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.)

.github/workflows/release-binaries.yml

@@ -39,7 +39,6 @@ jobs:
         uses: actions/checkout@v6
         with:
           ref: ${{ github.event.inputs.tag || github.ref }}
-          fetch-depth: 0 # full history so the changelog step can diff tags
 
       - name: Setup Rust toolchain
         uses: dtolnay/rust-toolchain@stable
@@ -117,6 +116,16 @@ jobs:
     permissions:
       contents: write
     steps:
+      # The release job is separate from the build matrix and has no repo by
+      # default — check it out (full history + tags) so the changelog step has a
+      # git repo to diff. Without this, `git` failed with "not a git repository"
+      # and the changelog came out empty.
+      - name: Checkout
+        uses: actions/checkout@v6
+        with:
+          ref: ${{ github.event.inputs.tag || github.ref }}
+          fetch-depth: 0
+
       - name: Download all artifacts
         uses: actions/download-artifact@v8
         with:
@@ -138,16 +147,31 @@ jobs:
           git fetch --tags --force --quiet origin 2>/dev/null || true
           TAG="${{ github.event.inputs.tag || github.ref_name }}"
           PREV="$(git describe --tags --abbrev=0 "${TAG}^" 2>/dev/null || true)"
+          RANGE="${TAG}"
+          [ -n "$PREV" ] && RANGE="${PREV}..${TAG}"
+          # Group commit subjects by conventional-commit type so the notes are
+          # scannable ("what's new / what's fixed") instead of a flat dev log.
+          LOG="$(git log "$RANGE" --no-merges --pretty='%s' | grep -v '^chore(release)' || true)"
+          # NOTE: the job runs under `bash -e`. grep returning 1 (no match) and
+          # the `[ -n "$body" ]` test returning 1 (empty section) must NOT abort
+          # the script — otherwise a release whose commit range lacks a whole
+          # category (e.g. only `feat`, no `fix`) dies before writing the closing
+          # heredoc delimiter and the whole release step fails. `|| true` +
+          # `return 0` keep section() always-succeeding.
+          section() { # $1=header  $2=grep-pattern
+            local body; body="$(printf '%s\n' "$LOG" | grep -E "$2" | sed 's/^/- /' || true)"
+            [ -n "$body" ] && printf '\n### %s\n%s\n' "$1" "$body"
+            return 0
+          }
           {
             echo "notes<<__NOTES_EOF__"
             echo "## What changed"
-            echo ""
+            section "✨ Features" '^feat'
+            section "🐛 Fixes" '^fix'
+            section "🔧 Other" '^(perf|refactor|docs|build|ci|test|style|revert)'
             if [ -n "$PREV" ]; then
-              git log "${PREV}..${TAG}" --no-merges --pretty='- %s' | grep -v '^- chore(release)' || true
               echo ""
               echo "**Full changelog**: https://github.com/${{ github.repository }}/compare/${PREV}...${TAG}"
-            else
-              git log "${TAG}" --no-merges --pretty='- %s' | grep -v '^- chore(release)' || true
             fi
             echo "__NOTES_EOF__"
           } >> "$GITHUB_OUTPUT"

cli/Cargo.lock

@@ -290,7 +290,7 @@ checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
 
 [[package]]
 name = "chrome-use"
-version = "1.2.3"
+version = "1.5.18"
 dependencies = [
  "aes",
  "aes-gcm",

cli/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "chrome-use"
-version = "1.2.3"
+version = "1.5.18"
 edition = "2021"
 description = "Fast browser automation CLI for AI agents"
 license = "Apache-2.0"

cli/build.rs

@@ -3,6 +3,23 @@ use std::env;
 use std::fs;
 use std::path::Path;
 
+/// Embed the version of the `ab-connect` extension this CLI ships alongside, so
+/// `doctor` can tell a connected extension "you're older than what this CLI
+/// expects, update it." Read from the extension manifest at build time so it
+/// stays in sync with whatever extension version is in the same checkout/release
+/// (the ext is on its own 0.4.x line, separate from the CLI version). Falls back
+/// to "unknown" if the manifest can't be read.
+fn embed_extension_version() {
+    let manifest = Path::new("../extensions/ab-connect/manifest.json");
+    println!("cargo:rerun-if-changed=../extensions/ab-connect/manifest.json");
+    let version = fs::read_to_string(manifest)
+        .ok()
+        .and_then(|s| serde_json::from_str::<serde_json::Value>(&s).ok())
+        .and_then(|v| v.get("version").and_then(|x| x.as_str()).map(String::from))
+        .unwrap_or_else(|| "unknown".to_string());
+    println!("cargo:rustc-env=AB_CONNECT_VERSION={}", version);
+}
+
 /// Ensure `packages/dashboard/out/` exists so `rust-embed` doesn't fail during
 /// Rust-only dev builds where the dashboard hasn't been built. The placeholder
 /// `index.html` is only written when the directory is completely absent.
@@ -20,6 +37,7 @@ fn ensure_dashboard_dir() {
 
 fn main() {
     ensure_dashboard_dir();
+    embed_extension_version();
 
     let protocol_dir = Path::new("cdp-protocol");
     let out_dir = env::var("OUT_DIR").unwrap();

cli/src/commands.rs

@@ -30,12 +30,104 @@ pub enum ParseError {
     InvalidSessionName { name: String },
 }
 
+/// Top-level commands an agent is likely to mistype, used for "did you mean"
+/// 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",
+];
+
+/// Levenshtein distance, capped — small inputs only (command names).
+fn edit_distance(a: &str, b: &str) -> usize {
+    let a: Vec<char> = a.chars().collect();
+    let b: Vec<char> = b.chars().collect();
+    let mut prev: Vec<usize> = (0..=b.len()).collect();
+    let mut curr = vec![0usize; b.len() + 1];
+    for (i, &ca) in a.iter().enumerate() {
+        curr[0] = i + 1;
+        for (j, &cb) in b.iter().enumerate() {
+            let cost = if ca == cb { 0 } else { 1 };
+            curr[j + 1] = (prev[j + 1] + 1).min(curr[j] + 1).min(prev[j] + cost);
+        }
+        std::mem::swap(&mut prev, &mut curr);
+    }
+    prev[b.len()]
+}
+
+/// Closest known command within a small edit distance, or a prefix/substring
+/// match — `None` if nothing is close enough to suggest confidently.
+fn nearest_command(input: &str) -> Option<String> {
+    let lower = input.to_lowercase();
+    // Exact prefix/substring hits first (e.g. "session" -> "sessions").
+    if let Some(c) = KNOWN_COMMANDS
+        .iter()
+        .find(|c| c.starts_with(&lower) || lower.starts_with(**c))
+    {
+        return Some(c.to_string());
+    }
+    // Tolerance scales with length: short words get distance 1, longer get 2.
+    let max_dist = if lower.len() <= 4 { 1 } else { 2 };
+    KNOWN_COMMANDS
+        .iter()
+        .map(|c| (*c, edit_distance(&lower, c)))
+        .filter(|(_, d)| *d <= max_dist)
+        .min_by_key(|(_, d)| *d)
+        .map(|(c, _)| c.to_string())
+}
+
 impl ParseError {
     pub fn format(&self) -> String {
         match self {
-            ParseError::UnknownCommand { command } => {
-                format!("Unknown command: {}", command)
-            }
+            ParseError::UnknownCommand { command } => match nearest_command(command) {
+                Some(suggestion) => format!(
+                    "Unknown command: {}\nDid you mean: chrome-use {}?",
+                    command, suggestion
+                ),
+                None => format!("Unknown command: {}", command),
+            },
             ParseError::UnknownSubcommand {
                 subcommand,
                 valid_options,
@@ -370,6 +462,12 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             if flags.provider.is_some() {
                 nav_cmd["waitUntil"] = json!("none");
             }
+            // `--reuse-tab`: adopt an existing tab already on this URL instead of
+            // navigating/spawning a new one (issue #21 — avoids duplicate tabs on
+            // rebind, preserves in-page state).
+            if rest.iter().any(|a| *a == "--reuse-tab" || *a == "--reuse") {
+                nav_cmd["reuseTab"] = json!(true);
+            }
             // Explicit readiness override (issue #10): SPAs whose `load` event
             // never fires (a long-lived XHR/websocket holds it open) hang out the
             // load-event wait. `--wait-until domcontentloaded` returns as soon as
@@ -418,6 +516,9 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
         // === Core Actions ===
         "click" => {
             let new_tab = rest.contains(&"--new-tab");
+            // `--follow`: if the click opens a new tab, switch the active tab to
+            // it (default reports the opened tab but stays put) (issue #24-A).
+            let follow = rest.contains(&"--follow");
             // Coordinate click as a first-class form (issue #8.4): when the only
             // handle is a pixel position, no element/selector is needed.
             //   click <x> <y>          e.g. click 449 320
@@ -426,23 +527,27 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             let coord_args: Vec<&str> = rest
                 .iter()
                 .copied()
-                .filter(|a| *a != "--new-tab" && *a != "--coords")
+                .filter(|a| !a.starts_with("--"))
                 .collect();
             if let Some((x, y)) = parse_coords(&coord_args) {
                 return Ok(json!({ "id": id, "action": "click", "x": x, "y": y }));
             }
             let sel = rest
                 .iter()
-                .find(|arg| **arg != "--new-tab")
+                .find(|arg| !arg.starts_with("--"))
                 .ok_or_else(|| ParseError::MissingArguments {
                     context: "click".to_string(),
-                    usage: "click <selector> | click <x> <y> | click --coords <x>,<y> [--new-tab]",
+                    usage:
+                        "click <selector> | click <x> <y> | click --coords <x>,<y> [--new-tab] [--follow]",
                 })?;
+            let mut cmd = json!({ "id": id, "action": "click", "selector": sel });
             if new_tab {
-                Ok(json!({ "id": id, "action": "click", "selector": sel, "newTab": true }))
-            } else {
-                Ok(json!({ "id": id, "action": "click", "selector": sel }))
+                cmd["newTab"] = json!(true);
             }
+            if follow {
+                cmd["follow"] = json!(true);
+            }
+            Ok(cmd)
         }
         "dblclick" => {
             let sel = rest.first().ok_or_else(|| ParseError::MissingArguments {
@@ -459,20 +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:
@@ -577,11 +693,27 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
 
         // === Keyboard ===
         "press" | "key" => {
-            let key = rest.first().ok_or_else(|| ParseError::MissingArguments {
-                context: "press".to_string(),
-                usage: "press <key>",
+            let key = rest.iter().find(|a| !a.starts_with("--")).ok_or_else(|| {
+                ParseError::MissingArguments {
+                    context: "press".to_string(),
+                    usage: "press <key> [--hold <ms>]",
+                }
             })?;
-            Ok(json!({ "id": id, "action": "press", "key": key }))
+            let mut c = json!({ "id": id, "action": "press", "key": key });
+            // `--hold <ms>`: hold the key down for <ms> then release, timed inside
+            // the daemon (one round-trip, no shell-sleep jitter) — for games and
+            // hold-to-charge where keydown+sleep+keyup over 3 round-trips is too
+            // imprecise.
+            if let Some(i) = rest.iter().position(|a| *a == "--hold") {
+                let ms = rest.get(i + 1).and_then(|s| s.parse::<u64>().ok()).ok_or(
+                    ParseError::MissingArguments {
+                        context: "press --hold".to_string(),
+                        usage: "press <key> --hold <ms>",
+                    },
+                )?;
+                c["hold"] = json!(ms);
+            }
+            Ok(c)
         }
         "keydown" => {
             let key = rest.first().ok_or_else(|| ParseError::MissingArguments {
@@ -647,10 +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 {
@@ -825,17 +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
@@ -866,6 +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);
             }
@@ -996,8 +1202,31 @@ 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" => Ok(json!({ "id": id, "action": "close" })),
+        "close" | "quit" | "exit" => {
+            // `close <tab>` closes only that tab (and the output says "Tab
+            // closed"); bare `close` closes the browser/session. `close --all` is
+            // intercepted earlier in the dispatcher. Previously `close t12` still
+            // ran a browser close and alarmingly printed "Browser closed" (#26).
+            if let Some(tab_ref) = rest.iter().find(|a| !a.starts_with("--")) {
+                Ok(json!({ "id": id, "action": "tab_close", "tabId": tab_ref }))
+            } else {
+                Ok(json!({ "id": id, "action": "close" }))
+            }
+        }
+
+        // The active tab's stable handle — `targetId` survives cross-process
+        // navigation and is reusable across sessions, so an agent can hold it
+        // instead of re-deriving "which tab is live" from `tabs` each step (#26).
+        "current" => Ok(json!({ "id": id, "action": "current" })),
 
         // === Inspect ===
         "inspect" => Ok(json!({ "id": id, "action": "inspect" })),
@@ -1252,6 +1481,11 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
         // === Get ===
         "get" => parse_get(&rest, &id),
 
+        // List every frame the session can reach (top + same-process child
+        // frames + out-of-process iframes), with a text-length per frame so you
+        // can see where a listing's description actually lives (issue #27).
+        "frames" => Ok(json!({ "id": id, "action": "frames" })),
+
         // Top-level shortcuts for `get <x>` status reads — users naturally type
         // `chrome-use url` / `cdp-url` / `title` without the `get` prefix
         // (and expect `cdp-url`/`cdp_url` to work interchangeably).
@@ -1553,11 +1787,16 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
                     }
                     Ok(cmd)
                 }
-                Some(tab_ref) => Ok(json!({
-                    "id": id,
-                    "action": "tab_switch",
-                    "tabId": tab_ref,
-                })),
+                Some(tab_ref) => {
+                    // `tab <ref> --activate` (alias `--front`) switches to the tab
+                    // AND raises it to the foreground — for handing a specific tab
+                    // to the human (SMS code, captcha) (issue #24-C).
+                    let mut cmd = json!({ "id": id, "action": "tab_switch", "tabId": tab_ref });
+                    if rest.iter().any(|a| *a == "--activate" || *a == "--front") {
+                        cmd["activate"] = json!(true);
+                    }
+                    Ok(cmd)
+                }
                 None => {
                     let mut cmd = json!({ "id": id, "action": "tab_list" });
                     if full {
@@ -2339,11 +2578,42 @@ fn parse_get(rest: &[&str], id: &str) -> Result<Value, ParseError> {
 
     match rest.first().copied() {
         Some("text") => {
-            let sel = rest.get(1).ok_or_else(|| ParseError::MissingArguments {
-                context: "get text".to_string(),
-                usage: "get text <selector>",
-            })?;
-            Ok(json!({ "id": id, "action": "gettext", "selector": sel }))
+            // `get text --all-frames` aggregates visible text across every
+            // frame, including out-of-process iframes invisible to the top
+            // document (issue #27). The selector is ignored in this mode.
+            let all_frames = rest[1..]
+                .iter()
+                .any(|a| matches!(*a, "--all-frames" | "--frames" | "-a"));
+            if all_frames {
+                return Ok(json!({ "id": id, "action": "gettext", "allFrames": true }));
+            }
+            // `get text --pierce` reads through CLOSED shadow DOM / child docs
+            // via the CDP DOM tree — content eval/innerText can't reach, e.g. an
+            // extension's injected panel in a closed shadow root (issue #30).
+            let pierce = rest[1..]
+                .iter()
+                .any(|a| matches!(*a, "--pierce" | "--shadow" | "--deep"));
+            if pierce {
+                return Ok(json!({ "id": id, "action": "gettext", "pierce": true }));
+            }
+            // `get text --main` returns the main-content region (readability),
+            // skipping header/nav/footer/sidebar boilerplate (issue #27).
+            let main = rest[1..]
+                .iter()
+                .any(|a| matches!(*a, "--main" | "--readable" | "-m"));
+            if main {
+                return Ok(json!({ "id": id, "action": "gettext", "main": true }));
+            }
+            // `get text` with no selector reads the WHOLE PAGE and now defaults
+            // to cross-frame aggregation, so an agent gets a page's iframed
+            // content (listing descriptions etc.) without having to know about
+            // `--all-frames` (#27). On a single-frame page this is identical to
+            // the old body read; multi-frame pages get the child frames too —
+            // a strict superset. An explicit selector stays element-scoped.
+            match rest.iter().skip(1).find(|a| !a.starts_with("--")).copied() {
+                Some(sel) => Ok(json!({ "id": id, "action": "gettext", "selector": sel })),
+                None => Ok(json!({ "id": id, "action": "gettext", "allFrames": true })),
+            }
         }
         Some("html") => {
             let sel = rest.get(1).ok_or_else(|| ParseError::MissingArguments {
@@ -3575,6 +3845,40 @@ mod tests {
         assert_eq!(cmd["url"], "https://example.com");
     }
 
+    #[test]
+    fn test_press_plain_and_hold() {
+        let cmd = parse_command(&args("press d"), &default_flags()).unwrap();
+        assert_eq!(cmd["action"], "press");
+        assert_eq!(cmd["key"], "d");
+        assert!(cmd.get("hold").is_none());
+
+        let held = parse_command(&args("press d --hold 800"), &default_flags()).unwrap();
+        assert_eq!(held["key"], "d");
+        assert_eq!(held["hold"], 800);
+
+        // Missing/invalid duration is an error, not a silent no-hold.
+        assert!(parse_command(&args("press d --hold"), &default_flags()).is_err());
+        assert!(parse_command(&args("press d --hold abc"), &default_flags()).is_err());
+    }
+
+    #[test]
+    fn test_navigate_reuse_tab_flag() {
+        let cmd = parse_command(
+            &args("open https://example.com --reuse-tab"),
+            &default_flags(),
+        )
+        .unwrap();
+        assert_eq!(cmd["action"], "navigate");
+        assert_eq!(cmd["reuseTab"], true);
+        // Alias.
+        let cmd2 =
+            parse_command(&args("open https://example.com --reuse"), &default_flags()).unwrap();
+        assert_eq!(cmd2["reuseTab"], true);
+        // Absent by default.
+        let cmd3 = parse_command(&args("open https://example.com"), &default_flags()).unwrap();
+        assert!(cmd3.get("reuseTab").is_none());
+    }
+
     #[test]
     fn test_navigate_with_headers() {
         let mut flags = default_flags();
@@ -3723,6 +4027,21 @@ mod tests {
         assert!(cmd.get("x").is_none());
     }
 
+    #[test]
+    fn test_click_follow_flag() {
+        // `--follow` sets the flag; the selector is still found even with the flag
+        // before it (issue #24-A).
+        let cmd = parse_command(&args("click @e5 --follow"), &default_flags()).unwrap();
+        assert_eq!(cmd["selector"], "@e5");
+        assert_eq!(cmd["follow"], true);
+        let cmd2 = parse_command(&args("click --follow @e5"), &default_flags()).unwrap();
+        assert_eq!(cmd2["selector"], "@e5");
+        assert_eq!(cmd2["follow"], true);
+        // Absent by default.
+        let plain = parse_command(&args("click @e5"), &default_flags()).unwrap();
+        assert!(plain.get("follow").is_none());
+    }
+
     #[test]
     fn test_tabs_alias_lists() {
         assert_eq!(
@@ -3848,6 +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]
@@ -3931,6 +4272,28 @@ mod tests {
         assert_eq!(cmd["tabId"], "docs");
     }
 
+    #[test]
+    fn test_close_tab_vs_browser() {
+        // `close <tab>` closes that tab (says "Tab closed"); bare `close` closes
+        // the browser (#26).
+        let tab = parse_command(&args("close t12"), &default_flags()).unwrap();
+        assert_eq!(tab["action"], "tab_close");
+        assert_eq!(tab["tabId"], "t12");
+        let browser = parse_command(&args("close"), &default_flags()).unwrap();
+        assert_eq!(browser["action"], "close");
+        // `quit`/`exit` aliases still browser-close.
+        assert_eq!(
+            parse_command(&args("quit"), &default_flags()).unwrap()["action"],
+            "close"
+        );
+    }
+
+    #[test]
+    fn test_current_command() {
+        let cmd = parse_command(&args("current"), &default_flags()).unwrap();
+        assert_eq!(cmd["action"], "current");
+    }
+
     #[test]
     fn test_tab_sends_string_tab_id() {
         let cmd = parse_command(&args("tab t2"), &default_flags()).unwrap();
@@ -4110,6 +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();
@@ -4630,12 +5011,99 @@ mod tests {
     }
 
     #[test]
-    fn test_get_text_missing_selector() {
-        let result = parse_command(&args("get text"), &default_flags());
-        assert!(result.is_err());
-        let err = result.unwrap_err();
-        assert!(matches!(err, ParseError::MissingArguments { .. }));
-        assert!(err.format().contains("get text"));
+    fn test_get_text_defaults_to_all_frames() {
+        // `get text` with no selector now reads the whole page across ALL frames
+        // by default (#27), so iframed content isn't silently missed. (Was: a
+        // top-frame `body` read, #24-D.)
+        let cmd = parse_command(&args("get text"), &default_flags()).unwrap();
+        assert_eq!(cmd["action"], "gettext");
+        assert_eq!(cmd["allFrames"], true);
+        assert!(cmd.get("selector").is_none());
+        // An explicit selector still wins and stays element-scoped.
+        let cmd2 = parse_command(&args("get text h1"), &default_flags()).unwrap();
+        assert_eq!(cmd2["selector"], "h1");
+        assert!(cmd2.get("allFrames").is_none());
+        // `text` top-level shortcut behaves the same.
+        let cmd3 = parse_command(&args("text"), &default_flags()).unwrap();
+        assert_eq!(cmd3["allFrames"], true);
+    }
+
+    #[test]
+    fn test_get_text_all_frames() {
+        // `--all-frames` switches to whole-page, cross-frame aggregation and
+        // drops the selector (issue #27).
+        for variant in ["get text --all-frames", "get text --frames", "text -a"] {
+            let cmd = parse_command(&args(variant), &default_flags()).unwrap();
+            assert_eq!(cmd["action"], "gettext", "{variant}");
+            assert_eq!(cmd["allFrames"], true, "{variant}");
+            assert!(cmd.get("selector").is_none(), "{variant}");
+        }
+        // A flag mixed with a selector still triggers all-frames.
+        let cmd = parse_command(&args("get text body --all-frames"), &default_flags()).unwrap();
+        assert_eq!(cmd["allFrames"], true);
+        // Without the flag, a leading flag-like token is skipped for the selector.
+        let cmd = parse_command(&args("get text main"), &default_flags()).unwrap();
+        assert_eq!(cmd["selector"], "main");
+        assert!(cmd.get("allFrames").is_none());
+    }
+
+    #[test]
+    fn test_frames_command() {
+        let cmd = parse_command(&args("frames"), &default_flags()).unwrap();
+        assert_eq!(cmd["action"], "frames");
+    }
+
+    #[test]
+    fn test_nearest_command_suggestions() {
+        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")
+        );
+        // Nonsense with no close match stays silent.
+        assert_eq!(nearest_command("xyzzy"), None);
+        // The unknown-command error embeds the suggestion.
+        let err = ParseError::UnknownCommand {
+            command: "sesions".to_string(),
+        };
+        assert!(err.format().contains("Did you mean: chrome-use sessions?"));
+    }
+
+    #[test]
+    fn test_get_text_main() {
+        for variant in ["get text --main", "get text --readable", "text -m"] {
+            let cmd = parse_command(&args(variant), &default_flags()).unwrap();
+            assert_eq!(cmd["action"], "gettext", "{variant}");
+            assert_eq!(cmd["main"], true, "{variant}");
+            assert!(cmd.get("selector").is_none(), "{variant}");
+        }
+    }
+
+    #[test]
+    fn test_get_text_pierce() {
+        for variant in ["get text --pierce", "get text --shadow", "text --deep"] {
+            let cmd = parse_command(&args(variant), &default_flags()).unwrap();
+            assert_eq!(cmd["action"], "gettext", "{variant}");
+            assert_eq!(cmd["pierce"], true, "{variant}");
+            assert!(cmd.get("selector").is_none(), "{variant}");
+        }
+    }
+
+    #[test]
+    fn test_tab_activate_flag() {
+        let plain = parse_command(&args("tab t3"), &default_flags()).unwrap();
+        assert_eq!(plain["action"], "tab_switch");
+        assert!(plain.get("activate").is_none());
+
+        let act = parse_command(&args("tab t3 --activate"), &default_flags()).unwrap();
+        assert_eq!(act["action"], "tab_switch");
+        assert_eq!(act["tabId"], "t3");
+        assert_eq!(act["activate"], true);
+        // `--front` alias.
+        let front = parse_command(&args("tab t3 --front"), &default_flags()).unwrap();
+        assert_eq!(front["activate"], true);
     }
 
     // === Protocol alignment tests ===
@@ -5529,6 +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 +449,69 @@ 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.
+fn relay_ext_version_path() -> PathBuf {
+    relay_url_path().with_file_name("relay-ext-version")
+}
+
+/// Version of the connected `ab-connect` extension, if the host learned it from
+/// the extension's `hello`. `None` when no extension has connected since the
+/// host started, or the extension predates version reporting.
+pub fn relay_ext_version() -> Option<String> {
+    let s = std::fs::read_to_string(relay_ext_version_path())
+        .ok()?
+        .trim()
+        .to_string();
+    if s.is_empty() {
+        None
+    } else {
+        Some(s)
+    }
+}
+
 /// Hidden `__nm-host` mode: launched by Chrome for the ab-connect extension.
 ///
 /// Bridges the extension (native-messaging stdio, envelope protocol) to a local
@@ -570,6 +633,15 @@ async fn nm_host_main() {
             Ok(v) => v,
             Err(_) => continue,
         };
+        // Extension version handshake: record it next to the relay URL so
+        // `doctor` can report which extension build is live (and whether it's
+        // behind). Best-effort; the message carries no CDP payload.
+        if v.get("method").and_then(|m| m.as_str()) == Some("hello") {
+            if let Some(ver) = v.get("version").and_then(|x| x.as_str()) {
+                let _ = std::fs::write(relay_ext_version_path(), ver);
+            }
+            continue;
+        }
         let outs = {
             let mut s = state.lock().await;
             s.handle_ext_message(&v, "")
@@ -602,6 +674,7 @@ async fn nm_host_main() {
     }
     nm_log("[nm-host] stdin EOF — Chrome closed the port");
     let _ = std::fs::remove_file(relay_url_path());
+    let _ = std::fs::remove_file(relay_ext_version_path());
 }
 
 #[allow(clippy::too_many_arguments)]

cli/src/connection.rs

@@ -642,6 +642,22 @@ fn kill_stale_daemon(session: &str) {
     cleanup_stale_files(session);
 }
 
+/// Kill every per-session daemon worker (SIGTERM→SIGKILL + sidecar cleanup),
+/// leaving the Chrome-launched `__nm-host` native-messaging bridge alone — it's
+/// not a tracked session daemon, so the extension relay stays up. Returns the
+/// session names that were stopped. Powers `chrome-use daemon restart`, which
+/// clears corrupted/cross-leaked daemon state (e.g. after a version-mismatch
+/// restart) without the user resorting to `pgrep`/`kill` (issue #20).
+pub fn restart_all_daemons() -> Vec<String> {
+    let inventory = walk_daemons();
+    let mut stopped = Vec::new();
+    for session in &inventory.sessions {
+        kill_stale_daemon(&session.name);
+        stopped.push(session.name.clone());
+    }
+    stopped
+}
+
 pub fn ensure_daemon(session: &str, opts: &DaemonOptions) -> Result<DaemonResult, String> {
     // Socket connectivity is the sole liveness check — no PID check — so
     // callers in a different PID namespace (e.g. unshare) can still reuse
@@ -1182,6 +1198,55 @@ mod tests {
         let _ = fs::remove_dir(&dir);
     }
 
+    #[test]
+    fn test_restart_all_daemons_empty_dir() {
+        let dir = std::env::temp_dir().join("ab-test-restart-empty");
+        let _ = fs::create_dir_all(&dir);
+        let _guard = EnvGuard::new(&["AGENT_BROWSER_SOCKET_DIR", "XDG_RUNTIME_DIR"]);
+        _guard.set("AGENT_BROWSER_SOCKET_DIR", dir.to_str().unwrap());
+
+        // No daemons registered → nothing to stop, and it must not blow up.
+        assert!(restart_all_daemons().is_empty());
+
+        let _ = fs::remove_dir(&dir);
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn test_restart_all_daemons_kills_live_session() {
+        let dir = std::env::temp_dir().join("ab-test-restart-live");
+        let _ = fs::create_dir_all(&dir);
+        let _guard = EnvGuard::new(&["AGENT_BROWSER_SOCKET_DIR", "XDG_RUNTIME_DIR"]);
+        _guard.set("AGENT_BROWSER_SOCKET_DIR", dir.to_str().unwrap());
+
+        // Spawn a real, killable child and register it as a session daemon.
+        let mut child = Command::new("sleep")
+            .arg("30")
+            .spawn()
+            .expect("spawn sleep");
+        let pid = child.id();
+        let _ = fs::write(dir.join("rktest.pid"), pid.to_string());
+        let _ = fs::write(get_socket_path("rktest"), b"");
+
+        let stopped = restart_all_daemons();
+        assert!(
+            stopped.contains(&"rktest".to_string()),
+            "stopped: {:?}",
+            stopped
+        );
+
+        // Reap the killed child first — until the parent waits, it lingers as a
+        // zombie that still answers `kill(pid, 0)`, so is_pid_alive would lie.
+        let _ = child.wait();
+        assert!(!is_pid_alive(pid));
+
+        // Sidecars are cleaned up.
+        assert!(!dir.join("rktest.pid").exists());
+        assert!(!get_socket_path("rktest").exists());
+
+        let _ = fs::remove_dir(&dir);
+    }
+
     #[test]
     fn test_cleanup_stale_files_removes_version() {
         let dir = std::env::temp_dir().join("ab-test-cleanup-version");

cli/src/doctor/mod.rs

@@ -18,6 +18,7 @@ mod launch;
 mod network;
 mod providers;
 mod security;
+mod versions;
 
 use serde_json::{json, Value};
 
@@ -97,6 +98,7 @@ pub fn run_doctor(opts: DoctorOptions) -> i32 {
     let mut fixed: Vec<String> = Vec::new();
 
     environment::check(&mut checks);
+    versions::check(&mut checks);
     chrome::check(&mut checks);
     daemon::check(&mut checks);
     config::check(&mut checks);

cli/src/doctor/versions.rs

@@ -0,0 +1,89 @@
+//! Version-coherence checks across all four moving parts: the CLI binary, the
+//! per-session daemons (covered by `daemon.rs`), the connected `ab-connect`
+//! extension, and the bundled skill. The extension was previously a black box —
+//! nothing reported which build was live — so a user could sit on an old
+//! extension with no signal. The extension now reports its version over the
+//! relay (`hello`), the host records it, and this surfaces it in one place.
+
+use super::{Check, Status};
+use crate::{connect, upgrade};
+
+pub(super) fn check(checks: &mut Vec<Check>) {
+    let category = "Versions";
+    let cli_version = env!("CARGO_PKG_VERSION");
+
+    // CLI — compare against the latest seen by the background update check.
+    match upgrade::cached_latest_version() {
+        Some(latest) if upgrade::version_is_newer(&latest, cli_version) => {
+            checks.push(
+                Check::new(
+                    "versions.cli",
+                    category,
+                    Status::Warn,
+                    format!("CLI {cli_version} (newer available: {latest})"),
+                )
+                .with_fix("chrome-use upgrade".to_string()),
+            );
+        }
+        _ => {
+            checks.push(Check::new(
+                "versions.cli",
+                category,
+                Status::Pass,
+                format!("CLI {cli_version}"),
+            ));
+        }
+    }
+
+    // Extension — the build this CLI shipped alongside (embedded at compile time
+    // from the extension manifest) is what we expect to be running.
+    let expected_ext = env!("AB_CONNECT_VERSION");
+    match connect::relay_ext_version() {
+        Some(ext) if upgrade::version_is_newer(expected_ext, &ext) => {
+            checks.push(
+                Check::new(
+                    "versions.extension",
+                    category,
+                    Status::Warn,
+                    format!("extension {ext} is behind the bundled {expected_ext}"),
+                )
+                .with_fix(
+                    "update ab-connect in Chrome: chrome://extensions \u{2192} reload \
+                     (or wait for the Web Store auto-update)"
+                        .to_string(),
+                ),
+            );
+        }
+        Some(ext) => {
+            checks.push(Check::new(
+                "versions.extension",
+                category,
+                Status::Pass,
+                format!("extension {ext}"),
+            ));
+        }
+        None => {
+            checks.push(Check::new(
+                "versions.extension",
+                category,
+                Status::Info,
+                format!(
+                    "extension not connected (or it predates version reporting — \
+                     expected {expected_ext})"
+                ),
+            ));
+        }
+    }
+
+    // Skill — ships inside the same release artifact as the binary, so it's
+    // version-locked here. Copies made elsewhere via `skills add` aren't.
+    checks.push(Check::new(
+        "versions.skill",
+        category,
+        Status::Info,
+        format!(
+            "skills bundled with this CLI ({cli_version}); copies made via `skills add` \
+             elsewhere may be stale — re-run to refresh"
+        ),
+    ));
+}

cli/src/main.rs

@@ -29,8 +29,8 @@ use windows_sys::Win32::System::Threading::OpenProcess;
 
 use commands::{gen_id, parse_command, ParseError};
 use connection::{
-    cleanup_stale_files, ensure_daemon, get_socket_dir, is_pid_alive, send_command, walk_daemons,
-    DaemonOptions,
+    cleanup_stale_files, ensure_daemon, get_socket_dir, is_pid_alive, restart_all_daemons,
+    send_command, walk_daemons, DaemonOptions,
 };
 use flags::{clean_args, parse_flags, Flags};
 use install::run_install;
@@ -320,6 +320,94 @@ fn run_session(args: &[String], session: &str, json_mode: bool) {
     }
 }
 
+/// `chrome-use daemon <restart|status>` — manage the per-session daemon workers
+/// without resorting to `pgrep`/`kill`. `restart` clears corrupted or
+/// cross-leaked daemon state (e.g. after a mid-session `chrome-use upgrade`
+/// where stale tab handles bleed across sessions, issue #20) by killing every
+/// session worker. The Chrome-launched `__nm-host` native-messaging bridge is
+/// NOT a tracked session daemon, so the extension relay survives a restart —
+/// the next command spins up a fresh, clean daemon against the same live Chrome.
+fn run_daemon(args: &[String], json_mode: bool) {
+    match args.get(1).map(|s| s.as_str()) {
+        Some("restart") => {
+            let stopped = restart_all_daemons();
+            let relay_up = connect::relay_url().is_some();
+            if json_mode {
+                print_json_value(json!({
+                    "success": true,
+                    "data": { "stopped": stopped, "count": stopped.len(), "relay": relay_up },
+                }));
+            } else if stopped.is_empty() {
+                println!("No session daemons running — nothing to restart.");
+                if relay_up {
+                    println!(
+                        "{}",
+                        color::dim("Extension relay still up; next command starts a fresh daemon.")
+                    );
+                }
+            } else {
+                for s in &stopped {
+                    println!("{} Stopped daemon: {}", color::green("✓"), s);
+                }
+                println!(
+                    "{}",
+                    color::dim(if relay_up {
+                        "Extension relay (__nm-host) left running; next command starts a fresh daemon."
+                    } else {
+                        "Next command starts a fresh daemon."
+                    })
+                );
+            }
+        }
+        Some("status") | Some("list") => {
+            let inventory = walk_daemons();
+            let relay_up = connect::relay_url().is_some();
+            if json_mode {
+                let sessions: Vec<_> = inventory
+                    .sessions
+                    .iter()
+                    .map(|s| json!({ "name": s.name, "pid": s.pid, "version": s.version }))
+                    .collect();
+                print_json_value(json!({
+                    "success": true,
+                    "data": { "sessions": sessions, "relay": relay_up },
+                }));
+            } else if inventory.sessions.is_empty() {
+                println!("No session daemons running.");
+                if relay_up {
+                    println!("{}", color::dim("Extension relay (__nm-host): up"));
+                }
+            } else {
+                println!("Session daemons:");
+                for s in &inventory.sessions {
+                    let ver = s
+                        .version
+                        .as_deref()
+                        .map(|v| format!(" {}", color::dim(&format!("(v{})", v))))
+                        .unwrap_or_default();
+                    println!("  {} pid {}{}", s.name, s.pid, ver);
+                }
+                if relay_up {
+                    println!("{}", color::dim("Extension relay (__nm-host): up"));
+                }
+            }
+        }
+        other => {
+            eprintln!(
+                "{} usage: chrome-use daemon <restart|status>",
+                color::error_indicator()
+            );
+            if let Some(unknown) = other {
+                eprintln!(
+                    "{}",
+                    color::dim(&format!("  unknown subcommand: {}", unknown))
+                );
+            }
+            exit(2);
+        }
+    }
+}
+
 fn get_dashboard_pid_path() -> std::path::PathBuf {
     get_socket_dir().join("dashboard.pid")
 }
@@ -799,6 +887,20 @@ fn main() {
         return;
     }
 
+    // Handle daemon management (doesn't talk to a daemon — it manages them).
+    if clean.first().map(|s| s.as_str()) == Some("daemon") {
+        run_daemon(&clean, flags.json);
+        return;
+    }
+
+    // `sessions` is a natural top-level guess for "list my sessions" (the skill
+    // advertises sessions as a feature) — route it to the daemon inventory the
+    // same way `daemon status` does (issue #29).
+    if clean.first().map(|s| s.as_str()) == Some("sessions") {
+        run_daemon(&["sessions".to_string(), "status".to_string()], flags.json);
+        return;
+    }
+
     // Handle close --all: close all active sessions
     if matches!(
         clean.first().map(|s| s.as_str()),
@@ -1267,6 +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

@@ -1332,6 +1332,7 @@ pub async fn execute_command(cmd: &Value, state: &mut DaemonState) -> Value {
         "uncheck" => handle_uncheck(cmd, state).await,
         "wait" => handle_wait(cmd, state).await,
         "gettext" => handle_gettext(cmd, state).await,
+        "frames" => handle_frames(cmd, state).await,
         "getattribute" => handle_getattribute(cmd, state).await,
         "isvisible" => handle_isvisible(cmd, state).await,
         "isenabled" => handle_isenabled(cmd, state).await,
@@ -1340,6 +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,
@@ -1395,6 +1397,7 @@ pub async fn execute_command(cmd: &Value, state: &mut DaemonState) -> Value {
         "count" => handle_count(cmd, state).await,
         "styles" => handle_styles(cmd, state).await,
         "bringtofront" => handle_bringtofront(state).await,
+        "current" => handle_current(state).await,
         "timezone" => handle_timezone(cmd, state).await,
         "locale" => handle_locale(cmd, state).await,
         "geolocation" => handle_geolocation(cmd, state).await,
@@ -2531,6 +2534,20 @@ async fn handle_navigate(cmd: &Value, state: &mut DaemonState) -> Result<Value,
     state.ref_map.clear();
     state.iframe_sessions.clear();
     state.active_frame_id = None;
+
+    // `--reuse-tab`: if a tab already shows this URL (same origin+path), switch
+    // to it instead of navigating — preserves any in-page state and stops
+    // re-`open` from piling up duplicate tabs on rebind (issue #21).
+    if cmd
+        .get("reuseTab")
+        .and_then(|v| v.as_bool())
+        .unwrap_or(false)
+    {
+        if let Ok(Some(switched)) = mgr.reuse_tab_for_url(url).await {
+            return Ok(switched);
+        }
+    }
+
     let result = mgr.navigate(url, wait_until).await?;
     // Adaptive humanize: sample the freshly loaded page for known behavioural
     // anti-bot vendors and escalate this session to Human if any are present.
@@ -2874,7 +2891,32 @@ async fn handle_snapshot(cmd: &Value, state: &mut DaemonState) -> Result<Value,
         })
         .collect();
 
-    Ok(json!({ "snapshot": tree, "origin": url, "refs": refs }))
+    let ref_count = refs.len();
+    let mut out = json!({ "snapshot": tree, "origin": url, "refs": refs });
+
+    // Canvas/WebGL apps (games, map/3D viewers, drawing tools) paint to a
+    // <canvas> and expose almost no accessibility tree, so `snapshot` comes back
+    // near-empty and agents get stuck looking for refs that will never exist
+    // (dogfood: the Dead Cell game). When the tree is sparse but a canvas
+    // dominates the viewport, tell them to switch to the screenshot-driven path.
+    if ref_count < 3 {
+        let canvas_js =
+            "(() => { const c = document.querySelector('canvas'); if (!c) return false; \
+                         const r = c.getBoundingClientRect(); \
+                         return r.width * r.height > innerWidth * innerHeight * 0.5; })()";
+        if let Ok(v) = mgr.evaluate(canvas_js, None).await {
+            if v.as_bool() == Some(true) {
+                out["note"] = json!(
+                    "This page renders to a <canvas> (game / WebGL / editor) and exposes almost no \
+                     accessibility tree — refs won't help. Use `screenshot` to see it, coordinate \
+                     `click <x> <y>` to interact, and `keydown`/`keyup`/`press` for keyboard \
+                     (hold-to-move: `keydown d` … `keyup d`)."
+                );
+            }
+        }
+    }
+
+    Ok(out)
 }
 
 /// Resolve a (possibly relative) saved-file path to an absolute one so the CLI
@@ -2958,6 +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 {
@@ -3077,6 +3127,15 @@ async fn handle_click(cmd: &Value, state: &mut DaemonState) -> Result<Value, Str
 
     let button = cmd.get("button").and_then(|v| v.as_str()).unwrap_or("left");
     let click_count = cmd.get("clickCount").and_then(|v| v.as_i64()).unwrap_or(1) as i32;
+    let follow = cmd.get("follow").and_then(|v| v.as_bool()).unwrap_or(false);
+
+    // Snapshot tracked targets so we can tell if this click opened a NEW tab
+    // (target=_blank link / window.open). On the relay the new tab is discovered
+    // passively and doesn't steal focus (#7/#8.1), so without surfacing it the
+    // post-click snapshot shows the OLD page and looks like the click failed
+    // (issue #24-A).
+    let before: std::collections::HashSet<String> =
+        mgr.pages_list().into_iter().map(|p| p.target_id).collect();
 
     interaction::click(
         &mgr.client,
@@ -3089,7 +3148,26 @@ async fn handle_click(cmd: &Value, state: &mut DaemonState) -> Result<Value, Str
     )
     .await?;
 
-    Ok(json!({ "clicked": selector }))
+    // Give a just-opened tab a moment to register, then look for it.
+    let mgr = state.browser.as_mut().ok_or("Browser not launched")?;
+    tokio::time::sleep(std::time::Duration::from_millis(150)).await;
+    let opened = mgr.adopt_newly_opened(&before).await;
+
+    let mut out = json!({ "clicked": selector });
+    if let Some(page) = opened {
+        let tab_id = super::browser::format_tab_id(page.tab_id);
+        out["openedTab"] = json!({ "tabId": tab_id, "url": page.url, "title": page.title });
+        // `--follow`: switch the active tab to the newly-opened one (default is
+        // to report it but stay put, so multi-tab flows aren't hijacked).
+        if follow {
+            state.ref_map.clear();
+            state.iframe_sessions.clear();
+            state.active_frame_id = None;
+            let _ = mgr.tab_switch_by_id(page.tab_id).await;
+            out["followed"] = json!(true);
+        }
+    }
+    Ok(out)
 }
 
 async fn handle_dblclick(cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {
@@ -3147,6 +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
@@ -3158,7 +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 }));
     }
 
@@ -3182,6 +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 }))
@@ -3295,6 +3389,16 @@ async fn handle_press(cmd: &Value, state: &mut DaemonState) -> Result<Value, Str
     // Parse modifier+key chords like "Control+a", "Shift+Enter", "Control+Shift+a"
     let (actual_key, modifiers) = parse_key_chord(key);
 
+    // `--hold <ms>`: keyDown, wait, keyUp — all inside the daemon so the hold
+    // duration is precise (no shell-sleep / round-trip jitter). For games
+    // (hold-to-move/charge) and any press-and-hold interaction.
+    if let Some(ms) = cmd.get("hold").and_then(|v| v.as_u64()) {
+        interaction::dispatch_single_key(&mgr.client, &session_id, &actual_key, "keyDown").await?;
+        tokio::time::sleep(std::time::Duration::from_millis(ms)).await;
+        interaction::dispatch_single_key(&mgr.client, &session_id, &actual_key, "keyUp").await?;
+        return Ok(json!({ "pressed": key, "heldMs": ms }));
+    }
+
     interaction::press_key_with_modifiers(&mgr.client, &session_id, &actual_key, modifiers).await?;
     Ok(json!({ "pressed": key }))
 }
@@ -3379,17 +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> {
@@ -3516,6 +3785,56 @@ async fn handle_wait(cmd: &Value, state: &mut DaemonState) -> Result<Value, Stri
 async fn handle_gettext(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();
+
+    // `get text --all-frames` aggregates visible text across every frame the
+    // session can reach — including out-of-process iframes that never show up
+    // in the top document (#27: Yahoo/Rakuten/Mercari listing descriptions).
+    if cmd.get("allFrames").and_then(|v| v.as_bool()) == Some(true) {
+        let frames = super::element::collect_all_frames_text(
+            &mgr.client,
+            &session_id,
+            &state.iframe_sessions,
+        )
+        .await?;
+        let mut combined = String::new();
+        let mut frame_count = 0usize;
+        for f in &frames {
+            let t = f.text.trim();
+            if t.is_empty() {
+                continue;
+            }
+            frame_count += 1;
+            if f.kind != "top" {
+                combined.push_str(&format!("\n\n----- frame [{}] {} -----\n", f.kind, f.url));
+            }
+            combined.push_str(t);
+        }
+        let url = mgr.get_url().await.unwrap_or_default();
+        return Ok(json!({
+            "text": combined,
+            "origin": url,
+            "frames": frame_count,
+            "allFrames": true,
+        }));
+    }
+
+    // `get text --pierce` reads text through CLOSED shadow roots and child
+    // documents via the CDP DOM tree — content `innerText`/`eval` can't see,
+    // e.g. an extension's injected panel in a closed shadow DOM (#30).
+    if cmd.get("pierce").and_then(|v| v.as_bool()) == Some(true) {
+        let text = super::element::get_pierced_text(&mgr.client, &session_id).await?;
+        let url = mgr.get_url().await.unwrap_or_default();
+        return Ok(json!({ "text": text, "origin": url, "pierce": true }));
+    }
+
+    // `get text --main` returns the page's main-content region (readability-lite),
+    // skipping global header/nav/footer/sidebar boilerplate (#27).
+    if cmd.get("main").and_then(|v| v.as_bool()) == Some(true) {
+        let text = super::element::get_main_content_text(&mgr.client, &session_id).await?;
+        let url = mgr.get_url().await.unwrap_or_default();
+        return Ok(json!({ "text": text, "origin": url, "main": true }));
+    }
+
     let selector = cmd
         .get("selector")
         .and_then(|v| v.as_str())
@@ -3533,6 +3852,29 @@ async fn handle_gettext(cmd: &Value, state: &mut DaemonState) -> Result<Value, S
     Ok(json!({ "text": text, "origin": url }))
 }
 
+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 list: Vec<Value> = frames
+        .iter()
+        .enumerate()
+        .map(|(i, f)| {
+            json!({
+                "index": i,
+                "kind": f.kind,
+                "url": f.url,
+                "frameId": f.frame_id,
+                "textLen": f.text.trim().chars().count(),
+            })
+        })
+        .collect();
+    let url = mgr.get_url().await.unwrap_or_default();
+    Ok(json!({ "frames": list, "count": list.len(), "origin": url }))
+}
+
 async fn handle_getattribute(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();
@@ -3967,6 +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();
@@ -4304,8 +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") => {
@@ -4355,8 +4812,12 @@ async fn handle_keyboard(cmd: &Value, state: &DaemonState) -> Result<Value, Stri
 // Phase 5 handlers
 // ---------------------------------------------------------------------------
 
-async fn handle_tab_list(cmd: &Value, state: &DaemonState) -> Result<Value, String> {
-    let mgr = state.browser.as_ref().ok_or("Browser not launched")?;
+async fn handle_tab_list(cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {
+    let mgr = state.browser.as_mut().ok_or("Browser not launched")?;
+    // Re-sync with the live browser so the list reflects tabs opened by other
+    // sessions or re-attached after a cross-process nav, and drops gone ones
+    // (issue #21). Best-effort: a stale list still beats erroring the command.
+    mgr.resync_targets().await.ok();
     let tabs = mgr.tab_list();
     // Echo `full` so the formatter prints untruncated URLs (issue #19).
     if cmd.get("full").and_then(|v| v.as_bool()).unwrap_or(false) {
@@ -4394,13 +4855,50 @@ async fn handle_tab_switch(cmd: &Value, state: &mut DaemonState) -> Result<Value
     let tab_ref_str = cmd
         .get("tabId")
         .and_then(|v| v.as_str())
-        .ok_or("Missing 'tabId' parameter (expected `t<N>` or a label)")?;
-    let tab_ref = super::browser::TabRef::parse(tab_ref_str)?;
-    let tab_id = mgr.resolve_tab_ref(&tab_ref)?;
+        .ok_or("Missing 'tabId' parameter (expected `t<N>`, a label, or a targetId)")?;
+    // Re-sync first so a tab opened by another session, or one that re-attached
+    // after a cross-process nav, is adoptable from here (issue #21).
+    mgr.resync_targets().await.ok();
+    // A CDP `targetId` (shown in `tab list`) is stable across sessions, so accept
+    // it directly for adopting a specific pre-existing tab — falling back to the
+    // per-session `t<N>` / label form.
+    let tab_id = match mgr.tab_id_for_target(tab_ref_str) {
+        Some(id) => id,
+        None => {
+            let tab_ref = super::browser::TabRef::parse(tab_ref_str)?;
+            mgr.resolve_tab_ref(&tab_ref)?
+        }
+    };
     state.ref_map.clear();
     state.iframe_sessions.clear();
     state.active_frame_id = None;
-    let result = mgr.tab_switch_by_id(tab_id).await?;
+    let mut result = mgr.tab_switch_by_id(tab_id).await?;
+
+    // Liveness probe: confirm the new session actually answers before we report
+    // success, so `tab <id>` doesn't print a misleading ✓ for a session that's
+    // stale and will fail on the very next command (issue #29.3). On the churned
+    // -tabId case the ext-0.4.9 targetId recovery (#24) self-heals within ~6s, so
+    // we surface a warning rather than a hard error to avoid a false failure
+    // during that window.
+    if mgr.evaluate("1", None).await.is_err() {
+        if let Some(obj) = result.as_object_mut() {
+            obj.insert(
+                "warning".to_string(),
+                json!("switched tab is not responding yet (session re-attaching); retry the next command"),
+            );
+        }
+    }
+
+    // `--activate`: raise this tab to the foreground (the switch made it active;
+    // bring_to_front acts on the active tab) — for handing a specific tab to the
+    // human (issue #24-C). Best-effort; don't fail the switch if it can't.
+    if cmd
+        .get("activate")
+        .and_then(|v| v.as_bool())
+        .unwrap_or(false)
+    {
+        let _ = mgr.bring_to_front().await;
+    }
 
     if let Some(ref server) = state.stream_server {
         if let Ok(dims) = mgr
@@ -5230,6 +5728,18 @@ async fn handle_bringtofront(state: &DaemonState) -> Result<Value, String> {
     Ok(json!({ "broughtToFront": true }))
 }
 
+async fn handle_current(state: &mut DaemonState) -> Result<Value, String> {
+    let mgr = state.browser.as_mut().ok_or("Browser not launched")?;
+    // Refresh so `current` reflects the live URL/title even after a cross-process
+    // nav (the relay's cached target_info can lag) (#26).
+    mgr.resync_targets().await.ok();
+    let mut info = mgr.active_page_info().ok_or("No active tab")?;
+    if let Some(obj) = info.as_object_mut() {
+        obj.insert("current".to_string(), json!(true));
+    }
+    Ok(info)
+}
+
 async fn handle_timezone(cmd: &Value, state: &DaemonState) -> Result<Value, String> {
     let mgr = state.browser.as_ref().ok_or("Browser not launched")?;
     let timezone = cmd
@@ -6781,6 +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,
@@ -7135,6 +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(
@@ -8731,13 +9264,7 @@ async fn handle_keydown(cmd: &Value, state: &DaemonState) -> Result<Value, Strin
         .and_then(|v| v.as_str())
         .ok_or("Missing 'key' parameter")?;
 
-    mgr.client
-        .send_command(
-            "Input.dispatchKeyEvent",
-            Some(json!({ "type": "keyDown", "key": key })),
-            Some(&session_id),
-        )
-        .await?;
+    interaction::dispatch_single_key(&mgr.client, &session_id, key, "keyDown").await?;
     Ok(json!({ "keydown": key }))
 }
 
@@ -8749,13 +9276,7 @@ async fn handle_keyup(cmd: &Value, state: &DaemonState) -> Result<Value, String>
         .and_then(|v| v.as_str())
         .ok_or("Missing 'key' parameter")?;
 
-    mgr.client
-        .send_command(
-            "Input.dispatchKeyEvent",
-            Some(json!({ "type": "keyUp", "key": key })),
-            Some(&session_id),
-        )
-        .await?;
+    interaction::dispatch_single_key(&mgr.client, &session_id, key, "keyUp").await?;
     Ok(json!({ "keyup": key }))
 }
 
@@ -8873,6 +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/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
@@ -796,16 +805,43 @@ pub(super) fn extract_ax_string(value: &Option<AXValue>) -> String {
 /// Build a JS expression that finds a DOM element by CSS selector or XPath.
 fn build_find_element_js(selector: &str) -> String {
     if let Some(xpath) = selector.strip_prefix("xpath=") {
-        format!(
+        return format!(
             "document.evaluate({}, document, null, XPathResult.FIRST_ORDERED_NODE_TYPE, null).singleNodeValue",
             serde_json::to_string(xpath).unwrap_or_default()
-        )
-    } else {
-        format!(
-            "document.querySelector({})",
-            serde_json::to_string(selector).unwrap_or_default()
-        )
+        );
     }
+    // Bare string (or explicit `text=`): try CSS first, then fall back to
+    // matching an interactive element by its VISIBLE TEXT. snapshot exposes
+    // buttons/links by their name, so `click "購入手続きへ"` should resolve by
+    // that label — previously it was fed straight to `querySelector` as CSS and
+    // failed as an invalid selector even though the button was right there
+    // (issue #24-B). CSS still wins when it matches, so existing selectors are
+    // unaffected; nested/non-ASCII labels now resolve too.
+    let text_only = selector.strip_prefix("text=");
+    let force_text = text_only.is_some();
+    let sel_json = serde_json::to_string(selector).unwrap_or_default();
+    let want_json = serde_json::to_string(text_only.unwrap_or(selector)).unwrap_or_default();
+    format!(
+        r#"(() => {{
+  const sel = {sel};
+  const css = {force_text} ? null : (() => {{ try {{ return document.querySelector(sel); }} catch (_e) {{ return null; }} }})();
+  if (css) return css;
+  const norm = s => (s == null ? '' : String(s)).replace(/\s+/g, ' ').trim();
+  const w = norm({want}); if (!w) return null;
+  const wl = w.toLowerCase();
+  const interactive = Array.from(document.querySelectorAll(
+    'button,a,[role=button],[role=link],[role=menuitem],[role=tab],[role=option],input[type=submit],input[type=button],input[type=reset],summary,label,[onclick]'));
+  const textOf = e => norm(e.innerText || e.textContent) || norm(e.value) ||
+    norm(e.getAttribute && e.getAttribute('aria-label')) || norm(e.getAttribute && e.getAttribute('title'));
+  let hit = interactive.find(e => textOf(e) === w) || interactive.find(e => textOf(e).toLowerCase().includes(wl));
+  if (hit) return hit;
+  const leaves = Array.from(document.querySelectorAll('*')).filter(e => !e.children.length);
+  return leaves.find(e => norm(e.textContent) === w) || leaves.find(e => norm(e.textContent).toLowerCase().includes(wl)) || null;
+}})()"#,
+        sel = sel_json,
+        want = want_json,
+        force_text = force_text
+    )
 }
 
 /// Build a JS expression that counts matching DOM elements by CSS selector or XPath.
@@ -948,6 +984,268 @@ pub async fn get_element_text(
         .unwrap_or_default())
 }
 
+/// Text content collected from a single frame of the page.
+#[derive(Debug, Clone)]
+pub struct FrameText {
+    pub frame_id: String,
+    pub url: String,
+    /// "top" | "inline" (same-process child frame) | "oopif" (out-of-process).
+    pub kind: &'static str,
+    pub text: String,
+}
+
+// The expression we run in every frame to read its visible text. innerText
+// honors CSS visibility (skips display:none), textContent is the fallback.
+const FRAME_INNERTEXT_JS: &str = "(function(){try{var b=document.body||document.documentElement;return b?(b.innerText||b.textContent||''):'';}catch(e){return '';}})()";
+
+async fn eval_text_default(client: &CdpClient, session_id: &str) -> String {
+    let res = client
+        .send_command(
+            "Runtime.evaluate",
+            Some(serde_json::json!({
+                "expression": FRAME_INNERTEXT_JS,
+                "returnByValue": true,
+            })),
+            Some(session_id),
+        )
+        .await;
+    res.ok()
+        .and_then(|v| v.get("result").and_then(|r| r.get("value")).cloned())
+        .and_then(|v| v.as_str().map(|s| s.to_string()))
+        .unwrap_or_default()
+}
+
+// Same-process child frames share the top renderer but live in their own
+// execution context. Page.createIsolatedWorld hands us a context id bound to
+// that frame so Runtime.evaluate reads the child document, not the parent.
+async fn eval_text_in_frame(client: &CdpClient, session_id: &str, frame_id: &str) -> String {
+    let ctx = client
+        .send_command(
+            "Page.createIsolatedWorld",
+            Some(serde_json::json!({ "frameId": frame_id, "worldName": "chrome_use_text" })),
+            Some(session_id),
+        )
+        .await
+        .ok()
+        .and_then(|v| v.get("executionContextId").and_then(|c| c.as_i64()));
+    let Some(ctx_id) = ctx else {
+        return String::new();
+    };
+    let res = client
+        .send_command(
+            "Runtime.evaluate",
+            Some(serde_json::json!({
+                "expression": FRAME_INNERTEXT_JS,
+                "returnByValue": true,
+                "contextId": ctx_id,
+            })),
+            Some(session_id),
+        )
+        .await;
+    res.ok()
+        .and_then(|v| v.get("result").and_then(|r| r.get("value")).cloned())
+        .and_then(|v| v.as_str().map(|s| s.to_string()))
+        .unwrap_or_default()
+}
+
+fn flatten_frame_tree(node: &Value, is_top: bool, out: &mut Vec<(String, String, bool)>) {
+    if let Some(frame) = node.get("frame") {
+        if let Some(id) = frame.get("id").and_then(|v| v.as_str()) {
+            let url = frame
+                .get("url")
+                .and_then(|v| v.as_str())
+                .unwrap_or("")
+                .to_string();
+            out.push((id.to_string(), url, is_top));
+        }
+    }
+    if let Some(children) = node.get("childFrames").and_then(|v| v.as_array()) {
+        for child in children {
+            flatten_frame_tree(child, false, out);
+        }
+    }
+}
+
+/// Collect visible text from every frame reachable in the active session,
+/// including out-of-process iframes (which never appear in the top frame's
+/// `Page.getFrameTree` and so are invisible to `document.body.innerText`).
+///
+/// Same-process child frames are read through `Page.createIsolatedWorld`;
+/// OOPIFs are read through their own auto-attached debugger session
+/// (`iframe_sessions`, keyed by frameId == targetId). This is the engine
+/// behind `get text --all-frames` and `chrome-use frames` — the fix for
+/// listing/marketplace pages whose description lives in a child frame (#27).
+pub async fn collect_all_frames_text(
+    client: &CdpClient,
+    top_session: &str,
+    iframe_sessions: &HashMap<String, String>,
+) -> Result<Vec<FrameText>, String> {
+    let mut out: Vec<FrameText> = Vec::new();
+    let mut seen: std::collections::HashSet<String> = std::collections::HashSet::new();
+
+    // 1. Top session: the top frame plus its same-process descendants. OOPIF
+    //    frames that happen to surface here are skipped — they're read via
+    //    their dedicated session in step 2 (cross-process isolated worlds fail).
+    let tree = client
+        .send_command_no_params("Page.getFrameTree", Some(top_session))
+        .await?;
+    let mut frames: Vec<(String, String, bool)> = Vec::new();
+    flatten_frame_tree(&tree["frameTree"], true, &mut frames);
+    for (fid, url, is_top) in frames {
+        if iframe_sessions.contains_key(&fid) {
+            continue;
+        }
+        if !seen.insert(fid.clone()) {
+            continue;
+        }
+        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,
+            )
+        };
+        out.push(FrameText {
+            frame_id: fid,
+            url,
+            kind,
+            text,
+        });
+    }
+
+    // 2. Each out-of-process iframe, read through its own session.
+    for (fid, sid) in iframe_sessions {
+        if !seen.insert(fid.clone()) {
+            continue;
+        }
+        let url = client
+            .send_command_no_params("Page.getFrameTree", Some(sid))
+            .await
+            .ok()
+            .and_then(|t| {
+                t.get("frameTree")
+                    .and_then(|ft| ft.get("frame"))
+                    .and_then(|f| f.get("url"))
+                    .and_then(|v| v.as_str())
+                    .map(|s| s.to_string())
+            })
+            .unwrap_or_default();
+        let text = eval_text_default(client, sid).await;
+        out.push(FrameText {
+            frame_id: fid.clone(),
+            url,
+            kind: "oopif",
+            text,
+        });
+    }
+
+    Ok(out)
+}
+
+// Readability-lite: prefer the page's semantic main-content region over the
+// whole body so global header/nav/footer chrome (and, on many listing pages,
+// the "related items" sidebar) doesn't drown out the actual content. Runs on
+// the live, rendered tree (innerText needs layout — a detached clone returns
+// empty), so we pick the densest <main>/<article> region rather than cloning
+// and stripping. Falls back to <body> when no substantial main region exists.
+const MAIN_CONTENT_JS: &str = r#"(function(){
+  function txt(el){try{return (el.innerText||'').trim();}catch(e){return '';}}
+  var sels=['main','[role=main]','article','#main','#contents','#l-content'];
+  var best=null,bestLen=0;
+  for(var i=0;i<sels.length;i++){
+    var els=document.querySelectorAll(sels[i]);
+    for(var j=0;j<els.length;j++){var l=txt(els[j]).length;if(l>bestLen){bestLen=l;best=els[j];}}
+  }
+  if(best&&bestLen>200)return txt(best);
+  return txt(document.body);
+})()"#;
+
+/// Extract the page's main-content text (readability-lite), preferring a
+/// semantic `<main>`/`<article>` region over the full body. Used by
+/// `get text --main` to avoid header/nav/sidebar boilerplate (#27).
+pub async fn get_main_content_text(client: &CdpClient, session_id: &str) -> Result<String, String> {
+    let res = client
+        .send_command(
+            "Runtime.evaluate",
+            Some(serde_json::json!({
+                "expression": MAIN_CONTENT_JS,
+                "returnByValue": true,
+            })),
+            Some(session_id),
+        )
+        .await?;
+    Ok(res
+        .get("result")
+        .and_then(|r| r.get("value"))
+        .and_then(|v| v.as_str())
+        .unwrap_or_default()
+        .to_string())
+}
+
+// Text nodes whose parent is one of these carry no visible content.
+fn is_noise_tag(name: &str) -> bool {
+    matches!(name, "SCRIPT" | "STYLE" | "NOSCRIPT" | "TEMPLATE" | "HEAD")
+}
+
+// Walk a CDP DOM.Node tree, collecting text-node values. Unlike `innerText`
+// (JS, blocked by CLOSED shadow roots), the CDP DOM tree from
+// `DOM.getDocument(pierce:true)` includes closed shadow roots and child
+// documents — so this reaches text JS can't. `parent_noise` carries whether an
+// ancestor was <script>/<style>/etc so their text is skipped.
+fn collect_dom_text(node: &Value, parent_noise: bool, out: &mut String) {
+    let node_type = node.get("nodeType").and_then(|v| v.as_i64()).unwrap_or(0);
+    let node_name = node.get("nodeName").and_then(|v| v.as_str()).unwrap_or("");
+    if node_type == 3 {
+        if !parent_noise {
+            if let Some(t) = node.get("nodeValue").and_then(|v| v.as_str()) {
+                let t = t.trim();
+                if !t.is_empty() {
+                    if !out.is_empty() {
+                        out.push(' ');
+                    }
+                    out.push_str(t);
+                }
+            }
+        }
+        return;
+    }
+    let noise = parent_noise || is_noise_tag(node_name);
+    if let Some(children) = node.get("children").and_then(|v| v.as_array()) {
+        for child in children {
+            collect_dom_text(child, noise, out);
+        }
+    }
+    if let Some(shadow) = node.get("shadowRoots").and_then(|v| v.as_array()) {
+        for sr in shadow {
+            collect_dom_text(sr, noise, out);
+        }
+    }
+    if let Some(doc) = node.get("contentDocument") {
+        collect_dom_text(doc, noise, out);
+    }
+}
+
+/// Extract text from the page via the CDP DOM tree with `pierce:true`, which
+/// reaches into CLOSED shadow roots and child documents that `innerText`/`eval`
+/// cannot. Lets an agent read content rendered into a closed shadow DOM (e.g. an
+/// extension's injected debug panel) without any extra Chrome permission — it
+/// rides the per-tab debugger session that's already attached (#30).
+pub async fn get_pierced_text(client: &CdpClient, session_id: &str) -> Result<String, String> {
+    let doc = client
+        .send_command(
+            "DOM.getDocument",
+            Some(serde_json::json!({ "depth": -1, "pierce": true })),
+            Some(session_id),
+        )
+        .await?;
+    let mut out = String::new();
+    if let Some(root) = doc.get("root") {
+        collect_dom_text(root, false, &mut out);
+    }
+    Ok(out)
+}
+
 pub async fn get_element_attribute(
     client: &CdpClient,
     session_id: &str,
@@ -1421,6 +1719,31 @@ mod tests {
         assert_eq!(parse_ref("@e123"), Some("e123".to_string()));
     }
 
+    #[test]
+    fn test_collect_dom_text_pierces_closed_shadow_and_skips_noise() {
+        // A CDP DOM.Node tree: a host element whose CLOSED shadow root holds the
+        // text, plus a <script> whose text must be skipped.
+        let tree = serde_json::json!({
+            "nodeType": 1, "nodeName": "BODY",
+            "children": [
+                { "nodeType": 1, "nodeName": "SCRIPT",
+                  "children": [ { "nodeType": 3, "nodeName": "#text", "nodeValue": "var secret=1;" } ] },
+                { "nodeType": 1, "nodeName": "DIV",
+                  "shadowRoots": [
+                    { "nodeType": 11, "nodeName": "#document-fragment",
+                      "children": [
+                        { "nodeType": 1, "nodeName": "SPAN",
+                          "children": [ { "nodeType": 3, "nodeName": "#text", "nodeValue": "DECRYPTED 42" } ] }
+                      ] }
+                  ] }
+            ]
+        });
+        let mut out = String::new();
+        collect_dom_text(&tree, false, &mut out);
+        assert_eq!(out, "DECRYPTED 42");
+        assert!(!out.contains("secret"), "script text must be skipped");
+    }
+
     #[test]
     fn test_parse_ref_equals_prefix() {
         assert_eq!(parse_ref("ref=e1"), Some("e1".to_string()));
@@ -1452,10 +1775,30 @@ mod tests {
     #[test]
     fn test_build_selector_js_css() {
         let js = build_selector_js("#submit-btn");
-        assert!(js.contains("document.querySelector(\"#submit-btn\")"));
+        // CSS is now tried via a `sel` variable, with a visible-text fallback
+        // appended (issue #24-B). It must still use querySelector (not xpath).
+        assert!(js.contains("const sel = \"#submit-btn\""));
+        assert!(js.contains("document.querySelector(sel)"));
         assert!(!js.contains("document.evaluate"));
     }
 
+    #[test]
+    fn test_build_find_element_js_text_fallback() {
+        // A bare label gets a text-matching fallback so `click "購入手続きへ"`
+        // resolves by visible text, not just CSS (issue #24-B).
+        let js = build_find_element_js("購入手続きへ");
+        assert!(js.contains("購入手続きへ"));
+        assert!(js.contains("interactive")); // the text-match branch
+        assert!(js.contains("textOf"));
+        // `text=` forces the text path (skips CSS).
+        let forced = build_find_element_js("text=Buy now");
+        assert!(forced.contains("true ? null")); // force_text => css skipped
+                                                 // xpath is unchanged.
+        let xp = build_find_element_js("xpath=//button");
+        assert!(xp.contains("document.evaluate"));
+        assert!(!xp.contains("interactive"));
+    }
+
     #[test]
     fn test_build_selector_js_xpath() {
         let js = build_selector_js("xpath=//button[@id='ok']");

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,
@@ -306,32 +589,50 @@ pub async fn fill(
     )
     .await?;
 
-    // Focus the element
-    client
-        .send_command_typed::<_, Value>(
-            "Runtime.callFunctionOn",
-            &CallFunctionOnParams {
-                function_declaration: "function() { this.focus(); }".to_string(),
-                object_id: Some(object_id.clone()),
-                arguments: None,
-                return_by_value: Some(true),
-                await_promise: Some(false),
-            },
-            Some(&effective_session_id),
-        )
-        .await?;
+    // Emulate a real edit so framework-controlled inputs (React/Vue) and
+    // site-side listeners actually see the change (issue #25): the old path set
+    // `this.value` directly and used Input.insertText, which left React's
+    // internal value-tracker out of sync and never fired change/blur — so
+    // dependent logic (e.g. Mercari's postal-code → 都道府県 autocomplete) never
+    // ran even though the value was visible. Set the value through the element's
+    // PROTOTYPE setter (which React's _valueTracker hooks), then dispatch
+    // input → change → blur/focusout. `type <sel> <text>` remains for sites that
+    // need per-keystroke events.
+    let fill_js = format!(
+        r#"function() {{
+            const el = this;
+            const v = {val};
+            try {{ el.focus(); }} catch (e) {{}}
+            const tag = el.tagName;
+            const fire = (type, ctor) => el.dispatchEvent(new (ctor || Event)(type, {{ bubbles: true }}));
+            if (tag === 'SELECT') {{
+                el.value = v; fire('input'); fire('change'); return true;
+            }}
+            if (el.isContentEditable) {{
+                el.textContent = v; fire('input', window.InputEvent || Event); fire('change');
+                try {{ el.blur(); }} catch (e) {{}} fire('focusout'); return true;
+            }}
+            const proto = tag === 'TEXTAREA' ? window.HTMLTextAreaElement.prototype
+                                             : window.HTMLInputElement.prototype;
+            const desc = Object.getOwnPropertyDescriptor(proto, 'value');
+            const set = desc && desc.set ? (x) => desc.set.call(el, x) : (x) => {{ el.value = x; }};
+            set('');                                   // reset the framework tracker
+            fire('input', window.InputEvent || Event);
+            set(v);                                    // native setter → React/Vue registers
+            fire('input', window.InputEvent || Event);
+            fire('change');
+            try {{ el.blur(); }} catch (e) {{}}
+            fire('focusout');                          // blur-triggered lookups/validation
+            return true;
+        }}"#,
+        val = serde_json::to_string(value).unwrap_or_default()
+    );
 
-    // Select all + delete to clear
     client
         .send_command_typed::<_, Value>(
             "Runtime.callFunctionOn",
             &CallFunctionOnParams {
-                function_declaration: r#"function() {
-                    this.select && this.select();
-                    this.value = '';
-                    this.dispatchEvent(new Event('input', { bubbles: true }));
-                }"#
-                .to_string(),
+                function_declaration: fill_js,
                 object_id: Some(object_id),
                 arguments: None,
                 return_by_value: Some(true),
@@ -341,17 +642,6 @@ pub async fn fill(
         )
         .await?;
 
-    // Insert text (keyboard input dispatched at page level, use parent session_id)
-    client
-        .send_command_typed::<_, Value>(
-            "Input.insertText",
-            &InsertTextParams {
-                text: value.to_string(),
-            },
-            Some(session_id),
-        )
-        .await?;
-
     Ok(())
 }
 
@@ -365,6 +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,
@@ -411,7 +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(
@@ -419,6 +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
@@ -468,6 +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
@@ -557,6 +889,48 @@ pub async fn press_key_with_modifiers(
     Ok(())
 }
 
+/// Dispatch a SINGLE key event (`keyDown` or `keyUp`) carrying the full key
+/// descriptor — `key`, `code`, `windowsVirtualKeyCode`/`nativeVirtualKeyCode`,
+/// and (on key-down) printable `text`. Powers the `keydown`/`keyup` commands.
+///
+/// The previous implementation sent only `{key}`, so games and shortcut handlers
+/// that read `event.code` (e.g. `"KeyD"`, `"ArrowRight"`) or `event.keyCode` saw
+/// nothing — a held key set no movement flag and did nothing (dogfood: holding a
+/// direction in a canvas platformer barely nudged the player). Sending the same
+/// descriptor `press` uses makes hold-to-move work regardless of which field the
+/// page keys off.
+pub async fn dispatch_single_key(
+    client: &CdpClient,
+    session_id: &str,
+    key: &str,
+    event_type: &str,
+) -> Result<(), String> {
+    let (key_name, code, key_code) = named_key_info(key);
+    // Printable text is only meaningful on key-down; key-up never inserts.
+    let text = if event_type == "keyDown" {
+        key_text(&key_name)
+    } else {
+        None
+    };
+    client
+        .send_command_typed::<_, Value>(
+            "Input.dispatchKeyEvent",
+            &DispatchKeyEventParams {
+                event_type: event_type.to_string(),
+                key: Some(key_name),
+                code: Some(code),
+                text: text.clone(),
+                unmodified_text: text,
+                windows_virtual_key_code: Some(key_code),
+                native_virtual_key_code: Some(key_code),
+                modifiers: None,
+            },
+            Some(session_id),
+        )
+        .await?;
+    Ok(())
+}
+
 pub async fn scroll(
     client: &CdpClient,
     session_id: &str,

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

@@ -186,6 +186,103 @@ pub fn print_response_with_opts(resp: &Response, action: Option<&str>, opts: &Ou
     }
 
     if let Some(data) = &resp.data {
+        // A click that opened a new tab: surface it so the agent doesn't read the
+        // unchanged old page as a failed click (issue #24-A).
+        if let Some(opened) = data.get("openedTab") {
+            let tid = opened.get("tabId").and_then(|v| v.as_str()).unwrap_or("?");
+            let url = opened.get("url").and_then(|v| v.as_str()).unwrap_or("");
+            let followed = data
+                .get("followed")
+                .and_then(|v| v.as_bool())
+                .unwrap_or(false);
+            let verb = if followed {
+                "switched to new tab"
+            } else {
+                "opened new tab"
+            };
+            eprintln!(
+                "{} {} [{}] {}",
+                color::cyan("→"),
+                verb,
+                tid,
+                color::dim(url)
+            );
+        }
+
+        // `current`: the active tab's stable handle (#26).
+        if data
+            .get("current")
+            .and_then(|v| v.as_bool())
+            .unwrap_or(false)
+        {
+            let tid = data.get("tabId").and_then(|v| v.as_str()).unwrap_or("?");
+            let title = data.get("title").and_then(|v| v.as_str()).unwrap_or("");
+            let url = data.get("url").and_then(|v| v.as_str()).unwrap_or("");
+            let target = data.get("targetId").and_then(|v| v.as_str()).unwrap_or("");
+            println!("{} [{}] {} - {}", color::cyan("→"), tid, title, url);
+            println!("      {}", color::dim(&format!("target: {}", target)));
+            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()) {
@@ -297,6 +394,43 @@ pub fn print_response_with_opts(resp: &Response, action: Option<&str>, opts: &Ou
         // Snapshot
         if let Some(snapshot) = data.get("snapshot").and_then(|v| v.as_str()) {
             print_with_boundaries(snapshot, origin, opts);
+            // Canvas-app hint: the tree was near-empty but the page paints to a
+            // <canvas>, so refs are a dead end — point at the screenshot path.
+            if let Some(note) = data.get("note").and_then(|v| v.as_str()) {
+                eprintln!("{}", color::dim(note));
+            }
+            return;
+        }
+        // Frame list (`chrome-use frames`)
+        if action == Some("frames") {
+            if let Some(list) = data.get("frames").and_then(|v| v.as_array()) {
+                let count = list.len();
+                println!(
+                    "{}",
+                    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);
+                    let kind = f.get("kind").and_then(|v| v.as_str()).unwrap_or("?");
+                    let url = f.get("url").and_then(|v| v.as_str()).unwrap_or("");
+                    let len = f.get("textLen").and_then(|v| v.as_i64()).unwrap_or(0);
+                    println!(
+                        "  [{}] {:<6} {} chars  {}",
+                        idx,
+                        kind,
+                        len,
+                        color::dim(if url.is_empty() { "(about:blank)" } else { url })
+                    );
+                }
+                eprintln!(
+                    "{}",
+                    color::dim("read everything with: chrome-use get text --all-frames")
+                );
+            }
             return;
         }
         // Title
@@ -512,25 +646,35 @@ pub fn print_response_with_opts(resp: &Response, action: Option<&str>, opts: &Ou
                 } else {
                     println!("{} [{}] {} - {}", marker, tab_id, title, url);
                 }
+                // `--full` also surfaces the stable cross-session CDP targetId so
+                // a stranded tab can be adopted from another session via
+                // `tab <targetId>` (issue #21).
+                if full {
+                    if let Some(target_id) = tab.get("targetId").and_then(|v| v.as_str()) {
+                        println!("      {}", color::dim(&format!("target: {}", target_id)));
+                    }
+                }
             }
             return;
         }
         // Tab switch
         if action == Some("tab_switch") {
             if let Some(tab_id) = data.get("tabId").and_then(|v| v.as_str()) {
-                if let Some(url) = data.get("url").and_then(|v| v.as_str()) {
-                    println!(
-                        "{} Switched to tab [{}] ({})",
-                        color::success_indicator(),
-                        tab_id,
-                        url
-                    );
+                let warning = data.get("warning").and_then(|v| v.as_str());
+                // A non-responding session isn't a real success — show a warning
+                // indicator instead of the green ✓ (issue #29.3).
+                let indicator = if warning.is_some() {
+                    color::warning_indicator()
                 } else {
-                    println!(
-                        "{} Switched to tab [{}]",
-                        color::success_indicator(),
-                        tab_id
-                    );
+                    color::success_indicator()
+                };
+                if let Some(url) = data.get("url").and_then(|v| v.as_str()) {
+                    println!("{} Switched to tab [{}] ({})", indicator, tab_id, url);
+                } else {
+                    println!("{} Switched to tab [{}]", indicator, tab_id);
+                }
+                if let Some(w) = warning {
+                    eprintln!("{}", color::dim(w));
                 }
                 return;
             }
@@ -1349,6 +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
@@ -1356,6 +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.)
@@ -1620,12 +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
@@ -1637,6 +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" => {
@@ -1717,6 +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.
@@ -1737,6 +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
@@ -1878,7 +2046,9 @@ Usage: chrome-use get <subcommand> [args]
 Retrieves various types of information from elements or the page.
 
 Subcommands:
-  text <selector>            Get text content of element
+  text [selector]            Element text; no selector = WHOLE PAGE, all frames
+  text --main                Main-content text only (skip nav/header/sidebar)
+  text --pierce              Read through CLOSED shadow DOM (injected panels)
   html <selector>            Get inner HTML of element
   value <selector>           Get value of input element
   attr <selector> <name>     Get attribute value
@@ -1894,7 +2064,10 @@ Global Options:
   --session <name>     Use specific session
 
 Examples:
-  chrome-use get text @e1
+  chrome-use get text                  # whole page across ALL frames (default)
+  chrome-use get text @e1              # one element
+  chrome-use get text --main           # main content, no nav/sidebar boilerplate
+  chrome-use frames                    # list frames + where the text lives
   chrome-use get html "#content"
   chrome-use get value "#email-input"
   chrome-use get attr "#link" href
@@ -2765,6 +2938,20 @@ Notes:
   - Streaming is always enabled. Set AGENT_BROWSER_STREAM_PORT to bind to a
     specific port instead of the default OS-assigned port.
 
+The WS is BIDIRECTIONAL — the high-throughput way to drive a live/real-time page
+(games, canvas apps) instead of one screenshot + one CLI call per action:
+  - Server -> client (JSON text frames):
+      {"type":"frame","data":"<base64 jpeg>"}   live screencast (~60fps)
+      plus status / tabs messages.
+  - Client -> server (send JSON text):
+      {"type":"input_keyboard","eventType":"keyDown|keyUp","key":" ","code":"Space",
+       "windowsVirtualKeyCode":32}
+      {"type":"input_mouse","eventType":"mousePressed|mouseReleased|mouseMoved",
+       "x":640,"y":360,"button":"left","clickCount":1}
+      {"type":"input_touch","eventType":"touchStart|touchEnd","touchPoints":[...]}
+  Connect once and run a tight local loop: read frames, send timed input — no
+  per-action process spawn, no round-trip. Works over the extension relay too.
+
 Global Options:
   --json               Output as JSON
   --session <name>     Use specific session
@@ -3057,7 +3244,12 @@ Core Commands:
   dblclick <sel>             Double-click element
   type <sel> <text>          Type into element
   fill <sel> <text>          Clear and fill
-  press <key>                Press key (Enter, Tab, Control+a)
+  press <key> [--hold <ms>]  Press key (Enter, Tab, Control+a). --hold keeps it
+                             down <ms> then releases — precise (in-daemon), for
+                             games/charge: `press d --hold 800`
+  keydown <key>              Hold a key down (no auto-release) — for games/shortcuts
+  keyup <key>                Release a held key. Pair with keydown to hold-to-move:
+                             `keydown d` … `keyup d`
   keyboard type <text>       Type text with real keystrokes (no selector)
   keyboard inserttext <text> Insert text without key events
   hover <sel>                Hover element
@@ -3085,10 +3277,15 @@ Navigation:
 
 Get Info:  chrome-use get <what> [selector]
   text, html, value, attr <name>, title, url, count, box, styles, cdp-url
+  text (no selector = whole page, all frames), text --main, frames (list)
 
 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
 
@@ -3116,7 +3313,12 @@ Storage:
   storage <local|session>    Manage web storage
 
 Tabs:
-  tab [new|list|close|<n>]   Manage tabs
+  tab [new|list|close|<ref>] Manage tabs (<ref> = t<N>, a label, or a CDP targetId)
+  tab list --full            Full URLs + stable cross-session targetId per tab
+  tab <targetId>             Adopt a specific tab (incl. another session's) by its
+                             stable targetId, no reload — preserves in-page state
+  open <url> --reuse-tab     Reuse an existing tab on that URL instead of spawning
+                             a duplicate (matches origin+path; preserves state)
 
 Diff:
   diff snapshot              Compare current vs last snapshot
@@ -3178,6 +3380,10 @@ Confirmation:
 Sessions:
   session                    Show current session name
   session list               List active sessions
+  sessions                   List running session daemons (alias of daemon status)
+  daemon status              List running session daemons (+ relay state)
+  daemon restart             Kill all session daemons; keeps the extension relay
+                             up. Clears stale/cross-leaked state after an upgrade.
 
 Chat (AI):
   chat <message>             Send a natural language instruction (single-shot)
@@ -3371,6 +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
 "#
     );
 }
@@ -3453,6 +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)]

cli/src/upgrade.rs

@@ -52,6 +52,27 @@ fn is_newer(latest: &str, current: &str) -> bool {
     matches!((parse_version(latest), parse_version(current)), (Some(l), Some(c)) if l > c)
 }
 
+/// Public semver-ish comparison (`latest` strictly newer than `current`), so
+/// `doctor` can flag a stale extension/CLI without re-implementing parsing.
+pub fn version_is_newer(latest: &str, current: &str) -> bool {
+    is_newer(latest, current)
+}
+
+/// The latest CLI version recorded by the background update check, if any.
+/// `doctor` uses it to show "a newer chrome-use is available" without a network
+/// call (the `__update-check` worker refreshes the cache out of band).
+pub fn cached_latest_version() -> Option<String> {
+    std::fs::read_to_string(update_cache_path())
+        .ok()
+        .and_then(|s| serde_json::from_str::<serde_json::Value>(&s).ok())
+        .and_then(|j| {
+            j.get("latest")
+                .and_then(|v| v.as_str())
+                .map(|s| s.to_string())
+        })
+        .filter(|s| !s.is_empty())
+}
+
 /// Hidden `__update-check` subcommand: fetch the latest release tag and cache it.
 /// Spawned detached by [`maybe_notify_update`] so the network call never blocks a
 /// real command. Uses `curl` (no extra deps, matches `upgrade`).
@@ -100,7 +121,7 @@ pub fn maybe_notify_update() {
     if first.starts_with("__")
         || matches!(
             first.as_str(),
-            "upgrade" | "install" | "doctor" | "dashboard"
+            "upgrade" | "install" | "doctor" | "dashboard" | "daemon"
         )
     {
         return;

extensions/ab-connect/background.js

@@ -30,6 +30,16 @@ const tabs = new Map()
 const sessionToTab = new Map()
 /** child (OOPIF/worker) sessionId -> tabId */
 const childSessionToTab = new Map()
+/** sessionId -> CDP targetId, kept ACROSS detach so a dead `cb-tab-<oldTabId>`
+ * session can be recovered by its stable targetId when the cross-process nav
+ * gave the tab a new Chrome tabId (issue #24). Capped to bound memory. */
+const sessionTargets = new Map()
+function rememberSessionTarget(sessionId, targetId) {
+  if (!sessionId || !targetId) return
+  sessionTargets.delete(sessionId)
+  sessionTargets.set(sessionId, targetId)
+  if (sessionTargets.size > 256) sessionTargets.delete(sessionTargets.keys().next().value)
+}
 /** tab-group name -> chrome tabGroups id (best-effort cache) */
 const groupIdByName = new Map()
 
@@ -108,6 +118,12 @@ function connectHost() {
     // reconnect. Keep chrome.debugger attached so reconnect is cheap.
     for (const tabId of tabs.keys()) setBadge(tabId, 'connecting')
   })
+  // Report our version so the host can tell the CLI/`doctor` which extension
+  // build is live (otherwise the extension version is a black box — the user
+  // can't tell they're on an old one). Best-effort; ignored by older hosts.
+  try {
+    postToHost({ method: 'hello', version: chrome.runtime.getManifest().version })
+  } catch {}
   // Tell the daemon about everything we already have attached, then attach
   // anything new.
   reannounceAttachedTabs()
@@ -149,6 +165,90 @@ function tabForTarget(targetId) {
   return null
 }
 
+// The STABLE Chrome tabId encoded in a `cb-tab-<tabId>` session id (#17), or
+// null for any other session shape (child/iframe sessions). The tabId is the
+// real source of truth: it survives the renderer-process swaps (cross-origin
+// OAuth/SSO navs) that tear down the page's CDP target — which is why binding to
+// it (like claude-in-chrome) rides through the hop that killed the old
+// target/sessionId binding (issue #23).
+function tabIdFromSession(sessionId) {
+  const m = /^cb-tab-(\d+)$/.exec(sessionId || '')
+  return m ? Number(m[1]) : null
+}
+
+// Ensure the debugger is attached to a `cb-tab-<tabId>` session's tab, re-attaching
+// across the transient window of a process swap (with a couple of short retries).
+// Returns the tabId on success, or null when the tab is genuinely gone
+// (closed / restricted). (issues #20.1, #23)
+async function recoverSessionTab(sessionId) {
+  const tabId = tabIdFromSession(sessionId)
+  // 1) Fast path: the encoded Chrome tabId still exists — re-attach it (covers
+  //    the common renderer-process swap where the tabId is preserved, #23).
+  if (tabId != null) {
+    for (let i = 0; i < 3; i++) {
+      const tab = await chrome.tabs.get(tabId).catch(() => null)
+      if (!eligible(tab)) break // tabId is gone — fall through to targetId recovery
+      try {
+        await attachTab(tabId)
+        if (tabs.has(tabId)) return tabId
+      } catch {
+        // mid-swap: tab exists but isn't attachable yet — back off and retry.
+      }
+      await new Promise((r) => setTimeout(r, 120 + i * 150))
+    }
+  }
+  // 2) The Chrome tabId is gone, but the CDP targetId is STABLE across the nav.
+  //    Some cross-process hops (Mercari's signin token exchange) give the tab a
+  //    NEW tabId while keeping the same target, so `cb-tab-<oldTabId>` can't be
+  //    recovered by tabId. Find the tab now hosting our remembered targetId via
+  //    chrome.debugger.getTargets(), attach it, and ALIAS the dead session to it
+  //    so the daemon's session id keeps resolving. Longer window: this hop can
+  //    take several seconds to settle (issue #24).
+  const targetId = sessionTargets.get(sessionId)
+  if (targetId) {
+    for (let i = 0; i < 6; i++) {
+      const targets = await chrome.debugger.getTargets().catch(() => null)
+      const t = targets && targets.find((x) => x.id === targetId && x.tabId != null)
+      if (t && t.tabId != null) {
+        const tab = await chrome.tabs.get(t.tabId).catch(() => null)
+        if (eligible(tab)) {
+          try {
+            await attachTab(t.tabId)
+            if (tabs.has(t.tabId)) {
+              sessionToTab.set(sessionId, t.tabId) // alias dead session -> live tab
+              return t.tabId
+            }
+          } catch {
+            // not attachable yet — keep waiting for the swap to settle.
+          }
+        }
+      }
+      await new Promise((r) => setTimeout(r, 300 + i * 300))
+    }
+  }
+  return null
+}
+
+// Send a CDP command to a tab, riding a debugger detach that can happen between
+// our attach check and the command itself (a renderer-process swap mid-flight).
+// On a detached-style failure, drop the stale handle, re-attach the stable tab,
+// and retry once — so a cross-process nav never surfaces as a hard error (#23).
+async function sendCdpToTab(tabId, method, params) {
+  const dbg = { tabId }
+  try {
+    return await chrome.debugger.sendCommand(dbg, method, params)
+  } catch (e) {
+    const msg = String((e && e.message) || e)
+    if (!/detached|not attached|target.*(closed|gone)|no target|cannot access|frame.*detached/i.test(msg)) {
+      throw e
+    }
+    detachTab(tabId, false)
+    const ok = await recoverSessionTab(`cb-tab-${tabId}`)
+    if (!ok) throw e
+    return await chrome.debugger.sendCommand(dbg, method, params)
+  }
+}
+
 function anyConnectedTab() {
   const it = tabs.keys().next()
   return it.done ? null : it.value
@@ -207,13 +307,26 @@ async function handleForwardCdpCommand(msg) {
   // Fail loudly instead so the agent sees an actionable error, not bad data.
   let tabId
   if (sessionId) {
-    tabId = tabForSession(sessionId)
-    if (!tabId) {
-      throw new Error(
-        `stale sessionId ${sessionId} for ${method}: its tab is gone (closed, ` +
-          `navigated across processes, or lost after an extension restart). ` +
-          `Re-attach by re-opening your target URL before retrying.`,
-      )
+    // The stable Chrome tabId encoded in `cb-tab-<tabId>` is the source of truth
+    // (it survives renderer-process swaps; the CDP target/sessionId does not).
+    // Resolve via it primarily — don't depend on a session→tab map entry that the
+    // detach handler may have cleared — and ensure the debugger is attached,
+    // re-attaching across a cross-process nav before failing (issues #20.1, #23).
+    // `tabForSession` still covers child/iframe sessions that aren't `cb-tab-*`.
+    tabId = tabIdFromSession(sessionId) ?? tabForSession(sessionId)
+    if (tabId == null) {
+      throw new Error(`unknown sessionId ${sessionId} for ${method}`)
+    }
+    if (!tabs.has(tabId)) {
+      const recovered = await recoverSessionTab(sessionId)
+      if (!recovered) {
+        throw new Error(
+          `stale sessionId ${sessionId} for ${method}: its tab is gone (closed, ` +
+            `navigated across processes, or lost after an extension restart). ` +
+            `Re-attach by re-opening your target URL before retrying.`,
+        )
+      }
+      tabId = recovered
     }
   } else if (typeof params?.targetId === 'string') {
     tabId = tabForTarget(params.targetId)
@@ -223,18 +336,17 @@ async function handleForwardCdpCommand(msg) {
     // applies to any attached tab.
     tabId = anyConnectedTab()
   }
-  if (!tabId) throw new Error(`no attached tab for ${method}`)
-  const dbg = { tabId }
+  if (tabId == null) throw new Error(`no attached tab for ${method}`)
 
   // Re-enabling Runtime can leave a stale state; bounce it (matches upstream).
   if (method === 'Runtime.enable') {
     try {
-      await chrome.debugger.sendCommand(dbg, 'Runtime.disable')
+      await sendCdpToTab(tabId, 'Runtime.disable', undefined)
       await new Promise((r) => setTimeout(r, 30))
     } catch {}
-    return await chrome.debugger.sendCommand(dbg, 'Runtime.enable', params)
+    return await sendCdpToTab(tabId, 'Runtime.enable', params)
   }
-  return await chrome.debugger.sendCommand(dbg, method, params)
+  return await sendCdpToTab(tabId, method, params)
 }
 
 // ---- attach / detach ------------------------------------------------------
@@ -273,6 +385,7 @@ async function attachTab(tabId) {
   const entry = { sessionId, targetId }
   tabs.set(tabId, entry)
   sessionToTab.set(sessionId, tabId)
+  rememberSessionTarget(sessionId, targetId)
   setBadge(tabId, port ? 'on' : 'connecting')
   postToHost({
     method: 'forwardCDPEvent',
@@ -355,9 +468,33 @@ chrome.debugger.onEvent.addListener((source, method, params) =>
   }),
 )
 
-chrome.debugger.onDetach.addListener((source) =>
-  void whenReady(() => {
-    if (source.tabId) detachTab(source.tabId, true)
+chrome.debugger.onDetach.addListener((source, reason) =>
+  void whenReady(async () => {
+    const tabId = source.tabId
+    if (!tabId) return
+    detachTab(tabId, true)
+    // A cross-process navigation (e.g. an SSO redirect like
+    // login.account.rakuten.com that swaps the render process / spawns OOPIFs)
+    // detaches the debugger, but the TAB survives. Without re-attaching, the
+    // session goes permanently stale and even open/navigate fails — exactly the
+    // #19 follow-up. So proactively re-attach (the stable `cb-tab-<tabId>`
+    // session id then restores the daemon's binding). Don't fight a detach the
+    // user or DevTools initiated.
+    if (reason === 'canceled_by_user' || reason === 'replaced_with_devtools') return
+    if (!port) return
+    // The swapped-in process needs a moment to settle; retry with backoff.
+    for (let i = 0; i < 6; i++) {
+      await new Promise((r) => setTimeout(r, 250 + i * 200))
+      if (tabs.has(tabId)) return // already re-attached (e.g. via onUpdated)
+      const tab = await chrome.tabs.get(tabId).catch(() => null)
+      if (!tab || !eligible(tab)) return // tab gone or now a restricted page
+      try {
+        await attachTab(tabId)
+        return
+      } catch (e) {
+        console.warn(`ab-connect: reattach attempt ${i + 1} for tab ${tabId} failed:`, e)
+      }
+    }
   }),
 )
 

extensions/ab-connect/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "chrome-use",
-  "version": "0.4.4",
+  "version": "0.4.9",
   "description": "Let chrome-use drive your logged-in Chrome \u2014 install once, no token, no per-use confirmation.",
   "key": "MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEA6vQIyscGIPYPZdSpPwPL0+0gxUROyRgCpmvCSDoc8XUm4qm97VbKnD9Ijc1lV22lNWZtE78gaRjt6BeSfuMgnBymnhLKjN1gU6AI5QUU0mrJyeHdWKvrKQR5FmsM2A7Xr1ykE2SiiS8zNUS3Y/6O5l+Nva7wrVy6E4a2dkBVQkOsu+DV+nEZvhIyuDY5D5SPXqNwUTWTaglwj5mjvHz36xSwCWlPmrtJ+ED0AUyrb2z4GIOmvk4kqtBVrh/UD058klLo4CkYOnIybB5aV6WYuwarfPY4bF/dLggPem+ewLNTUNBuwrxj/A4nUv0LJTuRO8rR7f8WR9qnRCY0Ic5saQIDAQAB",
   "icons": {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "chrome-use",
-  "version": "1.2.3",
+  "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:
@@ -207,7 +253,11 @@ assigned fresh on every snapshot.
 For unstructured reading (no refs needed):
 
 ```bash
-chrome-use get text @e1                # visible text of an element
+chrome-use get text                    # WHOLE PAGE — all frames by default (see below)
+chrome-use get text @e1                # visible text of one element (or a CSS selector)
+chrome-use get text --main             # main content only — skip nav/header/sidebar
+chrome-use get text --pierce           # read through CLOSED shadow DOM (injected panels)
+chrome-use frames                      # list every frame + where the text lives
 chrome-use get html @e1                # innerHTML
 chrome-use get attr @e1 href           # any attribute
 chrome-use get value @e1               # input value
@@ -216,6 +266,27 @@ chrome-use get url                     # current URL
 chrome-use get count ".item"           # count matching elements
 ```
 
+**Whole-page text is cross-frame by default.** `chrome-use get text` with no
+selector aggregates visible text across **every** frame — top document plus
+same-process child frames plus cross-origin iframes — so you never silently miss
+content that lives in an iframe (Yahoo Auctions / Rakuten / Mercari shop
+descriptions, embedded checkout/spec frames). Each child frame is delimited with
+a `----- frame [kind] url -----` marker. You do **not** need to remember a flag —
+the default already reads all frames. (`--all-frames` is still accepted as an
+explicit alias.)
+
+So: when text looks missing or wrong, you don't have to guess — just
+`chrome-use get text` reads everything. To **see** the structure (which frame
+holds what), run `chrome-use frames`. To **cut boilerplate** (global nav/header/
+footer, "related items" sidebars), use `chrome-use get text --main`. If content
+is lazy-loaded, `scroll` it into view first, then read.
+
+**Closed shadow DOM.** Some injected UI (browser-extension debug panels, web
+components) renders into a *closed* shadow root that `eval`/`innerText` cannot
+read. `chrome-use get text --pierce` reads through closed shadow roots and child
+documents via the CDP DOM tree — use it when content is clearly on screen (you
+see it in a screenshot) but `get text`/`eval` come back empty.
+
 ## Interacting
 
 ```bash
@@ -226,8 +297,15 @@ 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 press Enter                 # press a key at current focus
+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)
+chrome-use keyup d                     # release it — pair them to hold-to-move
+                                          # in a game: `keydown d; sleep; keyup d`
 chrome-use check @e3                   # check checkbox
 chrome-use uncheck @e3                 # uncheck
 chrome-use select @e4 "option-value"   # native <select> only
@@ -240,16 +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:
@@ -296,6 +390,54 @@ chrome-use click --coords 449,320   # same, explicit flag
 
 A bare-number argument is always a coordinate, never a selector.
 
+### Canvas / WebGL apps (games, map & 3D viewers, drawing tools)
+
+These paint everything to a `<canvas>` and expose **almost no accessibility
+tree**, so `snapshot` comes back near-empty and refs are a dead end. `snapshot`
+detects this and prints a one-line hint. Drive them the screenshot way:
+
+```bash
+chrome-use screenshot /tmp/s.png       # SEE the state (your only read path —
+                                          # eval/get text return nothing useful)
+chrome-use click 640 360               # interact by viewport coordinate
+chrome-use press d --hold 800          # hold-to-move, precise (timed in-daemon —
+                                          # NOT keydown+shell-sleep+keyup, which
+                                          # adds ~250ms jitter per round-trip)
+chrome-use press Space                 # discrete actions (jump/attack/confirm)
+```
+
+**Don't drive frame-by-frame with one CLI call per action** — that's the slowest,
+lowest-fidelity way (each call is a process spawn + round-trip). Script a *timed
+sequence in a single round-trip* with `batch` (it sends each step to the running
+daemon; `press --hold` and `wait` block in-daemon, so timing is precise):
+
+```bash
+chrome-use batch "press d --hold 900" "press j" "press j" "wait 200" "press d --hold 500"
+```
+
+Also try reading real state instead of pixels: `eval` runs in the page's main
+world, so for a framework/engine game you can often reach its globals (e.g. a
+Phaser/PIXI/Three instance, a store, `window.__GAME__`) and read positions/score
+directly — far better than guessing from a screenshot.
+
+**For genuinely real-time driving, drop the CLI entirely and use the WebSocket.**
+`chrome-use stream enable` opens a bidirectional WS (`stream status` prints the
+`ws://127.0.0.1:<port>`). Connect once and you get a live ~60fps screencast AND
+can send input on the same socket — no per-action process spawn, no round-trip,
+works over the extension relay:
+
+```js
+// node (global WebSocket): live frames + locally-timed input
+const ws = new WebSocket("ws://127.0.0.1:PORT")
+ws.onmessage = e => { const m = JSON.parse(e.data); if (m.type==="frame") {/* base64 jpeg */} }
+const k = (eventType,key,code,vk) => ws.send(JSON.stringify({type:"input_keyboard",eventType,key,code,windowsVirtualKeyCode:vk}))
+k("keyDown"," ","Space",32); setTimeout(()=>k("keyUp"," ","Space",32), 80)   // a jump
+// also: {type:"input_mouse",eventType:"mousePressed",x,y,button:"left",clickCount:1}
+```
+
+This is the difference between watching a slideshow and playing the game. Reserve
+screenshots for one-off checks; use the WS for any sustained real-time control.
+
 ## Waiting (read this)
 
 Agents fail more often from bad waits than from bad selectors. Pick the
@@ -501,6 +643,43 @@ the same browser's existing targets, so a second session's first `open` can
 navigate a sibling's tab. For concurrent agents on one real Chrome, use the
 extension (each with a distinct `--session`), not raw `--cdp`.
 
+Each session owns its own tab group and assigns its own `t<N>` indices (the same
+physical tab is `t8` in one session, `t1` in another), so `t<N>` is **not** a
+stable cross-session handle. To reach a *specific* tab from another session — e.g.
+a tab that was filled in a session whose handle later died — use the **stable CDP
+`targetId`**:
+
+```bash
+chrome-use tab list --full --session B   # re-syncs live tabs; prints `target: <id>` per row
+chrome-use tab <targetId> --session B    # adopt that exact tab, NO reload (state preserved)
+```
+
+`tab list` re-discovers the live tab set on every call, so a fresh session sees
+tabs other sessions opened (and re-attached ones), not just its own. Adopting by
+`targetId` lands session B on the stranded tab without reloading it, so a
+half-filled form survives. Still, the simplest recovery for a session whose own
+tab died is to recover *that* session (reload / re-`open` / `daemon restart`).
+
+To avoid piling up duplicate tabs when you re-`open` the same entry URL on
+rebind, pass **`--reuse-tab`**: if a tab already shows that URL (matched by
+origin+path), it switches to it instead of spawning a new one.
+
+### Reset stuck daemon state
+
+Each session runs a background daemon worker that holds the page handles. If a
+session starts misbehaving — commands hit the wrong tab, refs/handles look stale,
+or you upgraded `chrome-use` mid-session and old workers linger — restart the
+daemons instead of hunting PIDs with `pgrep`/`kill`:
+
+```bash
+chrome-use daemon status     # list running session daemons (+ relay state)
+chrome-use daemon restart    # kill every session daemon worker
+```
+
+`daemon restart` leaves the extension's native-messaging bridge (`__nm-host`)
+alone, so the relay to your live Chrome stays up — the next command just spins up
+a fresh, clean daemon against the same browser. It does **not** close any tabs.
+
 ### Mock network requests
 
 ```bash