feat: rich-editor fill, box centers, screenshot downscale, disabled+docs (#41-#45)

Dogfooding backlog from this session's embedded-form/editor work.

#41 fill on rich editors: detect CodeMirror 5 / Monaco / ProseMirror /
contenteditable and set via their own API or execCommand('insertText') so
beforeinput/input fire (a raw .value/textContent write no-op'd juejin's
CodeMirror and skipped React composers). Response echoes the `engine` used.
`fill <sel> --file <path>` / `--stdin` set large multiline text without
shell-escaping. `get value` now reads CodeMirror/Monaco/contenteditable too.

#42 screenshot --max-width/--max-height/--scale, plus a default 2000px
longest-edge cap (AGENT_BROWSER_SCREENSHOT_MAX_EDGE; 0 disables) so retina
full-page shots fit an agent's image reader and --scale 0.5 makes screenshot
px line up with click px. Annotated shots are never downscaled.

#43 `box @ref` (already a top-level alias of `get box`) now also returns
centerX/centerY/inViewport in CSS px — feed straight into `click x y` when a
ref-click no-ops (e.g. a button in a cross-origin iframe).

#44 no code change needed — disabled elements already list as
`button "Save" [disabled, ref=eN]`; the reporter's missing button was
DOM-gated on validity. Added a skill note: `find text` can't reach into a
cross-origin iframe — target those by snapshot @ref.

#45 core skill now distinguishes screenshot-to-locate (discouraged) from
screenshot-to-capture a reusable image asset via `screenshot [--clip] <file>`
(encouraged), so agents stop over-reading the prohibition.

#40 (group-scoped relay) stays deferred — needs an ab-connect extension change.

Verified live: fill --file round-trips multiline+CJK+backticks; contenteditable
engine=contenteditable + get value reads it back; box gives centerX/centerY/
inViewport; screenshot of retina example.com → 2000px; disabled button shows
[disabled]. 856 tests pass.

cli/Cargo.lock

@@ -290,7 +290,7 @@ checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
 
 [[package]]
 name = "chrome-use"
-version = "1.5.20"
+version = "1.5.21"
 dependencies = [
  "aes",
  "aes-gcm",

cli/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "chrome-use"
-version = "1.5.20"
+version = "1.5.21"
 edition = "2021"
 description = "Fast browser automation CLI for AI agents"
 license = "Apache-2.0"

cli/src/commands.rs

@@ -79,6 +79,7 @@ const KNOWN_COMMANDS: &[&str] = &[
     "dialog",
     "upload",
     "site",
+    "box",
 ];
 
 /// Levenshtein distance, capped — small inputs only (command names).
@@ -560,9 +561,37 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
         "fill" => {
             let sel = rest.first().ok_or_else(|| ParseError::MissingArguments {
                 context: "fill".to_string(),
-                usage: "fill <selector> <text>",
+                usage: "fill <selector> <text> | fill <selector> --file <path> | fill <selector> --stdin",
             })?;
-            Ok(json!({ "id": id, "action": "fill", "selector": sel, "value": rest[1..].join(" ") }))
+            // Large/multiline content without shell-escaping hell (issue #41):
+            // `fill <sel> --file <path>` reads the value from a UTF-8 file, and
+            // `fill <sel> --stdin` reads it from stdin — sent verbatim, so backticks,
+            // quotes, newlines and non-ASCII pass through untouched.
+            let value = match rest.get(1).copied() {
+                Some("--file") => {
+                    let path = rest.get(2).ok_or(ParseError::InvalidValue {
+                        message: "fill --file requires a path".to_string(),
+                        usage: "fill <selector> --file <path>",
+                    })?;
+                    std::fs::read_to_string(path).map_err(|e| ParseError::InvalidValue {
+                        message: format!("fill --file: cannot read {path}: {e}"),
+                        usage: "fill <selector> --file <path>",
+                    })?
+                }
+                Some("--stdin") => {
+                    use std::io::Read;
+                    let mut buf = String::new();
+                    io::stdin()
+                        .read_to_string(&mut buf)
+                        .map_err(|e| ParseError::InvalidValue {
+                            message: format!("fill --stdin: {e}"),
+                            usage: "fill <selector> --stdin",
+                        })?;
+                    buf
+                }
+                _ => rest[1..].join(" "),
+            };
+            Ok(json!({ "id": id, "action": "fill", "selector": sel, "value": value }))
         }
         "type" => {
             // `--key-events` (alias `--keys`): send real per-character keystrokes
@@ -1006,11 +1035,55 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             // path: file path (contains / or . or ends with known extension)
             let mut full_page = false;
             let mut clip: Option<Value> = None;
+            let mut max_width: Option<u32> = None;
+            let mut max_height: Option<u32> = None;
+            let mut scale: Option<f64> = None;
             let mut positional: Vec<&str> = Vec::new();
             let mut i = 0;
+            // Parse a numeric value for a downscale flag (issue #42).
+            let parse_num = |i: &mut usize, flag: &str| -> Result<String, ParseError> {
+                let v = rest
+                    .get(*i + 1)
+                    .ok_or_else(|| ParseError::MissingArguments {
+                        context: format!("screenshot {flag}"),
+                        usage: "screenshot [--max-width <px>] [--max-height <px>] [--scale <0..1>]",
+                    })?;
+                *i += 1;
+                Ok(v.to_string())
+            };
             while i < rest.len() {
                 match rest[i] {
                     "--full" | "-f" => full_page = true,
+                    // Downscale the saved image so retina/full-page shots fit an
+                    // agent's image reader and screenshot px line up with click px (#42).
+                    "--max-width" => {
+                        let v = parse_num(&mut i, "--max-width")?;
+                        max_width = Some(v.parse().map_err(|_| ParseError::InvalidValue {
+                            message: format!("--max-width expects a number, got '{v}'"),
+                            usage: "screenshot --max-width <px>",
+                        })?);
+                    }
+                    "--max-height" => {
+                        let v = parse_num(&mut i, "--max-height")?;
+                        max_height = Some(v.parse().map_err(|_| ParseError::InvalidValue {
+                            message: format!("--max-height expects a number, got '{v}'"),
+                            usage: "screenshot --max-height <px>",
+                        })?);
+                    }
+                    "--scale" => {
+                        let v = parse_num(&mut i, "--scale")?;
+                        let s: f64 = v.parse().map_err(|_| ParseError::InvalidValue {
+                            message: format!("--scale expects a number like 0.5, got '{v}'"),
+                            usage: "screenshot --scale <0..1>",
+                        })?;
+                        if s <= 0.0 || s > 1.0 {
+                            return Err(ParseError::InvalidValue {
+                                message: format!("--scale must be in (0, 1], got '{v}'"),
+                                usage: "screenshot --scale <0..1>",
+                            });
+                        }
+                        scale = Some(s);
+                    }
                     // `--clip x,y,w,h` captures a pixel region (issue #34).
                     "--clip" => {
                         let raw = rest
@@ -1073,6 +1146,15 @@ fn parse_command_inner(args: &[String], flags: &Flags) -> Result<Value, ParseErr
             if let Some(c) = clip {
                 cmd["clip"] = c;
             }
+            if let Some(w) = max_width {
+                cmd["maxWidth"] = json!(w);
+            }
+            if let Some(h) = max_height {
+                cmd["maxHeight"] = json!(h);
+            }
+            if let Some(s) = scale {
+                cmd["scale"] = json!(s);
+            }
             if let Some(ref fmt) = flags.screenshot_format {
                 cmd["format"] = json!(fmt);
             }

cli/src/native/actions.rs

@@ -3114,7 +3114,40 @@ async fn handle_screenshot(cmd: &Value, state: &mut DaemonState) -> Result<Value
     )
     .await?;
 
