Browser extension

Feedbacks captures client feedback and UI test findings with the context a developer’s coding agent needs. Your team needs a running Feedbacks server first. DevOps installs it once; reviewers need its URL, their own account and project access. Follow the complete setup order or the DevOps installation guide.

Install, pin and connect from Chrome Web Store

1. Open the Feedbacks Chrome Web Store listing. Choose Add to Chrome, then Add extension.
2. Open Chrome’s Extensions menu (the puzzle-piece button). Find Feedbacks and click its pin so the icon stays in the toolbar. Pinning is important for quick access: use that icon whenever you want to start or resume review.
3. Sign in to your team’s Feedbacks web app and open Setup.
4. Click the pinned Feedbacks icon while this page is active. An updated server identifies itself and the extension fills an empty server setting automatically. Set in extension, beside Copy server URL, opens extension Settings with that address after you have opened the icon on this tab. Before that first activation, the button explains how to open the icon; older versions can still use copy/paste.
5. Choose Connect to server, allow Chrome’s access to that server, sign in and approve pairing.

Chrome manages updates for Store installations. For your first capture: open the website → click the pinned icon → hover and right-click an element → write the note → Save point → Review & send to finalize and inspect the evidence → Send & Open. Saved points are local drafts until sending. See the complete reviewer walkthrough.

Detection uses the current tab only and adds no Chrome permissions. It never replaces a saved server, an address you are typing or pending pairing. It does not connect an account automatically. For a fresh local HTTP installation, paste the address and enable the explicit local-server option before connecting; HTTPS is detected automatically. Refreshing the page removes its setup bridge, so reopen the icon before using Set in extension again.

Install from source

1. Run npm ci and npm run build:extension.
2. Open chrome://extensions, enable Developer mode, choose Load unpacked, and select dist/extension/unpacked/ (the built folder includes the locally bundled replay recorder).
3. Open Feedbacks from Chrome's toolbar. Enter your server's origin, such as https://feedback.example.com.
4. Choose Connect to server, grant permission to that server and approve pairing in its web application.
5. Open a website registered in a project you can access. Start a review, capture or select a point, redact sensitive content, and send the feedback.

For an element-specific review, open Feedbacks to start review. Hover an element to see its outline, then right-click it. Feedbacks keeps the selected element outlined and opens a comment field beside it. Write a note and choose Save point, then right-click another element to add more numbered notes. Orange dashed pins and the compact not sent count mean the points are still only in this browser. Hover or focus an orange pin to open its note, unsent status and Edit point action; click it to edit before capture. Use Review & send beside the movable Page Controls icon when finished. Its N unsent count stays visible while the panel is collapsed. This opens the saved point originals without taking another screenshot; the editor shows every note and its order, with an optional overall comment and the drawing/redaction tools. Send & Open creates the shared thread first, then uploads its approved screenshots. The page switches from draft markers to published points as soon as the thread exists. If an upload fails, the editor shows that the thread is already published and keeps the remaining images for Retry Send; retry does not create a second thread.

Original views and responsive points

The comment editor opens immediately while its original visible screenshot is captured in the background. Capturing… means you can type while waiting; a small thumbnail and Original saved · Not sent confirm the view was retained locally. Save waits for that capture to finish. The saved pin also previews its original image. The movable Feedbacks icon counts unsent points and points outside the current view; hover, focus or click it to open tools, edit any point or inspect Original view. Drag the icon to move it. The tool panel collapses when the pointer leaves, except while editing a draft. Hide removes the controls while review stays active; reopen the Chrome popup to restore them. Exit stops review. A draft pin follows its original element when the layout changes. Hidden, removed, covered or offscreen elements are listed without placing a misleading pin elsewhere. Capturing only the visible page still includes the original images of points elsewhere on the page.

Suggest a text edit: during review, select visible page text and choose Suggest edit beside it. Check the exact Original text, enter a Suggested replacement, and choose Save suggestion. An empty replacement requests deletion. Inputs, editable regions and hidden/private content are excluded. Suggestions are local drafts and never change the website. You can edit the replacement again before sending; the thread and agent tools retain both versions.

Ordinary page review temporarily enables native text selection on pages that block it with CSS or document selection handlers. Drag across the copy or double-click a word, then choose Suggest edit. Site controls and editable/private fields keep their own interactions. Leaving review or entering recording mode restores the site's selection behavior. Same-origin embedded pages participate when Chrome permits access; inaccessible embedded content remains outside the review.

In the screenshot editor and each thread image, Show/Hide points, Show/Hide element outline, and Show/Hide text selection control independent evidence layers. Points and text selection start visible; element outlines start hidden. These controls change the view, while the saved geometry and suggestion remain available. Hover outlines on the website remain temporary targeting aids. Older screenshots with baked pins cannot hide those pixels.

The editor keeps Send & Return and Send & Open beside Discard draft in the top bar; both actions are also below the comment. Send & Return immediately focuses the reviewed website while this capture tab continues sending, then closes the capture tab after all images and selected diagnostics are shared. Send & Open keeps progress visible here and opens the completed thread in this same tab. Neither requires a confirmation screen. Keep the capture tab open until sending finishes. On failure, background sending brings this tab back with the draft and remaining uploads ready for retry. If the website tab has closed, choose Send & Open. While sending, each button's fill follows the percentage of images confirmed by the server, not transferred network bytes. Reduced-motion preferences disable the fill transition. Retry resumes the same published thread. In the editor, each point has View original image, and its image is labeled point-001-original.webp, with its original viewport and capture time. Review, annotate, redact or remove these images before Send. A failed original capture is labeled No original image; the point text remains available. Point originals remain separate from full-page sections and are excluded from the combined continuous-page image. Each point's text and element context remain linked in the resulting thread, API and agent tools. If the element changes before capture, check the saved point and the screenshot before sending.

When an authorized teammate opens the same page in review mode, published pins appear on confidently matched elements. Open reviews refresh while the page is visible. Hover or focus a pin to read its note, author and relative time, then open its Feedbacks thread. Members with resolve permission can choose Resolve point there without closing other points; a resolved pin is hidden by default and can be shown with Show resolved. Older paired extension tokens may lack the point-status scope and open the thread to resolve there instead; reconnect the extension to enable direct resolution after the server is updated. A marker hides when a website menu covers its target, and reappears when the menu closes. In the Feedbacks thread, Review on the page shows numbered points over the approved screenshot, their complete notes, and a screenshot selector. Tap a point or choose Show on screenshot on mobile. Pencil, arrow, rectangle and text marks remain visible in the approved image; the thread also summarizes their types.

The public Store installation and a first source installation start without a server address. Existing installations retain their saved server address and account connections. For local development, enter a loopback address such as http://localhost:3000 and enable Advanced → Allow local HTTP development server before connecting. Use HTTPS for shared team servers.

Private preset build

For a separate internal distribution, set FEEDBACKS_EXTENSION_DEFAULT_SERVER=https://feedback.example.com when running npm run build:extension. The build produces feedbacks-extension-internal-VERSION.zip with an Internal label and that server preset. It does not overwrite the public server download ZIP or its release metadata. Saved connections take precedence over the preset. Build without this variable for a public, server-neutral package; never include credentials in a preset.

Session replay and video debug context

Use Page Controls → Recording options → Session only to capture DOM replay, activity, console and network evidence without video. Record video + session captures video with that context. Review masking and request-body options before starting; collection is explicit and limited to five minutes or the local size limit. The thread provides replay and diagnostic tabs, and agents can export the same evidence. See session replay and debug bundles.

Recording audio

Settings → Recording & audio controls Include tab sound in video and Include microphone in video. Both default on; explicit saved choices stay unchanged after an update. If an older installation has either off, enable it once. Session-only recording captures DOM and diagnostics; it does not create an audio track.

On first microphone use, recording opens this Settings section in a visible extension tab so Chrome can ask for access. Choose Allow, return to the website and start recording again. With permission granted, subsequent starts stay on the website. Allow microphone or Test microphone also checks access from Settings; the test immediately stops its stream and does not save audio.

If you previously chose Block, the extension cannot force Chrome to show another prompt. Choose Chrome microphone settings (Chrome → Settings → Privacy and security → Site settings → Microphone), find the Feedbacks extension under “Not allowed”, then Allow it or remove the blocked entry. Settings displays the extension’s exact chrome-extension://…/ address so you can identify this installation. Return and choose Check microphone again. A dismissed prompt can be requested again with Allow microphone. A managed browser may need an administrator to change its policy.

Chrome permission and computer permission are separate. On macOS, allow Chrome in System Settings → Privacy & Security → Microphone. On Windows, enable microphone access for desktop apps in Settings → Privacy & security → Microphone. Select the intended input device in Chrome, check hardware mute, and restart Chrome if your computer requests it. Missing and busy devices have their own messages and can be tested again after correction.

Tab sound comes from the selected website tab, independently of microphone permission. A silent page records a silent track; one-click tab capture does not need a separate microphone grant for tab sound. If Chrome’s manual tab picker is used, enable Share tab audio. When a requested source is unavailable, recording does not silently continue without it: the page controls show the cause, Settings opens once with recovery help, and capture remains stopped. You can explicitly switch that source off and retry. Browser and OS permission dialogs require manual verification on the target computer; automated audio checks use synthetic devices.

Sources: Chrome’s microphone permissions and recovery guide, Chrome recording APIs, and Chrome Extensions DevRel on microphone prompts in visible extension pages, checked 30 September 2026.

Audio Settings after a successful test uses synthetic permission state and shows a saved microphone opt-out. Browser checks verify this section in light/dark themes at 1100 px and 390 px without horizontal overflow. Separate packaged-extension checks use Chrome’s actual permission API with its fake microphone device; they do not verify a physical device or the OS permission dialog.

Permissions and data

activeTab and scripting support user-initiated review; storage keeps connections and drafts in trusted extension contexts; contextMenus provides the review action; alarms polls pending pairing and enforces recording limits. debugger enables detailed console and network capture only during an explicitly started recording; Chrome displays its debugger indicator. Reconnect old paired devices to receive the new recording-upload scope. Pairing requests access only to the server you entered. The popup's Chrome access rows show whether server access is allowed, review and capture are ready on the current tab, and optional all-site access is granted. If server access was revoked, choose Allow server access there. Allow access on all websites is a separate one-time Chrome grant; granting it does not start review on a page. Open the Feedbacks popup to start review, and choose Exit review at the top to stop it. A green current-tab row means review started using the temporary activeTab grant; there is no separate persistent screenshot permission. Full-page capture errors are independent of Chrome permissions.

After connecting, the popup shows Capture view, icon controls for full-page capture, Page Controls and preview sizes, plus one Record video + session action. Icon help appears immediately on pointer hover or keyboard focus in the popup, Page Controls and review screens. Team feedback, More review tools and Keyboard shortcuts expand when needed; Settings keeps server/account setup. The toolbar popup closes when the pointer leaves; an ordinary tab showing the popup page does not. While review is on and you are not typing, press a single key: M mobile, T tablet, D desktop, W reset width, S screenshot, P full page, or R stop review. Inputs, text areas, editable regions, editor controls and composing text ignore these keys. Modified key combinations are left to the website/browser.

After Capture view or Full page, a small status on the source page names the current work: preparing, collecting available network and console data, streaming DOM and page diagnostics, capturing screenshot pixels, processing, and saving the local draft. Only stages that run are shown. The status briefly disappears for each screenshot so it is absent from the captured image, then returns for processing. It is removed when capture finishes or stops. Large pages can take several seconds; keep the original tab active until the review opens.

To open Feedbacks from outside review, the suggested Chrome shortcut is Command+Shift+Y on Mac or Ctrl+Shift+Y elsewhere. Chrome's commands API requires a modifier for extension opening shortcuts; bare review keys run only on the active reviewed page. The popup shows the actual assigned opening shortcut. Change it at chrome://extensions/shortcuts if Chrome or your operating system reserves it.

Captured pixels, selected element context and comments are sent to the chosen server when you submit. Review screenshots before sending; masking is not a guarantee that all sensitive information was detected. Tokens and pending drafts are stored locally. Server access is scoped to your project grants.

Record video starts capture on the current website tab with one click. The recorder stays in the background until Stop opens the review. Use Pause video, Resume video and Stop video in the page controls. The recorder also has these controls. Paused time does not count toward the five-minute recording limit. Stop opens the recorder for review. Leaving review stops recording; closing the recorder clears its local clip. Navigation in the selected tab continues recording, with the starting page and project retained as feedback context. Tab sound and Microphone default on in Settings → Recording & audio; saved on/off choices are preserved. Microphone permission is separate from these defaults. First use opens the visible Settings page for Chrome’s prompt; allow access, return to the website, and start recording again. Future starts stay on the website while permission remains granted. Unavailable selected audio stops the start with a recovery message. See recording audio. The recorder uses WebM VP9/Opus with VP8 fallback, targets 1 Mbps video and 64 Kbps audio at up to 24 fps, and discards recordings above 40 MiB. Preview the clip and write a comment, then choose Send & Return to return to the website immediately or Send & Open to wait here and open the finished thread in this tab. Background sending closes the capture tab only after video, selected diagnostics and saved frames are shared; a failure returns to this tab for retry. Keep this tab open until sending finishes. Discard recording or closing the recorder tab clears the local clip. A submitted video follows the server operator's project retention and backup policy; there is no automatic deletion timer. If upload fails after draft feedback creation, keep the review tab open and retry Send video; the draft link does not prove the diagnostics or screenshot points were uploaded. Server access is checked again when a teammate opens the attachment. Video cannot be redacted by the screenshot tools, so review the whole clip for sensitive information before sending.

