_prune_ids(&pages, &empty, Some("A"), &mut misses).is_empty());
+        // Miss 3 (== RELAY_PRUNE_MISSES): genuinely gone → pruned.
+        assert_eq!(
+            debounced_prune_ids(&pages, &empty, Some("A"), &mut misses),
+            vec!["B".to_string()]
+        );
+    }
+
+    #[test]
+    fn debounced_prune_resets_on_reappearance_and_protects_pin() {
+        let pages = vec![page("A"), page("B")];
+        let mut misses = HashMap::new();
+        let empty: HashSet<String> = HashSet::new();
+        let mut live_b: HashSet<String> = HashSet::new();
+        live_b.insert("B".to_string());
+        // Two misses for B, then it reappears → counter resets, so it survives
+        // indefinitely under intermittent churn.
+        debounced_prune_ids(&pages, &empty, Some("A"), &mut misses);
+        debounced_prune_ids(&pages, &empty, Some("A"), &mut misses);
+        assert!(debounced_prune_ids(&pages, &live_b, Some("A"), &mut misses).is_empty());
+        assert!(debounced_prune_ids(&pages, &empty, Some("A"), &mut misses).is_empty()); // back to miss 1
+                                                                                         // The pinned active "A" is never pruned no matter how many misses.
+        for _ in 0..5 {
+            let gone = debounced_prune_ids(&pages, &empty, Some("A"), &mut misses);
+            assert!(!gone.contains(&"A".to_string()));
+        }
+    }
+
     #[test]
     fn resolve_active_index_pin_survives_passive_background_tab() {
         // A foreign tab ("Z") gets appended by passive discovery after we pinned

cli/src/native/e2e_tests.rs

@@ -35,6 +35,7 @@ fn native_test_fixture_html(name: &str) -> &'static str {
         "html5_drag_probe" => include_str!("test_fixtures/html5_drag_probe.html"),
         "pointer_capture_probe" => include_str!("test_fixtures/pointer_capture_probe.html"),
         "upload_probe" => include_str!("test_fixtures/upload_probe.html"),
+        "iframe_button_probe" => include_str!("test_fixtures/iframe_button_probe.html"),
         _ => panic!("Unknown native test fixture: {}", name),
     }
 }
@@ -573,6 +574,76 @@ async fn e2e_snapshot_and_click_ref() {
     assert_success(&resp);
 }
 
+/// Clicking a button INSIDE an iframe by `@ref` must deliver a TRUSTED activation
+/// (`event.isTrusted === true`), not a synthetic DOM `.click()`. Security-sensitive
+/// embedded forms (Google Payments' `保存`) reject `isTrusted:false` clicks, so an
+/// enabled submit button silently no-op'd (issue #39). The fix routes iframe-ref
+/// clicks to a real `Input.dispatchMouseEvent` on the element's own frame session.
+/// The fixture's iframe button writes `clicked:<isTrusted>` into its own text on
+/// click, which the cross-frame snapshot reads back.
+#[tokio::test]
+#[ignore]
+async fn e2e_iframe_button_click_is_trusted() {
+    let mut state = DaemonState::new();
+
+    let resp = execute_command(
+        &json!({ "id": "1", "action": "launch", "headless": true }),
+        &mut state,
+    )
+    .await;
+    assert_success(&resp);
+
+    let resp = execute_command(
+        &json!({ "id": "2", "action": "navigate", "url": native_test_fixture_url("iframe_button_probe") }),
+        &mut state,
+    )
+    .await;
+    assert_success(&resp);
+
+    // Snapshot (interactive) — the button lives in the iframe and must appear with
+    // a ref; that ref carries the frame_id so the click resolves into the frame.
+    let resp = execute_command(
+        &json!({ "id": "3", "action": "snapshot", "interactive": true }),
+        &mut state,
+    )
+    .await;
+    assert_success(&resp);
+    let snapshot = get_data(&resp)["snapshot"].as_str().unwrap_or("");
+    let ref_id = snapshot
+        .lines()
+        .find(|l| l.contains("button \"save\""))
+        .and_then(|l| l.split("ref=").nth(1))
+        .map(|r| r.trim_end_matches(']').trim())
+        .unwrap_or_else(|| panic!("iframe button not found in snapshot:\n{snapshot}"));
+
+    // Click it by ref.
+    let resp = execute_command(
+        &json!({ "id": "4", "action": "click", "selector": ref_id }),
+        &mut state,
+    )
+    .await;
+    assert_success(&resp);
+
+    tokio::time::sleep(tokio::time::Duration::from_millis(300)).await;
+
+    // The button rewrote its own text with the click's isTrusted flag; read it
+    // back across frames.
+    let resp = execute_command(
+        &json!({ "id": "5", "action": "snapshot", "interactive": true }),
+        &mut state,
+    )
+    .await;
+    assert_success(&resp);
+    let after = get_data(&resp)["snapshot"].as_str().unwrap_or("");
+    assert!(
+        after.contains("clicked:true"),
+        "iframe button click must be trusted (isTrusted:true); snapshot:\n{after}"
+    );
+
+    let resp = execute_command(&json!({ "id": "99", "action": "close" }), &mut state).await;
+    assert_success(&resp);
+}
+
 // ---------------------------------------------------------------------------
 // Screenshot
 // ---------------------------------------------------------------------------

cli/src/native/element.rs

@@ -100,6 +100,15 @@ impl RefMap {
         self.map.get(ref_id)
     }
 
+    /// Whether `selector_or_ref` is a `@ref` whose snapshot entry lives inside an
+    /// iframe (has a `frame_id`). Pointer interactions use this to choose
+    /// DOM-dispatch over coordinates for OOPIF elements (issue #36).
+    pub fn ref_is_in_iframe(&self, selector_or_ref: &str) -> bool {
+        parse_ref(selector_or_ref)
+            .and_then(|r| self.map.get(&r).map(|e| e.frame_id.is_some()))
+            .unwrap_or(false)
+    }
+
     pub fn entries_sorted(&self) -> Vec<(String, RefEntry)> {
         let mut entries = self
             .map
@@ -1019,7 +1028,9 @@ async fn eval_text_in_frame(client: &CdpClient, session_id: &str, frame_id: &str
         .await
         .ok()
         .and_then(|v| v.get("executionContextId").and_then(|c| c.as_i64()));
-    let Some(ctx_id) = ctx else { return String::new() };
+    let Some(ctx_id) = ctx else {
+        return String::new();
+    };
     let res = client
         .send_command(
             "Runtime.evaluate",
@@ -1090,7 +1101,10 @@ pub async fn collect_all_frames_text(
         let (kind, text) = if is_top {
             ("top", eval_text_default(client, top_session).await)
         } else {
-            ("inline", eval_text_in_frame(client, top_session, &fid).await)
+            (
+                "inline",
+                eval_text_in_frame(client, top_session, &fid).await,
+            )
         };
         out.push(FrameText {
             frame_id: fid,

cli/src/native/interaction.rs

@@ -7,6 +7,17 @@ use super::cdp::types::*;
 use super::element::{parse_ref, resolve_element_center, resolve_element_object_id, RefMap};
 use super::humanize;
 
+/// Whether a pointer interaction should be DOM-dispatched (invoke the event on
+/// the element in its own session) rather than dispatched at a viewport
+/// coordinate via `Input.dispatchMouseEvent`. True when the target is inside an
+/// iframe (an OOPIF element's box can't be mapped to a top-viewport point) or we
+/// drive over the extension relay (a coordinate Input event isn't confined to the
+/// target tab on a busy real Chrome — it drifts onto the foreground tab; issues
+/// #31/#36). DOM-dispatch always hits the right element in the right tab.
+fn prefer_dom_dispatch(ref_map: &RefMap, selector_or_ref: &str) -> bool {
+    ref_map.ref_is_in_iframe(selector_or_ref) || crate::connect::relay_url().is_some()
+}
+
 pub async fn click(
     client: &CdpClient,
     session_id: &str,
@@ -45,6 +56,45 @@ pub async fn click(
         .await;
     }
 
+    // An element INSIDE an iframe needs a TRUSTED activation: a DOM `.click()` is
+    // `isTrusted:false`, which security-sensitive embedded forms reject — Google
+    // Payments' enabled `保存` button silently no-ops on a synthetic click (issue
+    // #39). A coordinate `Input.dispatchMouseEvent` can't help either: `getBoxModel`
+    // for a sub-frame node returns frame-local coordinates that don't compose the
+    // iframe's offset, so the click lands in the wrong place. The frame-agnostic
+    // trusted path is keyboard activation — focus the element in its own frame, then
+    // dispatch a real Enter on the page session; Chrome routes the key to the
+    // focused element regardless of frame (same as `type --focused`), and Enter on a
+    // focused button/link fires a trusted `click`. `coord` mode opts out.
+    let in_iframe = ref_map.ref_is_in_iframe(selector_or_ref);
+    if mode != "coord" && button == "left" && click_count == 1 && in_iframe {
+        return dom_activate(
+            client,
+            session_id,
+            ref_map,
+            selector_or_ref,
+            iframe_sessions,
+        )
+        .await;
+    }
+    // On the relay (the user's real Chrome) a TOP-document coordinate click used to
+    // drift onto the foreground tab; that root cause is fixed (#5: the agent drives
+    // its own pinned tab), but DOM-dispatch stays the conservative default here.
+    if mode != "coord"
+        && button == "left"
+        && click_count == 1
+        && crate::connect::relay_url().is_some()
+    {
+        return dom_click(
+            client,
+            session_id,
+            ref_map,
+            selector_or_ref,
+            iframe_sessions,
+        )
+        .await;
+    }
+
     let resolved = resolve_element_center(
         client,
         session_id,
@@ -235,6 +285,112 @@ async fn dom_click(
     Ok(())
 }
 
+/// Trusted activation of an element inside an iframe (issue #39). Focuses the
+/// element in its own frame session, then dispatches a real Enter/Space on the
+/// page session — Chrome routes the key to the focused element across frames, and
+/// Enter/Space on a focused button/link/checkbox fires a `click` with
+/// `isTrusted: true`, which security-sensitive embedded forms (Google Payments
+/// `保存`) require. Non-activatable roles (a `div[onclick]`) can't be keyboard-
+/// activated, so they fall back to a DOM `.click()`.
+async fn dom_activate(
+    client: &CdpClient,
+    session_id: &str,
+    ref_map: &RefMap,
+    selector_or_ref: &str,
+    iframe_sessions: &HashMap<String, String>,
+) -> Result<(), String> {
+    let role = parse_ref(selector_or_ref)
+        .and_then(|r| ref_map.get(&r).map(|e| e.role.clone()))
+        .unwrap_or_default();
+    // Space toggles checkbox-like controls; Enter activates buttons/links/menus.
+    let key = match role.as_str() {
+        "checkbox" | "radio" | "switch" | "option" | "menuitemcheckbox" | "menuitemradio" => {
+            Some("space")
+        }
+        "button" | "link" | "menuitem" | "tab" | "treeitem" => Some("enter"),
+        _ => None,
+    };
+    let Some(key) = key else {
+        // Not keyboard-activatable — best effort via DOM .click() (untrusted).
+        return dom_click(
+            client,
+            session_id,
+            ref_map,
+            selector_or_ref,
+            iframe_sessions,
+        )
+        .await;
+    };
+
+    let (object_id, effective_session_id) = resolve_element_object_id(
+        client,
+        session_id,
+        ref_map,
+        selector_or_ref,
+        iframe_sessions,
+    )
+    .await?;
+    // Focus the element in its OWN frame session so the keystroke lands on it.
+    client
+        .send_command_typed::<_, Value>(
+            "Runtime.callFunctionOn",
+            &CallFunctionOnParams {
+                function_declaration: "function() { this.focus(); }".to_string(),
+                object_id: Some(object_id),
+                arguments: None,
+                return_by_value: Some(true),
+                await_promise: Some(false),
+            },
+            Some(&effective_session_id),
+        )
+        .await?;
+    // Trusted key on the page session — routed to the focused (in-frame) element.
+    press_key(client, session_id, key).await?;
+    wait_for_paint_settled(client, &effective_session_id).await;
+    Ok(())
+}
+
+/// DOM-dispatch a double-click on the element in its own session (no coordinates)
+/// — the relay/iframe-safe counterpart to a coordinate dblclick. Fires the full
+/// click,click,dblclick sequence so handlers bound to any of them respond.
+async fn dom_dblclick(
+    client: &CdpClient,
+    session_id: &str,
+    ref_map: &RefMap,
+    selector_or_ref: &str,
+    iframe_sessions: &HashMap<String, String>,
+) -> Result<(), String> {
+    let (object_id, effective_session_id) = resolve_element_object_id(
+        client,
+        session_id,
+        ref_map,
+        selector_or_ref,
+        iframe_sessions,
+    )
+    .await?;
+    client
+        .send_command_typed::<_, Value>(
+            "Runtime.callFunctionOn",
+            &CallFunctionOnParams {
+                function_declaration: r#"function() {
+                    const opts = { bubbles: true, cancelable: true, view: window };
+                    this.dispatchEvent(new MouseEvent('click', opts));
+                    this.dispatchEvent(new MouseEvent('click', { ...opts, detail: 2 }));
+                    this.dispatchEvent(new MouseEvent('dblclick', opts));
+                }"#
+                .to_string(),
+                object_id: Some(object_id),
+                arguments: None,
+                return_by_value: Some(true),
+                await_promise: Some(false),
+            },
+            Some(&effective_session_id),
+        )
+        .await?;
+    wait_for_paint_settled(client, &effective_session_id).await;
+    Ok(())
+}
+
 pub async fn dblclick(
     client: &CdpClient,
     session_id: &str,
@@ -242,6 +398,20 @@ pub async fn dblclick(
     selector_or_ref: &str,
     iframe_sessions: &HashMap<String, String>,
 ) -> Result<(), String> {
+    // Same relay/iframe drift hazard as a single click — DOM-dispatch the
+    // double-click there instead of a coordinate one (issues #31/#36).
+    if std::env::var("AGENT_BROWSER_CLICK_MODE").as_deref() != Ok("coord")
+        && prefer_dom_dispatch(ref_map, selector_or_ref)
+    {
+        return dom_dblclick(
+            client,
+            session_id,
+            ref_map,
+            selector_or_ref,
+            iframe_sessions,
+        )
+        .await;
+    }
     click(
         client,
         session_id,
@@ -254,6 +424,50 @@ pub async fn dblclick(
     .await
 }
 
+/// DOM-dispatch a hover (pointer/mouse enter+move) on the element in its own
+/// session — reaches OOPIF elements and never drifts to the foreground tab over
+/// the relay, unlike a coordinate `mouseMoved` (issues #31/#36).
+async fn dom_hover(
+    client: &CdpClient,
+    session_id: &str,
+    ref_map: &RefMap,
+    selector_or_ref: &str,
+    iframe_sessions: &HashMap<String, String>,
+) -> Result<(), String> {
+    let (object_id, effective_session_id) = resolve_element_object_id(
+        client,
+        session_id,
+        ref_map,
+        selector_or_ref,
+        iframe_sessions,
+    )
+    .await?;
+    client
+        .send_command_typed::<_, Value>(
+            "Runtime.callFunctionOn",
+            &CallFunctionOnParams {
+                function_declaration: r#"function() {
+                    const r = this.getBoundingClientRect();
+                    const cx = r.left + r.width / 2, cy = r.top + r.height / 2;
+                    const base = { bubbles: true, cancelable: true, view: window, clientX: cx, clientY: cy };
+                    this.dispatchEvent(new PointerEvent('pointerover', base));
+                    this.dispatchEvent(new PointerEvent('pointerenter', { ...base, bubbles: false }));
+                    this.dispatchEvent(new MouseEvent('mouseover', base));
+                    this.dispatchEvent(new MouseEvent('mouseenter', { ...base, bubbles: false }));
+                    this.dispatchEvent(new MouseEvent('mousemove', base));
+                }"#
+                .to_string(),
+                object_id: Some(object_id),
+                arguments: None,
+                return_by_value: Some(true),
+                await_promise: Some(false),
+            },
+            Some(&effective_session_id),
+        )
+        .await?;
+    Ok(())
+}
+
 pub async fn hover(
     client: &CdpClient,
     session_id: &str,
@@ -261,6 +475,18 @@ pub async fn hover(
     selector_or_ref: &str,
     iframe_sessions: &HashMap<String, String>,
 ) -> Result<(), String> {
+    // Coordinate `mouseMoved` drifts to the foreground tab over the relay and
+    // can't reach an OOPIF — DOM-dispatch the hover there (issues #31/#36).
+    if prefer_dom_dispatch(ref_map, selector_or_ref) {
+        return dom_hover(
+            client,
+            session_id,
+            ref_map,
+            selector_or_ref,
+            iframe_sessions,
+        )
+        .await;
+    }
     let (x, y, _w, _h, effective_session_id) = resolve_element_center(
         client,
         session_id,
@@ -289,6 +515,63 @@ pub async fn hover(
     Ok(())
 }
 
+/// DOM-dispatch an HTML5 drag-and-drop from `source` to `target` in their shared
+/// session — the relay/iframe-safe counterpart to the coordinate drag, which
+/// drifts to the foreground tab over the relay and can't reach an OOPIF (issues
+/// #31/#36). Covers HTML5 DnD (sortable lists, file/card boards); pointer-driven
+/// drag (canvas, sliders) still needs the coordinate path. Errors if source and
+/// target live in different frames — a synthetic cross-frame DnD isn't reliable.
+pub async fn dom_drag(
+    client: &CdpClient,
+    session_id: &str,
+    ref_map: &RefMap,
+    source: &str,
+    target: &str,
+    iframe_sessions: &HashMap<String, String>,
+) -> Result<(), String> {
+    let (src_obj, src_session) =
+        resolve_element_object_id(client, session_id, ref_map, source, iframe_sessions).await?;
+    let (tgt_obj, tgt_session) =
+        resolve_element_object_id(client, session_id, ref_map, target, iframe_sessions).await?;
+    if src_session != tgt_session {
+        return Err(
+            "drag source and target are in different frames; cross-frame drag-and-drop over the \
+             relay isn't supported — drag within a single frame, or use a launched browser with \
+             AGENT_BROWSER_CLICK_MODE=coord"
+                .to_string(),
+        );
+    }
+    client
+        .send_command_typed::<_, Value>(
+            "Runtime.callFunctionOn",
+            &CallFunctionOnParams {
+                function_declaration: r#"function(target) {
+                    const dt = new DataTransfer();
+                    const ev = (type, el) => el.dispatchEvent(
+                        new DragEvent(type, { bubbles: true, cancelable: true, dataTransfer: dt }));
+                    ev('dragstart', this);
+                    ev('drag', this);
+                    ev('dragenter', target);
+                    ev('dragover', target);
+                    ev('drop', target);
+                    ev('dragend', this);
+                }"#
+                .to_string(),
+                object_id: Some(src_obj),
+                arguments: Some(vec![CallArgument {
+                    value: None,
+                    object_id: Some(tgt_obj),
+                }]),
+                return_by_value: Some(true),
+                await_promise: Some(false),
+            },
+            Some(&src_session),
+        )
+        .await?;
+    wait_for_paint_settled(client, &src_session).await;
+    Ok(())
+}
+
 pub async fn fill(
     client: &CdpClient,
     session_id: &str,
@@ -372,6 +655,7 @@ pub async fn type_text(
     clear: bool,
     delay_ms: Option<u64>,
     iframe_sessions: &HashMap<String, String>,
+    key_events: bool,
 ) -> Result<(), String> {
     let (object_id, effective_session_id) = resolve_element_object_id(
         client,
@@ -418,7 +702,7 @@ pub async fn type_text(
             .await?;
     }
 
-    type_text_into_active_context(client, session_id, text, delay_ms).await
+    type_text_into_active_context(client, session_id, text, delay_ms, key_events).await
 }
 
 pub async fn type_text_into_active_context(
@@ -426,6 +710,7 @@ pub async fn type_text_into_active_context(
     session_id: &str,
     text: &str,
     delay_ms: Option<u64>,
+    key_events: bool,
 ) -> Result<(), String> {
     // Per-character timing: an explicit `delay_ms` wins (caller asked for a
     // fixed cadence); otherwise fall back to humanize — variable, human-like
@@ -475,6 +760,46 @@ pub async fn type_text_into_active_context(
                     Some(session_id),
                 )
                 .await?;
+        } else if key_events {
+            // Real keystrokes (keyDown+keyUp carrying `text`) for autocomplete /
+            // combobox widgets that only react to key events and ignore the
+            // `input` that `Input.insertText` fires — e.g. Google's address
+            // postal-code → city/prefecture lookup (issue #36 / #4). The keyDown's
+            // `text` still inserts the character, so the field also fills.
+            let (key, code, key_code) = char_to_key_info(ch);
+            let s = ch.to_string();
+            client
+                .send_command_typed::<_, Value>(
+                    "Input.dispatchKeyEvent",
+                    &DispatchKeyEventParams {
+                        event_type: "keyDown".to_string(),
+                        key: Some(key.clone()),
+                        code: Some(code.clone()),
+                        text: Some(s.clone()),
+                        unmodified_text: Some(s),
+                        windows_virtual_key_code: Some(key_code),
+                        native_virtual_key_code: Some(key_code),
+                        modifiers: None,
+                    },
+                    Some(session_id),
+                )
+                .await?;
+            client
+                .send_command_typed::<_, Value>(
+                    "Input.dispatchKeyEvent",
+                    &DispatchKeyEventParams {
+                        event_type: "keyUp".to_string(),
+                        key: Some(key),
+                        code: Some(code),
+                        text: None,
+                        unmodified_text: None,
+                        windows_virtual_key_code: Some(key_code),
+                        native_virtual_key_code: Some(key_code),
+                        modifiers: None,
+                    },
+                    Some(session_id),
+                )
+                .await?;
         } else {
             // VS Code/Electron webviews reject repeated dispatchKeyEvent calls
             // carrying printable `text`. Insert printable characters directly

cli/src/native/screenshot.rs

@@ -60,6 +60,9 @@ pub struct ScreenshotOptions {
     pub quality: Option<i32>,
     pub annotate: bool,
     pub output_dir: Option<String>,
+    /// Explicit pixel region (x, y, width, height) — `--clip` (issue #34). Takes
+    /// precedence over selector/full_page.
+    pub clip: Option<(f64, f64, f64, f64)>,
 }
 
 impl Default for ScreenshotOptions {
@@ -72,6 +75,7 @@ impl Default for ScreenshotOptions {
             quality: None,
             annotate: false,
             output_dir: None,
+            clip: None,
         }
     }
 }
@@ -187,7 +191,16 @@ async fn capture_screenshot_base64(
         capture_beyond_viewport: if options.full_page { Some(true) } else { None },
     };
 
-    if options.full_page {
+    if let Some((x, y, width, height)) = options.clip {
+        // Explicit pixel region wins over selector/full_page (issue #34).
+        params.clip = Some(Viewport {
+            x,
+            y,
+            width,
+            height,
+            scale: 1.0,
+        });
+    } else if options.full_page {
         let metrics: Value = client
             .send_command_no_params("Page.getLayoutMetrics", Some(session_id))
             .await?;

cli/src/native/snapshot.rs

@@ -330,6 +330,13 @@ impl RoleNameTracker {
     }
 }
 
+/// Max iframe nesting depth `take_snapshot` expands. Embedded payment/checkout
+/// widgets nest a few frames deep (e.g. AdSense → payments.google.com → an inner
+/// form frame); expanding past the first level is what gives those inner refs a
+/// `frame_id` so clicks resolve into the right frame (issue #36). Capped to keep
+/// a pathological frame tree from blowing up the snapshot.
+const MAX_IFRAME_DEPTH: usize = 3;
+
 pub async fn take_snapshot(
     client: &CdpClient,
     session_id: &str,
@@ -337,6 +344,28 @@ pub async fn take_snapshot(
     ref_map: &mut RefMap,
     frame_id: Option<&str>,
     iframe_sessions: &HashMap<String, String>,
+) -> Result<String, String> {
+    take_snapshot_at_depth(
+        client,
+        session_id,
+        options,
+        ref_map,
+        frame_id,
+        iframe_sessions,
+        0,
+    )
+    .await
+}
+
+#[allow(clippy::too_many_arguments)]
+async fn take_snapshot_at_depth(
+    client: &CdpClient,
+    session_id: &str,
+    options: &SnapshotOptions,
+    ref_map: &mut RefMap,
+    frame_id: Option<&str>,
+    iframe_sessions: &HashMap<String, String>,
+    depth: usize,
 ) -> Result<String, String> {
     client
         .send_command_no_params("DOM.enable", Some(session_id))
@@ -606,10 +635,11 @@ pub async fn take_snapshot(
     }
 
     // Recurse into child iframes: for each Iframe node with a backend_node_id,
-    // resolve the child frame ID and take a snapshot of its content.
-    // We only recurse from the main frame (frame_id == None) to avoid
-    // unbounded depth; nested iframes within iframes are not expanded.
-    if frame_id.is_none() {
+    // resolve the child frame ID and snapshot its content. Recurse to
+    // MAX_IFRAME_DEPTH (not just the main frame) so refs inside nested
+    // payment/checkout widgets get a `frame_id` and clicks resolve into the right
+    // frame (issue #36); the cap bounds a pathological frame tree.
+    if depth < MAX_IFRAME_DEPTH {
         let mut iframe_snapshots: Vec<(String, String)> = Vec::new(); // (ref_id, child_snapshot)
         for node in tree_nodes.iter() {
             if node.role != "Iframe" || !node.has_ref {
@@ -622,13 +652,14 @@ pub async fn take_snapshot(
             if let Ok(child_fid) = resolve_iframe_frame_id(client, session_id, bid).await {
                 // Snapshot the child frame; errors are silently ignored
                 // (e.g. cross-origin iframes)
-                if let Ok(child_text) = Box::pin(take_snapshot(
+                if let Ok(child_text) = Box::pin(take_snapshot_at_depth(
                     client,
                     session_id,
                     options,
                     ref_map,
                     Some(&child_fid),
                     iframe_sessions,
+                    depth + 1,
                 ))
                 .await
                 {

cli/src/native/test_fixtures/iframe_button_probe.html

@@ -0,0 +1,28 @@
+<!doctype html>
+<html>
+  <head>
+    <meta charset="utf-8" />
+    <title>iframe button probe</title>
+  </head>
+  <body>
+    <h1>iframe button probe</h1>
+    <iframe
+      id="frame"
+      width="320"
+      height="140"
+      srcdoc="
+        <!doctype html>
+        <html>
+          <body style='margin:24px'>
+            <button id='b' style='padding:24px;font-size:22px'>save</button>
+            <script>
+              document.getElementById('b').addEventListener('click', function (e) {
+                this.textContent = 'clicked:' + e.isTrusted;
+              });
+            </script>
+          </body>
+        </html>
+      "
+    ></iframe>
+  </body>
+</html>

cli/src/output.rs

@@ -224,6 +224,65 @@ pub fn print_response_with_opts(resp: &Response, action: Option<&str>, opts: &Ou
             return;
         }
 
+        // Cloudflare challenge/clearance preflight (`cf-status`). Checked early
+        // because its response carries `url`/`title`, which later generic
+        // renderers would otherwise swallow.
+        if action == Some("cf_status") {
+            let challenged = data
+                .get("challenged")
+                .and_then(|v| v.as_bool())
+                .unwrap_or(false);
+            let rec = data
+                .get("recommendation")
+                .and_then(|v| v.as_str())
+                .unwrap_or("?");
+            let cl = data.get("clearance");
+            let present = cl
+                .and_then(|c| c.get("present"))
+                .and_then(|v| v.as_bool())
+                .unwrap_or(false);
+            let expired = cl
+                .and_then(|c| c.get("expired"))
+                .and_then(|v| v.as_bool())
+                .unwrap_or(false);
+            let expires_in = cl.and_then(|c| c.get("expiresIn")).and_then(|v| v.as_i64());
+            let device = data
+                .get("deviceVerified")
+                .and_then(|v| v.as_bool())
+                .unwrap_or(false);
+
+            let (icon, headline) = match rec {
+                "proceed" => (color::success_indicator().to_string(), "cleared — no challenge, proceed"),
+                "solve" => (color::warning_indicator().to_string(), "Cloudflare challenge active, no valid clearance — solve it"),
+                "reissue" => (color::warning_indicator().to_string(), "challenge active but a clearance cookie exists — stale (IP/UA changed?), re-solve"),
+                _ => (color::cyan("•").to_string(), "unknown"),
+            };
+            println!("{} {}", icon, headline);
+            println!(
+                "  challenged:     {}",
+                if challenged { "yes" } else { "no" }
+            );
+            let cl_desc = if !present {
+                "absent".to_string()
+            } else if expired {
+                "present but EXPIRED".to_string()
+            } else if let Some(s) = expires_in {
+                format!("valid, expires in {}m {}s", s / 60, s % 60)
+            } else {
+                "present (session)".to_string()
+            };
+            println!("  cf_clearance:   {}", cl_desc);
+            println!(
+                "  device trusted: {}",
+                if device {
+                    "yes (CF_VERIFIED_DEVICE)"
+                } else {
+                    "no"
+                }
+            );
+            return;
+        }
+
         // Dialog status response
         if action == Some("dialog") {
             if let Some(has_dialog) = data.get("hasDialog").and_then(|v| v.as_bool()) {
@@ -348,7 +407,11 @@ pub fn print_response_with_opts(resp: &Response, action: Option<&str>, opts: &Ou
                 let count = list.len();
                 println!(
                     "{}",
-                    color::bold(&format!("{} frame{}", count, if count == 1 { "" } else { "s" }))
+                    color::bold(&format!(
+                        "{} frame{}",
+                        count,
+                        if count == 1 { "" } else { "s" }
+                    ))
                 );
                 for f in list {
                     let idx = f.get("index").and_then(|v| v.as_i64()).unwrap_or(0);
@@ -1430,6 +1493,12 @@ Usage: chrome-use type <selector> <text>
 Types text into the specified element character by character.
 Unlike fill, this does not clear existing content first.
 
+Options:
+  --key-events         Send real per-character keyDown/keyUp instead of
+  (alias --keys)       Input.insertText. Use for autocomplete / combobox fields
+                       that only react to key events — e.g. a postal-code box
+                       that auto-fills city/prefecture, or Google Places.
+
 Global Options:
   --json               Output as JSON
   --session <name>     Use specific session
@@ -1437,6 +1506,7 @@ Global Options:
 Examples:
   chrome-use type "#search" "hello"
   chrome-use type @e2 "additional text"
+  chrome-use type @e5 "201-0001" --key-events   # trigger the address autocomplete
 
 See Also:
   For typing into contenteditable editors (Lexical, ProseMirror, etc.)
@@ -1701,12 +1771,23 @@ Usage: chrome-use scroll [direction] [amount] [options]
 
 Scrolls the page or a specific element in the specified direction.
 
+Without --selector, scroll dispatches a real (isTrusted) mouse wheel at a
+viewport coordinate, so it scrolls whatever container is under the pointer —
+including cross-origin iframes (Google Payments, Stripe, embedded checkout/KYC)
+that plain page scroll can't reach.
+
 Arguments:
   direction            up, down, left, right (default: down)
   amount               Pixels to scroll (default: 300)
 
 Options:
-  -s, --selector <sel> CSS selector for a scrollable container
+  -s, --selector <sel> CSS selector for a scrollable container (same-origin)
+  --at <x,y>           Dispatch the wheel at this viewport pixel (read it from a
+                       screenshot) — precise way into a cross-origin iframe
+  --frame <n>          Scroll the n-th frame from `chrome-use frames` (wheel at
+                       that frame's center)
+
+Without --selector/--at/--frame the wheel lands at the viewport center.
 
 Global Options:
   --json               Output as JSON
@@ -1718,6 +1799,8 @@ Examples:
   chrome-use scroll up 200
   chrome-use scroll left 100
   chrome-use scroll down 500 --selector "div.scroll-container"
+  chrome-use scroll down 700 --at 640,400      # wheel at a pixel over an iframe
+  chrome-use scroll down 700 --frame 2         # scroll frame 2 from `frames`
 "##
         }
         "scrollintoview" | "scrollinto" => {
@@ -1798,6 +1881,8 @@ Pass --hide-scrollbars false when launching to keep native scrollbars visible.
 
 Options:
   --full, -f           Capture full page (not just viewport)
+  [selector]           Capture just an element (CSS or @ref), e.g. `screenshot ".header" h.png`
+  --clip <x,y,w,h>     Capture a pixel region, e.g. `screenshot --clip 0,0,200,40 corner.png`
   --annotate           Overlay numbered labels on interactive elements.
                        Each label [N] corresponds to ref @eN from snapshot.
                        Prints a legend mapping labels to element roles/names.
@@ -1818,6 +1903,8 @@ Examples:
   chrome-use screenshot
   chrome-use screenshot ./screenshot.png
   chrome-use screenshot --full ./full-page.png
+  chrome-use screenshot ".header .indicator" corner.png  # just one element
+  chrome-use screenshot --clip 1600,0,200,40 corner.png  # a pixel region
   chrome-use screenshot --annotate              # Labeled screenshot + legend
   chrome-use screenshot --annotate ./page.png   # Save annotated screenshot
   chrome-use screenshot --annotate --json       # JSON output with annotations
@@ -3195,6 +3282,10 @@ Get Info:  chrome-use get <what> [selector]
 Check State:  chrome-use is <what> <selector>
   visible, enabled, checked
 
+Anti-bot:  chrome-use stealth | cf-status
+  stealth         stealth self-check (webdriver/UA/plugins + overrides)
+  cf-status       Cloudflare challenge + cf_clearance preflight (skip re-solving)
+
 Find Elements:  chrome-use find <locator> <value> <action> [text]
   role, text, label, placeholder, alt, title, testid, first, last, nth
 
@@ -3486,6 +3577,9 @@ iOS Simulator (requires Xcode and Appium):
   chrome-use -p ios device list                         # List simulators
   chrome-use -p ios swipe up                            # Swipe gesture
   chrome-use -p ios tap @e1                             # Touch element
+
+Hit a bug or rough edge? A 30-second issue genuinely sharpens this tool:
+  https://github.com/leeguooooo/chrome-use/issues
 "#
     );
 }
@@ -3568,6 +3662,7 @@ fn print_screenshot_diff(data: &serde_json::Map<String, serde_json::Value>) {
 
 pub fn print_version() {
     println!("chrome-use {}", env!("CARGO_PKG_VERSION"));
+    println!("report bugs / rough edges: https://github.com/leeguooooo/chrome-use/issues");
 }
 
 #[cfg(test)]

package.json

@@ -1,6 +1,6 @@
 {
   "name": "chrome-use",
-  "version": "1.5.4",
+  "version": "1.5.18",
   "description": "chrome-use — drive your real, logged-in Chrome from any AI agent, stealth by default",
   "type": "module",
   "packageManager": "pnpm@11.1.3",

skill-data/core/SKILL.md

@@ -36,6 +36,29 @@ Refs (`@e1`, `@e2`, ...) are assigned fresh on every snapshot. They become
 submits, dynamic re-renders, dialog opens. Always re-snapshot before your
 next ref interaction.
 
+> **Hard rule: snapshot-first, never screenshot-to-locate.** For form fields and
+> buttons, ALWAYS `snapshot -i` and act on refs/selectors. Do **not** reach for
+> `screenshot` + coordinate clicks to find or hit an element — `snapshot -i` now
+> pierces **cross-origin iframes** (embedded Google Payments / Stripe / checkout /
+> KYC forms) and lists their elements by `@ref`, including input values. Use
+> coordinates only for canvas/WebGL, or when `snapshot` genuinely returns nothing
+> for your target. Screenshots are for *visual verification you report*, never the
+> agent's own input — and a full-page `screenshot` of a real retina browser is
+> often too large for an image reader anyway. (If you ever feel you *need* a
+> screenshot to read state or locate something, that's a bug — please file it.)
+
+> **Snapshot-first, always. Never default to `screenshot` + coordinate clicking
+> for form fields or buttons.** Run `snapshot -i` and act on `@refs`. Use
+> coordinates only for canvas/WebGL, or when `snapshot` genuinely returns nothing
+> for your target. This holds **even inside cross-origin embedded iframes** —
+> since v1.5.12 `snapshot -i` pierces out-of-process iframes (Google Payments,
+> Stripe, embedded checkout/KYC) and lists their elements with refs, so
+> `click @e` / `type @e` / `fill @e` work directly. A screenshot is for a genuine
+> *visual* check you report to the user — not your own input. (Full-page
+> screenshots of a real retina Chrome are often too large for the image reader
+> anyway.) Driving off pixels on the relay also risks a coordinate event drifting
+> onto the user's foreground tab — refs never do. See issue #37.
+
 ## Before you automate: pick the cheapest tool
 
 Driving a browser is the heavy option. chrome-use earns its keep when you
@@ -108,7 +131,17 @@ Each `--session` that connects gets its **own colored Chrome tab group** (named
 after the session) and drives only its own tabs — multiple agents share the one
 real browser without cross-talk, and the user's own tabs are never grouped. CDP
 drives the page without moving the user's mouse/keyboard, so it doesn't fight
-them for control. **Anti-detection ranking: this real logged-in Chrome (extension
+them for control.
+
+**Strict multi-agent isolation.** A session over the relay tracks and drives
+**only the tabs it created** (its own group). It does **not** adopt the user's
+existing tabs, other agents' tabs, or pop-ups (e.g. an OAuth/login window — that's
+the user's), so several agents (and other tools opening tabs) can work in the same
+real Chrome concurrently without ever dropping or stealing each other's tabs —
+another agent's tab churn can't make your bound tab vanish or drift your commands
+onto the wrong page. Consequence: `tab list` shows only *your* session's tabs; to
+drive a specific page, navigate to it in your own tab instead of expecting a
+pre-existing or popped-up tab to appear in the list. **Anti-detection ranking: this real logged-in Chrome (extension
 connect) > a headed launched browser > headless (forbidden).** A genuine human
 browser has no headless/automation tells at all, so prefer it for anything
 anti-bot-sensitive.
@@ -127,6 +160,19 @@ cadence, and scroll/drag ease. Default `off`; a per-navigation detector
 auto-escalates pages guarded by Akamai/PerimeterX/DataDome to `human`. Leave it
 on auto; force `human` only when you already know the target scores behaviour.
 
+**Cloudflare clearance — solve once, reuse.** Passing a Cloudflare challenge
+mints a `cf_clearance` cookie (HttpOnly — invisible to `eval`/`document.cookie`;
+read it via `chrome-use cookies`). It's bound to your **IP + User-Agent**: reuse
+the same exit IP and UA and you skip the challenge until it expires. Driving the
+user's real Chrome (relay) persists it natively; for isolated sessions,
+`--session-name <name>` save/restores it. Before spending effort solving, run
+`chrome-use cf-status` (aliases `cf`, `clearance`): it reports whether the page
+is *currently* a Cloudflare challenge and whether a still-valid `cf_clearance`
+exists, with a recommendation — `proceed` (already cleared, don't re-solve),
+`solve` (challenge up, no clearance), or `reissue` (clearance present but page
+still blocks → IP/UA drifted, re-solve). Use it as a preflight to avoid
+re-solving what you already cleared.
+
 ## Two ways to drive a page — and when to drop to `eval`
 
 You have a **real Chrome with the user's DOM**. Two layers, mix them freely:
@@ -251,6 +297,10 @@ chrome-use hover @e1                   # hover
 chrome-use focus @e1                   # focus (useful before keyboard input)
 chrome-use fill @e2 "hello"            # clear then type
 chrome-use type @e2 " world"           # type without clearing
+chrome-use type @e5 "201-0001" --key-events  # real keystrokes (not insertText) —
+                                          # use for autocomplete/combobox fields that
+                                          # only react to key events (e.g. a postal box
+                                          # that auto-fills city/prefecture, Google Places)
 chrome-use press Enter                 # press a key at current focus (down+up)
 chrome-use press Control+a             # key combination
 chrome-use keydown d                   # HOLD a key down (no auto-release)
@@ -268,16 +318,32 @@ chrome-use pick @e4 --option "Europe"  # ANY combobox (react-select / ARIA /
                                           # (no silent no-op). Use this for custom
                                           # dropdowns where `select` returns ✓ but
                                           # changes nothing.
-chrome-use upload @e5 file1.pdf        # upload file(s) — NOTE: needs a --launch/direct-CDP
-                                          # session. Over the extension relay it CANNOT work
-                                          # (Chrome's chrome.debugger forbids it); chrome-use
-                                          # errors with a hint. Carry your login into a launched
-                                          # session via `cookies export` | `cookies set --curl`.
+chrome-use upload @e5 file1.pdf        # upload file(s) — works over the extension relay too:
+                                          # chrome.debugger forbids setFileInputFiles, so the
+                                          # file's bytes are streamed into the page and rebuilt as
+                                          # a File there (chunked under native-messaging's 1 MiB cap).
+                                          # Works on file <input>s and drop/paste composers (e.g. X).
 chrome-use scroll down 500             # scroll page (up/down/left/right)
+chrome-use scroll down 700 --at 640,400 # wheel at a pixel — scrolls a cross-origin
+                                          # iframe (Payments/Stripe/checkout/KYC) that
+                                          # plain page scroll can't reach
+chrome-use scroll down 700 --frame 2    # scroll frame 2 from `chrome-use frames`
 chrome-use scrollintoview @e1          # scroll element into view
 chrome-use drag @e1 @e2                # drag and drop
 ```
 
+**Cross-origin iframes (embedded payment / checkout / KYC widgets — Google
+Payments, Stripe, etc.) — drive them by ref, never by screenshot.** `snapshot -i`
+pierces these out-of-process iframes and lists their elements by `@ref`
+(including input values); `get text --all-frames` reads their text. Then just act
+on the refs: `click @e`, `type @e`, `hover @e`, `dblclick @e`, `drag @a @b` all
+work into the iframe. Over the extension relay these are dispatched through the
+DOM (in the element's own frame), so they hit the right element in the right tab
+— a coordinate click/scroll there can drift onto whatever tab is in the
+foreground, so prefer refs. For below-the-fold content in such a frame, scroll it
+with `scroll down N --at x,y` (a pixel over the frame) or `--frame n`. For a
+postal/autocomplete box inside the frame, `type @e "…" --key-events`.
+
 ### When refs don't work or you don't want to snapshot
 
 Use semantic locators: