fix: preserve browser scroll position and configure commit attribution

scripts/check-core-boundaries.test.mjs

@@ -3081,6 +3081,10 @@ test('third-party capability profiles reject ambient feature unions and unreview
       uses_default_features: false,
       features: ['vendored-libgit2'],
     }]),
+    packageAt('openbitfun-webdriver', 'src/crates/adapters/webdriver/Cargo.toml', [{
+      name: 'image', kind: null, optional: false, uses_default_features: false,
+      features: ['jpeg', 'png'],
+    }]),
   ];
 
   assert.deepEqual(findThirdPartyCapabilityFeatureViolations(validPackages), []);
@@ -3091,6 +3095,7 @@ test('third-party capability profiles reject ambient feature unions and unreview
   mutatedPackages[2].dependencies[0].features.push('rustls-tls-native-roots');
   mutatedPackages[3].dependencies[0].features.push('https');
   mutatedPackages[3].dependencies[0].rename = 'private-git2';
+  mutatedPackages[4].dependencies[0].features = ['png', 'webp'];
   mutatedPackages.push(packageAt('future-image-owner', 'src/apps/future/Cargo.toml', [{
     name: 'image',
     kind: null,
@@ -3107,6 +3112,8 @@ test('third-party capability profiles reject ambient feature unions and unreview
   assert.match(messages, /openbitfun-core Tokio Tungstenite dependency has unexpected features: rustls-tls-native-roots/);
   assert.match(messages, /openbitfun-services-core Git2 dependency has unexpected features: https/);
   assert.match(messages, /openbitfun-services-core Git2 dependency does not match its reviewed owner shape/);
+  assert.match(messages, /openbitfun-webdriver Image dependency missing features: jpeg/);
+  assert.match(messages, /openbitfun-webdriver Image dependency has unexpected features: webp/);
   assert.match(messages, /future-image-owner Image dependency is missing a reviewed owner profile/);
 });
 

scripts/core-boundaries/cargo-dependency-boundaries.mjs

@@ -574,7 +574,8 @@ const THIRD_PARTY_CAPABILITY_PROFILES = new Map([
       ['openbitfun-skin-market-service', dependencyProfile(['gif', 'jpeg', 'png', 'webp'], {
         useDefaultFeatures: false,
       })],
-      ['openbitfun-webdriver', dependencyProfile(['png'], {
+      // The embedded bridge captures PNG and encodes JPEG previews itself.
+      ['openbitfun-webdriver', dependencyProfile(['jpeg', 'png'], {
         useDefaultFeatures: false,
       })],
     ]),

src/crates/adapters/webdriver/Cargo.toml

@@ -16,10 +16,10 @@ tokio = { workspace = true, features = ["net", "rt", "sync", "time"] }
 serde = { workspace = true }
 serde_json = { workspace = true }
 log = { workspace = true }
-tauri = { workspace = true }
+tauri = { workspace = true, features = ["unstable"] }
 uuid = { workspace = true }
 base64 = { workspace = true }
-image = { workspace = true, features = ["png"] }
+image = { workspace = true, features = ["jpeg", "png"] }
 
 [target.'cfg(target_os = "macos")'.dependencies]
 block2 = { workspace = true }

src/crates/adapters/webdriver/src/runtime/api/element.rs

@@ -67,11 +67,11 @@ pub(crate) fn clear() -> &'static str {
 }
 
 pub(crate) fn send_keys() -> &'static str {
-    "(id, text) => { const el = window.__openbitfunWd.getElement(id); if (!el) { throw new Error('Element not found'); } window.__openbitfunWd.insertText(el, text); return null; }"
+    "(id, text) => { const el = window.__openbitfunWd.getElement(id); if (!el) { throw new Error('Element not found'); } if ('value' in el || el.isContentEditable) { window.__openbitfunWd.scrollElementIntoView(el); window.__openbitfunWd.focusWithoutScroll(el); } window.__openbitfunWd.insertText(el, text); return null; }"
 }
 
 pub(crate) fn screenshot_metadata() -> &'static str {
-    "(id) => { const el = window.__openbitfunWd.getElement(id); if (!el || !el.isConnected) { throw new Error('stale element reference'); } el.scrollIntoView({ block: 'center', inline: 'center' }); const rect = el.getBoundingClientRect(); return { x: rect.x, y: rect.y, width: rect.width, height: rect.height, devicePixelRatio: window.devicePixelRatio || 1 }; }"
+    "(id) => { const el = window.__openbitfunWd.getElement(id); if (!el || !el.isConnected) { throw new Error('stale element reference'); } window.__openbitfunWd.scrollElementIntoView(el); const rect = el.getBoundingClientRect(); return { x: rect.x, y: rect.y, width: rect.width, height: rect.height, devicePixelRatio: window.devicePixelRatio || 1 }; }"
 }
 
 pub(crate) fn get_shadow_root() -> &'static str {

src/crates/adapters/webdriver/src/runtime/script/core/execution.rs

@@ -78,6 +78,8 @@ pub(super) fn script() -> &'static str {
       validateFrameElement,
       getShadowRoot,
       isDisplayed,
+      scrollElementIntoView,
+      focusWithoutScroll,
       clearElement,
       insertText,
       setElementText,

src/crates/adapters/webdriver/src/runtime/script/core/interaction.rs

@@ -0,0 +1,28 @@
+pub(super) fn script() -> &'static str {
+    r####"
+    // Reveal an offscreen target with minimal movement. Centering also scrolls
+    // overflow:hidden ancestors, which can displace the surrounding app chrome.
+    const scrollElementIntoView = (element) => {
+      element.scrollIntoView({ behavior: "instant", block: "nearest", inline: "nearest" });
+    };
+
+    // Positioning belongs to the caller; focus must not move it again, especially
+    // after pointer coordinates have already been resolved.
+    const focusWithoutScroll = (element) => {
+      if (typeof element.focus === "function") {
+        element.focus({ preventScroll: true });
+      }
+    };
+
+    // A large target can remain only partly visible after nearest scrolling.
+    // Use its visible center rather than a point outside the viewport.
+    const getInViewCenter = (element) => {
+      const rect = element.getClientRects()[0] || element.getBoundingClientRect();
+      const ownerWindow = element.ownerDocument.defaultView || window;
+      return {
+        x: (Math.max(0, rect.left) + Math.min(ownerWindow.innerWidth, rect.right)) / 2,
+        y: (Math.max(0, rect.top) + Math.min(ownerWindow.innerHeight, rect.bottom)) / 2
+      };
+    };
+"####
+}

src/crates/adapters/webdriver/src/runtime/script/core/mod.rs

@@ -2,6 +2,7 @@ mod alert;
 mod context;
 mod cookie;
 mod execution;
+mod interaction;
 mod locator;
 mod runtime;
 mod shadow;
@@ -10,13 +11,14 @@ mod visibility;
 
 pub(super) fn head() -> String {
     format!(
-        "{runtime}{store}{context}{locator}{shadow}{visibility}",
+        "{runtime}{store}{context}{locator}{shadow}{visibility}{interaction}",
         runtime = runtime::script(),
         store = store::script(),
         context = context::script(),
         locator = locator::script(),
         shadow = shadow::script(),
-        visibility = visibility::script()
+        visibility = visibility::script(),
+        interaction = interaction::script()
     )
 }
 

src/crates/adapters/webdriver/src/runtime/script/input.rs

@@ -72,8 +72,11 @@ pub(super) fn script() -> &'static str {
       if (!element) {
         return;
       }
+      if ("value" in element || element.isContentEditable) {
+        scrollElementIntoView(element);
+        focusWithoutScroll(element);
+      }
       if ("value" in element) {
-        element.focus();
         if (!dispatchBeforeInputEvent(element, "deleteContentBackward", null)) {
           return;
         }
@@ -82,7 +85,6 @@ pub(super) fn script() -> &'static str {
         return;
       }
       if (element.isContentEditable) {
-        element.focus();
         element.textContent = "";
         emitInputEvents(element, "deleteContentBackward", null);
       }
@@ -112,7 +114,7 @@ pub(super) fn script() -> &'static str {
           return;
         }
         const selection = ownerWindow.getSelection();
-        element.focus();
+        focusWithoutScroll(element);
         if (selection && selection.rangeCount > 0) {
           selection.deleteFromDocument();
           selection.getRangeAt(0).insertNode(element.ownerDocument.createTextNode(text));

src/crates/adapters/webdriver/src/runtime/script/pointer/mouse.rs

@@ -37,15 +37,16 @@ pub(super) fn script() -> &'static str {
         };
       }
 
-      if (origin && typeof origin === "object" && typeof origin[ELEMENT_KEY] === "string") {
-        const element = getElement(origin[ELEMENT_KEY]);
+      // Script arguments have already deserialized wire element references.
+      if (isElementLike(origin) || (origin && typeof origin === "object" && typeof origin[ELEMENT_KEY] === "string")) {
+        const element = isElementLike(origin) ? origin : getElement(origin[ELEMENT_KEY]);
         if (!element) {
           throw new Error("Element not found");
         }
-        const rect = element.getBoundingClientRect();
+        const { x, y } = getInViewCenter(element);
         return {
-          x: rect.left + rect.width / 2 + (Number(action?.x) || 0),
-          y: rect.top + rect.height / 2 + (Number(action?.y) || 0),
+          x: x + (Number(action?.x) || 0),
+          y: y + (Number(action?.y) || 0),
           target: element
         };
       }
@@ -108,13 +109,9 @@ pub(super) fn script() -> &'static str {
       if (!element) {
         throw new Error("Element not found");
       }
-      element.scrollIntoView({ block: "center", inline: "center" });
-      if (typeof element.focus === "function") {
-        element.focus();
-      }
-      const rect = element.getBoundingClientRect();
-      const x = rect.left + rect.width / 2;
-      const y = rect.top + rect.height / 2;
+      scrollElementIntoView(element);
+      focusWithoutScroll(element);
+      const { x, y } = getInViewCenter(element);
       updatePointerTarget(getFrameContext(), x, y, element);
       const runtime = ensureRuntimeState();
       const buttonMask = pointerButtonMask(button);

src/crates/adapters/webdriver/src/runtime/script/pointer/pointer_source.rs

@@ -35,9 +35,7 @@ pub(super) fn script() -> &'static str {
           if (action.duration) {
             await sleep(action.duration);
           }
-          if (typeof target.focus === "function") {
-            target.focus();
-          }
+          focusWithoutScroll(target);
           runtime.pointer.buttons |= buttonMask;
           dispatchMouseEvent(target, "mousedown", runtime.pointer.x, runtime.pointer.y, button, runtime.pointer.buttons, frameContext);
           continue;

src/crates/adapters/webdriver/src/runtime/script/pointer/wheel.rs

@@ -1,22 +1,41 @@
 pub(super) fn script() -> &'static str {
     r####"
-    const findScrollableTarget = (target, doc) => {
+    const findScrollableTarget = (target, doc, deltaX, deltaY) => {
       let current = target;
       while (current && current !== doc.body && current !== doc.documentElement) {
-        if (
-          (current.scrollHeight > current.clientHeight || current.scrollWidth > current.clientWidth) &&
-          current instanceof Element
-        ) {
-          return current;
+        if (isElementLike(current)) {
+          const style = getOwnerWindow(current).getComputedStyle(current);
+          const wheelScrollable = (overflow) => ["auto", "scroll", "overlay"].includes(overflow);
+          // overflow:hidden can be scrolled by script, but never by a user wheel.
+          if (
+            (deltaX && current.scrollWidth > current.clientWidth && wheelScrollable(style.overflowX)) ||
+            (deltaY && current.scrollHeight > current.clientHeight && wheelScrollable(style.overflowY))
+          ) {
+            return current;
+          }
         }
         current = current.parentElement;
       }
-      return doc.scrollingElement || doc.documentElement || doc.body;
+      const viewport = doc.scrollingElement || doc.documentElement || doc.body;
+      if (!viewport) {
+        return null;
+      }
+      const ownerWindow = doc.defaultView || window;
+      const style = ownerWindow.getComputedStyle(viewport);
+      const bodyStyle = doc.body ? ownerWindow.getComputedStyle(doc.body) : style;
+      // Root overflow:visible propagates the body's overflow to the viewport.
+      const overflowX = style.overflowX === "visible" ? bodyStyle.overflowX : style.overflowX;
+      const overflowY = style.overflowY === "visible" ? bodyStyle.overflowY : style.overflowY;
+      const wheelScrollable = (overflow) => overflow !== "hidden" && overflow !== "clip";
+      return (
+        (deltaX && viewport.scrollWidth > viewport.clientWidth && wheelScrollable(overflowX)) ||
+        (deltaY && viewport.scrollHeight > viewport.clientHeight && wheelScrollable(overflowY))
+      ) ? viewport : null;
     };
 
     const applyWheelScroll = (target, deltaX, deltaY, frameContext = currentFrameContext) => {
       const doc = getCurrentDocument(frameContext);
-      const scrollTarget = findScrollableTarget(target, doc);
+      const scrollTarget = findScrollableTarget(target, doc, deltaX, deltaY);
       if (!scrollTarget) {
         return;
       }

src/crates/assembly/core/builtin_skills/commit-push-pr/SKILL.md

@@ -13,9 +13,11 @@ Invoking this skill is authorization for the requested commit/push/PR flow. It i
 
 ## OpenBitFun attribution contract
 
-Unless the user explicitly opts out, always use the PR footer below exactly once. When this flow creates one or more real commits, also use the commit trailer exactly once in each new commit:
+Unless the user explicitly opts out, use the PR footer below exactly once. Commit co-author attribution follows the execution host's setting, resolved when this built-in skill is loaded:
 
-- Commit trailer for each new commit, which GitHub uses for contributor/co-author attribution:
+{{OPENBITFUN_GIT_COMMIT_COAUTHOR_POLICY}}
+
+- Commit trailer when co-author attribution is enabled, which GitHub uses for contributor/co-author attribution:
 
   ```text
   Co-authored-by: OpenBitFun <318544290+bitfun-ai@users.noreply.github.com>
@@ -37,7 +39,7 @@ Do not duplicate either attribution. Preserve an equivalent existing trailer/foo
 2. Confirm scope; stage only intended paths.
 3. Create `openbitfun/<description>` only when currently on the base branch.
 4. Run the repository's focused verification.
-5. Commit real changes with the exact OpenBitFun trailer once.
+5. Commit real changes, following the resolved commit co-author setting above.
 6. Push normally; never force-push automatically.
 7. Update the exact matching PR or create a draft PR.
 8. End the PR body with the exact OpenBitFun link footer once.
@@ -104,11 +106,11 @@ If verification remains failed, blocked, or was explicitly declined by the user:
 
 - Derive the subject from the actual diff and repository conventions; do not create a generic `prepare PR` commit.
 - Keep the real human author/committer identity. The OpenBitFun line is a co-author trailer, not a replacement author.
-- Add a blank line before the exact OpenBitFun trailer.
+- Only when commit co-author attribution is enabled and the user has not opted out, add a blank line before the exact OpenBitFun trailer. When disabled, do not add this trailer to new commits.
 - If the intended changes are already committed, do not create an empty attribution commit.
 - Do not squash multiple existing commits or amend published commits without explicit approval.
 
-Example:
+Example when commit co-author attribution is enabled:
 
 ```text
 fix: preserve session state during reconnect

src/crates/assembly/core/src/agentic/tools/browser_control/actions.rs

@@ -648,10 +648,12 @@ impl<'a> BrowserActions<'a> {
         format!(
             r#"(function(){{
                 {js}
-                el.scrollIntoView({{ block: 'center', inline: 'center', behavior: 'instant' }});
-                const rect = el.getBoundingClientRect();
-                const localX = rect.x + rect.width / 2;
-                const localY = rect.y + rect.height / 2;
+                // Reveal the target without repositioning already visible content.
+                el.scrollIntoView({{ block: 'nearest', inline: 'nearest', behavior: 'instant' }});
+                const rect = el.getClientRects()[0] || el.getBoundingClientRect();
+                const ownerWindow = el.ownerDocument.defaultView || window;
+                const localX = (Math.max(0, rect.left) + Math.min(ownerWindow.innerWidth, rect.right)) / 2;
+                const localY = (Math.max(0, rect.top) + Math.min(ownerWindow.innerHeight, rect.bottom)) / 2;
                 let x = localX;
                 let y = localY;
                 try {{
@@ -721,7 +723,7 @@ impl<'a> BrowserActions<'a> {
     pub async fn fill(&self, selector: &str, value: &str) -> OpenBitFunResult<Value> {
         let js = Self::resolve_element_js(selector);
         let focus_js = format!(
-            r#"(function(){{ {} el.focus(); el.value = ''; el.dispatchEvent(new Event('input', {{ bubbles: true }})); return true; }})()"#,
+            r#"(function(){{ {} el.scrollIntoView({{ block: 'nearest', inline: 'nearest', behavior: 'instant' }}); el.focus({{ preventScroll: true }}); el.value = ''; el.dispatchEvent(new Event('input', {{ bubbles: true }})); return true; }})()"#,
             js
         );
         self.evaluate(&focus_js).await?;

src/crates/assembly/core/src/agentic/tools/implementations/skills/mod.rs

@@ -8,6 +8,7 @@ pub mod mode_overrides;
 pub mod policy;
 pub mod registry;
 pub mod resolver;
+mod runtime_settings;
 #[cfg(feature = "file-watch")]
 mod source_cache;
 pub mod types;

src/crates/assembly/core/src/agentic/tools/implementations/skills/registry.rs

@@ -8,6 +8,7 @@ use super::mode_overrides::{
     load_globally_disabled_project_skills, load_globally_disabled_user_skills,
     load_user_mode_skill_overrides, SkillPolicyWorkspace, UserModeSkillOverrides,
 };
+use super::runtime_settings::apply_builtin_runtime_settings;
 #[cfg(feature = "file-watch")]
 use super::source_cache::{LocalSkillWatchMonitor, LocalSkillWatchRoot, VersionedSnapshotCache};
 use super::types::{
@@ -1996,6 +1997,7 @@ impl SkillRegistry {
         data.source_id = info.source_id;
         data.source_label = info.source_label;
         data.dir_name = info.dir_name;
+        apply_builtin_runtime_settings(&mut data).await?;
         Ok(data)
     }
 
@@ -2040,6 +2042,7 @@ impl SkillRegistry {
         data.source_id = info.source_id;
         data.source_label = info.source_label;
         data.dir_name = info.dir_name;
+        apply_builtin_runtime_settings(&mut data).await?;
         Ok(data)
     }
 
@@ -2075,6 +2078,7 @@ impl SkillRegistry {
         data.source_id = info.source_id;
         data.source_label = info.source_label;
         data.dir_name = info.dir_name;
+        apply_builtin_runtime_settings(&mut data).await?;
         Ok(data)
     }
 
@@ -2118,6 +2122,7 @@ impl SkillRegistry {
         data.source_id = info.source_id;
         data.source_label = info.source_label;
         data.dir_name = info.dir_name;
+        apply_builtin_runtime_settings(&mut data).await?;
         Ok(data)
     }