Trim video shows a thumbnail timeline directly below the preview. Drag either end to trim, scrub to inspect frames, and use Play selection to play only the selected range. Handles support arrow keys (Shift for larger steps) and Home/End. Start/end seconds are under Precise adjustments; Crop frame has drag selection and percentage fields. Changes must be applied before sending. Export runs at playback speed; Cancel keeps the last preview and Restore original undoes edits. Keep the recorder tab open. Page Controls offer Highlight (normally on) and Click indicators. Recording starts with highlighting off and navigation allowed; stopping restores that document's pre-recording settings. Click indicators default on while recording. For a workflow that redirects to another website, add its exact origin under Capture across redirect sites, authorize those sites, and then start recording. The original project remains the destination. Unknown origins suspend diagnostics with a visible gap; Pause/Stop remains available in the recorder. Chrome desktop is the target on macOS, Windows and Ubuntu. Native permission prompts and audio devices still need OS-specific verification.

Recorder preview visual checks: desktop, 1440 × 1200 and mobile, 390 × 1100. These show the actual recorder HTML and CSS in a local browser with a synthetic project, comment and generated test video. The preview state was set by the fixture. These screenshots verify layout at those sizes; they do not verify Chrome's tab picker, recording permission or upload flow.

Capture this page takes the visible browser area. Capture full page scrolls the top-level page and saves a sequence of separate screenshots, then restores your scroll position. Each screenshot is named in order, such as full-page-001-of-043.webp, and adjacent captures are cropped to avoid repeated overlap. In the editor, choose a numbered section beside the image to annotate or redact it. Choose Full page preview to see all retained sections as one scrollable image before sending; return to a numbered section to change its marks. The preview is made locally and does not upload anything. If you select Also include one combined image, those same marks appear in the combined copy; you do not need to annotate it a second time. Long combined copies are scaled to the server's decoder limits, while numbered screenshots stay at their captured resolution. The screenshots stay as local browser Blobs until Send; Feedbacks uploads them in order as individual private attachments. The editor shows the number and percentage of images confirmed uploaded by the server; it does not claim to measure network bytes. If a send fails, Retry Send resumes the same draft and thread from the first unsent image. The thread groups numbered screenshots in a compact gallery; open any image for its full resolution. Agents can request each asset individually rather than decode one oversized combined image. There is no arbitrary screenshot-count cap, but browser storage, page stability and the server's 10 MiB per-image limit still apply. A page that moves or a capture that fails keeps the pages captured so far for review and allows a retry on the original tab. The full-page option captures cross-origin frames only as the browser renders them in each screenshot. Sticky elements may appear more than once, and lazy loading can shift content. Restricted browser pages cannot be captured. The extension loads no remote executable code.

If a visible Capture view PNG exceeds the local draft limit, the extension keeps the full pixel dimensions, encodes a WebP within the server's per-image limit and stores it as a private browser Blob. The editor can still annotate, redact, export and send it as one screenshot. If even the compressed image exceeds the limit, the draft retains page context and offers a retry.

Each screenshot asset records its filename, page range and normalized point/drawing locations. Each point records its original viewport and capture time in its anchor. The original screenshot links to that point through markings[].annotationId, including its point and selected-element rectangle; later page screenshots only mark points whose elements were available in that view. The optional combined asset also records each retained source section's page range and position within the merged image, so removed sections are not mistaken for continuous page content. Agents should use the separate numbered images for readable visual detail and the metadata to connect a mark to its comment.

For captures sent by the current extension, point locations are kept in metadata and drawn separately in the thread view. Extension Settings → Review defaults → Screenshot marker saves your default marker and size. The hover Page Controls → Screenshot marker choices apply to this review: No marker, outline circle, solid circle, arrow or numbered pin, with small, medium and large sizes. The default is a small red outline circle, so each point's own screenshot indicates the click without repeating its number or drawing a selection box. No marker still saves the point, its location and its screenshot. Screenshots with markers have their own Hide pins / Show pins control. Other approved drawing marks remain in the image. Older captures have pins baked into their image pixels; the thread labels these and cannot reveal pixels that were covered during capture.

Local screenshot tools and exports

Full-page preview stacks annotated sections at their original resolution. It shows Edit section to return to the markup tools; choosing another section also opens it for editing. Copy and Download in this view export the full page. Fit width, 100% and 200% control the view; the preview no longer uses the reduced combined upload image. Numbered sections remain editable.

Use the adjacent Copy image and Download controls to export the selected screenshot. When a full-page capture has multiple sections, choose This screenshot or Full page beside those controls. Copy image copies annotated PNG pixels. Download offers PNG, JPEG, WebP and PDF. PDF keeps one captured section per page. Raster downloads retain original resolution within browser dimension and memory limits; oversized pages show an error with PDF or individual screenshots as alternatives instead of silently reducing quality. Clipboard access depends on the browser; PNG download remains available if access is denied.

Alongside pencil, arrow, rectangle, text and permanent redaction, the editor offers highlighter, numbered steps, blur, simple stamps and local PNG/JPEG/WebP images. Select an inserted image to move it or resize its corner. These marks are flattened into approved screenshots and exports; point, element and text-selection evidence follow the visibility controls in local exports and remain separate from images sent to the thread. Asset metadata records tool types and normalized bounds for the platform and agent tools; imported image source bytes stay out of that metadata. Update the server with this extension release before submitting the new tools. Blur is reversible visual softening; use Redact for private pixels. Crop export crops only local Copy and downloads of the current screenshot. The complete screenshot and its point coordinates stay intact when sending feedback.

Release and updates

The build produces a versioned ZIP, SHA-256 checksum and server download metadata. Unpacked installations require a manual update/reload. Server update checks offer an archive; they never execute a downloaded update automatically. Chrome-managed installations use their configured update channel and do not show manual ZIP update notices. Chrome Web Store publication is a separate release step and is not implied by the ZIP build.

Before a release, check pairing, capture, redaction, submit/retry, project routing, existing connection persistence, origin rejection, permission denial and update notices in Chrome. Include license/notice files in the archive. See releasing.

Screenshot diagnostic evidence

Every new screenshot review captures the available page DOM, live form and storage values, page environment and performance context at capture time. Open shadow roots and same-origin frames are included where Chrome permits access. A page over 5 MiB is stored in ordered, hashed chunks; the extension does not place it in the ordinary feedback message. The local and server artifact limit is 256 MiB, with at most 2 MiB per upload chunk. Coverage identifies unavailable, partial and stopped channels.

For console and network history, choose Start diagnostics in the page controls before reproducing the issue. The extension then records Chrome debugger events and available request and response data until the diagnostic DOM snapshot starts, navigation, review exit or the capture time limit. A simultaneous session recording and screenshot trace share the Chrome debugger attachment; stopping either capture leaves the other attached. Screenshot pixels are captured after the DOM snapshot, so coverage marks that intervening trace gap. A screenshot taken without starting diagnostics still has the point-in-time page snapshot, but cannot reconstruct earlier requests or console messages. Chrome may withhold cross-origin frames, workers or individual bodies; inspect the coverage before treating an absent event as evidence.

The screenshot editor shows one Include captured diagnostics with this review checkbox. It starts checked for new drafts; Settings → Review defaults changes that starting value for future drafts. The choice on an open draft is retained. The review panel shows size, counts, coverage and a short text preview. Download diagnostics exports the complete locally captured .tar.gz archive with a manifest. Extracted DOM files use a .html.txt suffix so opening them displays source text. Large local downloads stream through browser file saving or private disk-backed browser storage; temporary private archives are removed after an hour, including after a browser restart. Browsers without either capability report that the large archive cannot be saved. If the checkbox is unchecked, no raw diagnostic chunks are uploaded. A failed upload retains the local draft and pending thread for Retry Send; a successful send or discard removes the local chunks. The private thread also offers bounded previews and an archive download.

New screenshot artifacts retain the raw values Chrome exposes, including URLs, headers, cookies, console arguments, request and response bodies, and storage values. No automatic masking is applied to them. Captured page text and events are untrusted data; do not execute it or replay requests. Older screenshot threads retain their small, masked console/network packet and its per-entry controls. New screenshot artifacts use separate storage and permissions from session recordings.

Screenshot and video review offer Priority and Assign to before sending. Normal and Unassigned are the defaults. Open Assign to for an internal search field and member list; use More or Previous for pages of at most ten members, or arrow keys and Enter. Choose Unassigned in the list to clear selection; Escape or clicking outside closes the dropdown. Capture assigns one responsible developer to the whole task. Changing a screenshot's project clears its assignee. Only active writable project members and organization owners are eligible; a stale or revoked selection fails on the server rather than creating partially assigned feedback. The screenshot draft retains the selection after reopening, and sending freezes priority/assignment for exact retries. Assignment records responsibility without starting another person's agent or creating a GitHub issue.

These controls require an updated server and newly approved extension pairing. Older connections retain their existing permissions and show reconnect guidance; default captures still work. Member search returns a bounded name/id list without profiles or email. Initial priority and assignment are committed with the thread, and retry does not duplicate either. See the capture triage plan for checks.

Category and comma-separated tags are optional in the capture editor. See review workflow for navigation, saved filters and screenshot comparisons.

Optional page QA scan

While reviewing a project, choose Scan page for QA findings in the extension popup. This is a one-time scan of the current page, not a background monitor. It lists up to ten images without an alt attribute and checks up to twelve distinct same-origin links with HTTP HEAD requests, three at a time. Only 404 and 410 responses are reported as broken. Unsupported HEAD responses, timeouts, redirects and cross-origin links are unknown, not passing results. The requests use the current site's credentials and do not contact Feedbacks until you choose Send.

When findings exist, the extension opens an ordinary screenshot draft with a prefilled report. Inspect the report, image and any private URL paths before sending; edit or discard the draft as needed. No issue or feedback thread is created automatically. The scan runs only on the active page and selected project after an explicit button press.

Settings and unobtrusive page controls

Open Settings beside the version in the popup, or choose Options from Chrome’s extension menu. The full settings page contains server/account connection, recent server addresses, website permissions, review defaults, shortcuts and update help. Review defaults configures navigation locking, hover highlighting, click indicators, pin visibility and resolved points for new reviews. Separate Recording defaults configures navigation, highlighting and click indicators during recording. Page controls override the current session; recording restores the earlier review controls when it ends. Defaults never start review or diagnostics automatically. Start / resume review shows the shortcut actually assigned by Chrome; Change shortcut opens Chrome’s shortcut controls. Navigation locked in the page tools prevents links, forms and cancellable page navigation from leaving the review. Menu toggles and in-page anchors remain usable. Select Navigation allowed to follow links again. Chrome's address bar and browser controls remain available.

The default start/resume shortcut is Command+Shift+Y on macOS and Ctrl+Shift+Y on Windows/Linux. It opens the extension popup, which starts review on the active website when connected; otherwise it shows connection setup. Settings reports an unassigned shortcut rather than a guessed default if Chrome could not bind it or the user cleared it. No combination is guaranteed free of other extensions or OS shortcuts. The existing _execute_action command keeps user remappings and does not add page listeners or permissions. See Chrome’s Commands API for binding and conflict behavior (checked 2026-09-28). The simple review keys run only during review. Disable Use single-key shortcuts during review when you prefer buttons; open reviews update immediately.

Drag the dock’s grip to move it, or focus the grip and use arrow keys. Hide removes it until you open the Chrome extension again. Exit stops review. Draft points stay on the same page for when you resume. The primary capture is the visible view with each point’s saved original image. Full-page capture (P) and Record video + session are directly available in the popup and Page Controls. They remain optional: additional images and long recordings may increase an agent’s processing and token use. The panel shows comment count without raw screen dimensions.

Page comments opens threads for the current page across screen sizes on the configured server. Expand Review tools for navigation, highlighting, click indicators, pin and resolved visibility, screenshot marker style and size, diagnostics, and preview sizes. Diagnostics are collected locally after an explicit click; stop discards them. Capture within five minutes to review selected entries before sharing.

The popup’s collapsed Team feedback summary shows the thread count. Expand it for project totals and filters for the exact page, its hostname, or the page at the current mobile/tablet/desktop size. Open the link to see matching threads on your configured server. Individual point resolution needs the updated server and an extension token with threads.annotationStatus; older connections can use the web thread or reconnect.

The Page Controls icon has a dotted drag grip. Drag it to move the controls, or focus the icon and use arrow keys. Its tooltip and accessible description explain the movement keys; hiding controls keeps review active.