+    // Downscale the saved image so retina/full-page shots fit an agent's image
+    // reader and screenshot pixels line up with `click x y` CSS px (issue #42).
+    // Explicit --scale / --max-width / --max-height win; otherwise a default cap
+    // (2000px longest edge, AGENT_BROWSER_SCREENSHOT_MAX_EDGE overrides, 0 = off)
+    // applies. Annotated shots are left untouched so ref overlays stay aligned.
+    let mut resized: Option<(u32, u32)> = None;
+    if !annotate {
+        let scale = cmd.get("scale").and_then(|v| v.as_f64());
+        let max_w = cmd
+            .get("maxWidth")
+            .and_then(|v| v.as_u64())
+            .map(|v| v as u32);
+        let max_h = cmd
+            .get("maxHeight")
+            .and_then(|v| v.as_u64())
+            .map(|v| v as u32);
+        let default_edge = if scale.is_none() && max_w.is_none() && max_h.is_none() {
+            std::env::var("AGENT_BROWSER_SCREENSHOT_MAX_EDGE")
+                .ok()
+                .and_then(|s| s.parse::<u32>().ok())
+                .or(Some(2000))
+                .filter(|&e| e > 0)
+        } else {
+            None
+        };
+        resized = downscale_screenshot(&result.path, scale, max_w, max_h, default_edge);
+    }
+
     let mut response = json!({ "path": absolutize_saved_path(&result.path) });
+    if let Some((w, h)) = resized {
+        response["width"] = json!(w);
+        response["height"] = json!(h);
+        response["resized"] = json!(true);
+    }
     if !result.annotations.is_empty() {
         response["annotations"] = serde_json::to_value(&result.annotations)
             .map_err(|e| format!("Failed to serialize annotations: {}", e))?;
@@ -3130,6 +3163,56 @@ async fn handle_screenshot(cmd: &Value, state: &mut DaemonState) -> Result<Value
     Ok(response)
 }
 
+/// Downscale a saved screenshot in place (issue #42). Resolves the target longest
+/// edge from `scale` (fraction of current), explicit `max_w`/`max_h` caps, or a
+/// `default_edge` cap — whichever yields the smaller image. Only ever shrinks;
+/// no-op (returns None) if the image is already within bounds or can't be read.
+/// Returns the new (width, height) when it actually resized.
+fn downscale_screenshot(
+    path: &str,
+    scale: Option<f64>,
+    max_w: Option<u32>,
+    max_h: Option<u32>,
+    default_edge: Option<u32>,
+) -> Option<(u32, u32)> {
+    let img = image::open(path).ok()?;
+    let (w, h) = (img.width(), img.height());
+    if w == 0 || h == 0 {
+        return None;
+    }
+
+    // Collect candidate scale factors (≤ 1.0); the smallest wins.
+    let mut factor = 1.0f64;
+    if let Some(s) = scale {
+        factor = factor.min(s);
+    }
+    if let Some(mw) = max_w {
+        if w > mw {
+            factor = factor.min(mw as f64 / w as f64);
+        }
+    }
+    if let Some(mh) = max_h {
+        if h > mh {
+            factor = factor.min(mh as f64 / h as f64);
+        }
+    }
+    if let Some(edge) = default_edge {
+        let longest = w.max(h);
+        if longest > edge {
+            factor = factor.min(edge as f64 / longest as f64);
+        }
+    }
+
+    if factor >= 1.0 {
+        return None; // already within bounds — never upscale
+    }
+    let nw = ((w as f64 * factor).round() as u32).max(1);
+    let nh = ((h as f64 * factor).round() as u32).max(1);
+    let resized = img.resize(nw, nh, image::imageops::FilterType::Lanczos3);
+    resized.save(path).ok()?;
+    Some((resized.width(), resized.height()))
+}
+
 async fn handle_click(cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {
     // First-class coordinate click (issue #8.4): click a raw viewport point with
     // no element resolution. Parsed from `click <x> <y>` / `click --coords x,y`.
@@ -3288,7 +3371,7 @@ async fn handle_fill(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();
 
-    interaction::fill(
+    let engine = interaction::fill(
         &mgr.client,
         &session_id,
         &state.ref_map,
@@ -3297,7 +3380,9 @@ async fn handle_fill(cmd: &Value, state: &mut DaemonState) -> Result<Value, Stri
         &state.iframe_sessions,
     )
     .await?;
-    Ok(json!({ "filled": selector }))
+    // Echo the input path used (input/contenteditable/codemirror5/monaco/select)
+    // so the agent can confirm a rich editor was handled, not silently no-op'd (#41).
+    Ok(json!({ "filled": selector, "engine": engine }))
 }
 
 async fn handle_type(cmd: &Value, state: &mut DaemonState) -> Result<Value, String> {

cli/src/native/element.rs

@@ -1529,9 +1529,27 @@ pub async fn get_element_input_value(
         .send_command_typed(
             "Runtime.callFunctionOn",
             &CallFunctionOnParams {
-                function_declaration:
-                    "function() { return typeof this.value === 'string' ? this.value : ''; }"
-                        .to_string(),
+                // Read rich-editor content too (issue #41): CodeMirror 5 / Monaco
+                // keep their text in a model, not `.value`; contenteditable keeps
+                // it as innerText. Falls back to `.value` for plain inputs.
+                function_declaration: r#"function() {
+                    const el = this;
+                    const cm5 = el.closest && el.closest('.CodeMirror');
+                    if (cm5 && cm5.CodeMirror) return cm5.CodeMirror.getValue();
+                    if (window.monaco && monaco.editor) {
+                        try {
+                            const eds = monaco.editor.getEditors ? monaco.editor.getEditors() : [];
+                            const ed = eds.find(e => e.getDomNode && e.getDomNode().contains(el)) || eds[0];
+                            if (ed) return ed.getValue();
+                            const m = monaco.editor.getModels ? monaco.editor.getModels() : [];
+                            if (m[0]) return m[0].getValue();
+                        } catch (e) {}
+                    }
+                    if (typeof el.value === 'string') return el.value;
+                    if (el.isContentEditable) return el.innerText;
+                    return '';
+                }"#
+                .to_string(),
                 object_id: Some(object_id),
                 arguments: None,
                 return_by_value: Some(true),
@@ -1609,7 +1627,15 @@ pub async fn get_element_bounding_box(
             &CallFunctionOnParams {
                 function_declaration: r#"function() {
                     const r = this.getBoundingClientRect();
-                    return { x: r.x, y: r.y, width: r.width, height: r.height };
+                    const inViewport = r.bottom > 0 && r.right > 0
+                        && r.top < (innerHeight || document.documentElement.clientHeight)
+                        && r.left < (innerWidth || document.documentElement.clientWidth);
+                    return {
+                        x: r.x, y: r.y, width: r.width, height: r.height,
+                        centerX: Math.round(r.x + r.width / 2),
+                        centerY: Math.round(r.y + r.height / 2),
+                        inViewport,
+                    };
                 }"#
                 .to_string(),
                 object_id: Some(object_id),

cli/src/native/interaction.rs

@@ -579,7 +579,7 @@ pub async fn fill(
     selector_or_ref: &str,
     value: &str,
     iframe_sessions: &HashMap<String, String>,
-) -> Result<(), String> {
+) -> Result<String, String> {
     let (object_id, effective_session_id) = resolve_element_object_id(
         client,
         session_id,
@@ -590,14 +590,15 @@ pub async fn fill(
     .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.
+    // site-side listeners actually see the change (issue #25): set the value
+    // through the element's PROTOTYPE setter (which React's _valueTracker hooks),
+    // then dispatch input → change → blur/focusout. Beyond plain inputs, detect
+    // rich editors and use their own API/events (issue #41): CodeMirror 5 and
+    // Monaco have a model that `.value`/`textContent` can't touch; ProseMirror /
+    // contenteditable need `execCommand('insertText')` so beforeinput/input fire
+    // (a raw `textContent =` corrupts PM's doc and skips React composers).
+    // Returns the engine used so the caller can report it. `type <sel> <text>`
+    // remains for sites that need per-keystroke events.
     let fill_js = format!(
         r#"function() {{
             const el = this;
@@ -605,13 +606,43 @@ pub async fn fill(
             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;
+
+            // CodeMirror 5: a hidden <textarea> inside .CodeMirror with a live instance.
+            const cm5 = el.closest && el.closest('.CodeMirror');
+            if (cm5 && cm5.CodeMirror) {{ cm5.CodeMirror.setValue(v); return 'codemirror5'; }}
+
+            // Monaco: global `monaco`; prefer the editor whose DOM contains el.
+            if (window.monaco && monaco.editor) {{
+                try {{
+                    const eds = monaco.editor.getEditors ? monaco.editor.getEditors() : [];
+                    const ed = eds.find(e => e.getDomNode && e.getDomNode().contains(el)) || eds[0];
+                    if (ed) {{ ed.setValue(v); return 'monaco'; }}
+                    const models = monaco.editor.getModels ? monaco.editor.getModels() : [];
+                    if (models[0]) {{ models[0].setValue(v); return 'monaco'; }}
+                }} catch (e) {{}}
             }}
+
+            if (tag === 'SELECT') {{ el.value = v; fire('input'); fire('change'); return 'select'; }}
+
             if (el.isContentEditable) {{
-                el.textContent = v; fire('input', window.InputEvent || Event); fire('change');
-                try {{ el.blur(); }} catch (e) {{}} fire('focusout'); return true;
+                // ProseMirror / contenteditable: select-all then insertText fires
+                // beforeinput/input that PM and React composers listen for.
+                let ok = false;
+                try {{
+                    const sel = window.getSelection();
+                    const range = document.createRange();
+                    range.selectNodeContents(el);
+                    sel.removeAllRanges();
+                    sel.addRange(range);
+                    ok = document.execCommand('insertText', false, v);
+                }} catch (e) {{}}
+                if (!ok) {{ el.textContent = v; fire('input', window.InputEvent || Event); }}
+                fire('change');
+                try {{ el.blur(); }} catch (e) {{}}
+                fire('focusout');
+                return ok ? 'contenteditable' : 'contenteditable-fallback';
             }}
+
             const proto = tag === 'TEXTAREA' ? window.HTMLTextAreaElement.prototype
                                              : window.HTMLInputElement.prototype;
             const desc = Object.getOwnPropertyDescriptor(proto, 'value');
@@ -623,13 +654,13 @@ pub async fn fill(
             fire('change');
             try {{ el.blur(); }} catch (e) {{}}
             fire('focusout');                          // blur-triggered lookups/validation
-            return true;
+            return 'input';
         }}"#,
         val = serde_json::to_string(value).unwrap_or_default()
     );
 
-    client
-        .send_command_typed::<_, Value>(
+    let result: EvaluateResult = client
+        .send_command_typed(
             "Runtime.callFunctionOn",
             &CallFunctionOnParams {
                 function_declaration: fill_js,
@@ -642,7 +673,11 @@ pub async fn fill(
         )
         .await?;
 
-    Ok(())
+    Ok(result
+        .result
+        .value
+        .and_then(|v| v.as_str().map(String::from))
+        .unwrap_or_else(|| "input".to_string()))
 }
 
 #[allow(clippy::too_many_arguments)]

cli/src/output.rs

@@ -489,6 +489,17 @@ pub fn print_response_with_opts(resp: &Response, action: Option<&str>, opts: &Ou
                 println!("y:      {}", y);
                 println!("width:  {}", w);
                 println!("height: {}", h);
+                if let (Some(cx), Some(cy)) = (
+                    obj.get("centerX").and_then(|v| v.as_i64()),
+                    obj.get("centerY").and_then(|v| v.as_i64()),
+                ) {
+                    // Echoed in click-ready CSS px so the agent can paste straight
+                    // into `click <centerX> <centerY>` (issue #43).
+                    println!("center: {} {}", cx, cy);
+                }
+                if let Some(iv) = obj.get("inViewport").and_then(|v| v.as_bool()) {
+                    println!("inViewport: {}", iv);
+                }
             }
             return;
         }
@@ -1490,9 +1501,20 @@ Examples:
 chrome-use fill - Clear and fill an input field
 
 Usage: chrome-use fill <selector> <text>
+       chrome-use fill <selector> --file <path>
+       chrome-use fill <selector> --stdin
 
-Clears the input field and fills it with the specified text.
-This replaces any existing content in the field.
+Clears the field and fills it with the text, replacing existing content.
+Works on rich editors too (issue #41): CodeMirror 5, Monaco, ProseMirror and
+plain contenteditable are detected and set via their own API / input events,
+not a raw `.value` write — and the response echoes which `engine` was used.
+For framework inputs (React/Vue/Angular) the value goes through the native
+setter so the form registers it (no more "pristine" Save no-ops).
+
+Options:
+  --file <path>        Read the value from a UTF-8 file (large/multiline text,
+                       backticks/quotes/newlines/non-ASCII — no shell escaping)
+  --stdin              Read the value from stdin
 
 Global Options:
   --json               Output as JSON
@@ -1501,7 +1523,8 @@ Global Options:
 Examples:
   chrome-use fill "#email" "user@example.com"
   chrome-use fill @e3 "Hello World"
-  chrome-use fill "input[name='search']" "query"
+  chrome-use fill ".CodeMirror" --file ./article.md     # set a CodeMirror editor
+  cat post.md | chrome-use fill @e7 --stdin
 "##
         }
         "type" => {
@@ -1903,6 +1926,13 @@ 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`
+  --max-width <px>     Downscale so the image's width ≤ px (preserves aspect)
+  --max-height <px>    Downscale so the image's height ≤ px
+  --scale <0..1>       Downscale by a factor, e.g. 0.5 (DPR-1, so screenshot px
+                       line up 1:1 with `click x y`)
+                       Default: capped at 2000px longest edge unless overridden
+                       (AGENT_BROWSER_SCREENSHOT_MAX_EDGE; 0 disables). Annotated
+                       shots are never downscaled, so ref overlays stay aligned.
   --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.
@@ -1925,6 +1955,8 @@ Examples:
   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 --scale 0.5 ./half.png   # DPR-1: screenshot px == click px
+  chrome-use screenshot --max-width 1400 ./shot.png  # cap width for image readers
   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
@@ -3263,7 +3295,8 @@ Core Commands:
   click <sel|x y>            Click element/@ref, or a viewport coordinate
   dblclick <sel>             Double-click element
   type <sel> <text>          Type into element
-  fill <sel> <text>          Clear and fill
+  fill <sel> <text>          Clear and fill (handles CodeMirror/Monaco/ProseMirror/
+                             contenteditable; `--file <path>`/`--stdin` for large text)
   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`
@@ -3283,7 +3316,8 @@ Core Commands:
   scroll <dir> [px]          Scroll (up/down/left/right)
   scrollintoview <sel>       Scroll element into view
   wait <sel|ms>              Wait for element or time
-  screenshot [path]          Take screenshot
+  screenshot [path]          Take screenshot (auto-downscaled to ≤2000px long edge;
+                             --max-width/--max-height/--scale to override)
   pdf <path>                 Save as PDF
   snapshot                   Accessibility tree with refs (for AI)
   eval <js>                  Run JavaScript
@@ -3297,6 +3331,8 @@ Navigation:
 
 Get Info:  chrome-use get <what> [selector]
   text, html, value, attr <name>, title, url, count, box, styles, cdp-url
+  box <sel>  →  x,y,width,height,centerX,centerY,inViewport in CSS px (feed
+               centerX/centerY into `click x y`); value reads CodeMirror/Monaco too
   text (no selector = whole page, all frames), text --main, frames (list)
 
 Check State:  chrome-use is <what> <selector>

package.json

@@ -1,6 +1,6 @@
 {
   "name": "chrome-use",
-  "version": "1.5.20",
+  "version": "1.5.21",
   "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

@@ -59,6 +59,18 @@ next ref interaction.
 > 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.
 
+> **Two different intents — only one is discouraged.** The rule above is about
+> *screenshot-to-locate* (using a picture to find/hit an element) — that's the bug.
+> *screenshot-to-capture* — saving a region or element to a file as a **reusable
+> image asset** (maps, charts, og-images, visual-diff baselines, report figures) —
+> is fully supported and encouraged: `screenshot [selector] [--clip x,y,w,h] <file>`.
+> Capturing a rendered map region to a PNG for a blog post is the right tool, not a
+> smell. Screenshots are auto-downscaled to ≤2000px (longest edge) so they fit an
+> image reader and their pixels line up with `click x y`; override with
+> `--max-width`/`--max-height`/`--scale`. To click something you couldn't hit by
+> ref, `box @ref` gives the element's CSS-px box + `centerX/centerY` to feed
+> straight into `click <centerX> <centerY>` — no screenshot needed.
+
 ## Before you automate: pick the cheapest tool
 
 Driving a browser is the heavy option. chrome-use earns its keep when you
@@ -373,6 +385,12 @@ foreground, so prefer refs. For below-the-fold content in such a frame, scroll i
 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`.
 
+> **Caveat: `find text "…"` can't reach into a cross-origin iframe** — it errors
+> "Element not found" even though `snapshot -i` lists those nodes and
+> `get text` reads them. Inside cross-origin iframes, target elements by their
+> **snapshot `@ref`**, not by `find`. (`box @ref` also works on iframe refs when
+> you need a coordinate fallback.)
+
 ### When refs don't work or you don't want to snapshot
 
 Use semantic locators: