[teamleaderleo]

Leo-ask: ask-49f4ca2104

Takes back part of 45fe246 ("fix(security): refuse unguarded app screenshots", lawrencecchen, 2026-10-20): it made window.screenshot, pane.screenshot and tab.screenshot person-only, so an agent had no way to see a terminal or a chat window. Leo: "cmux only lets a person run the window screenshot action. doesnt seem right". This PR keeps that fix's guarantee: no unmasked browser pixels reach automation.

Summary

• An automation run (CLI, MCP, scripts) of cmux window|pane|tab screenshot runs again. A person's run (palette, menu) is unchanged.
• What an automation image hides (ScreenshotPrivacyMask): these areas are painted opaque black, because the app can't reproduce the browser host's secret mask (vault, TOTP, typed secrets):
  • every pane showing a browser tab (the whole browser view, toolbar included);
  • every WKWebView not held by the app's own agent chat (AgentPaneView) or page tabs (InternalPageView), so an unlisted web view fails closed;
  • every Chromium page child window.
• What it still shows: the app's own React pages on Chromium are painted back. The overlay panels and the app's other windows are left out.
• Refused for automation: a browser tab, or a pane showing one. The refusal points at the browser REPL's page.screenshot(), which masks secrets.
• The cmux and cmux-browser skills now say what an agent's screenshot shows.

Risk

risk:high: this changes a security policy (a teammate's fail-closed screenshot gate). It takes back part of a commit pushed straight to feat-cmux-next, which has no PR number.

Changelog

Fixed: Agents can screenshot cmux windows, panes and terminal tabs again; browser pages in the image are blacked out.

Testing

Test first:

• ActionOriginTests/appScreenshotsRunForEveryCaller: CLI, MCP, script and no-origin runs reach the handler as non-person runs.
• ScreenshotPrivacyMaskTests:
  • a foreign web view is masked, including when nested;
  • the app's page view is not masked;
  • a hidden web view is not masked;
  • the mask layer paints opaque black.

Observed / Not run

• Review: an independent review of 238d980 failed on three points, all fixed there:
  • an app's user-origin run (during a person's gesture) was unmasked;
  • only direct Chromium child windows were masked;
  • the masks were measured after the capture.
    A re-review of 238d980 passed (SHIP). A review of the follow-up 91e026a, where a masked capture whose masks cannot be painted now throws instead of sending the unmasked image, also passed (SHIP).
• Fleet app compile: passed at 238d980 (job 04bdb0f38d2c, 80 s) and at the head 91e026a (job 7a50e9fb5e1e, 86 s).
• Red and green test runs: package-test-lane ActionOriginTests,ScreenshotPrivacyMaskTests on scratch refs (the red tests, then the fix, each with ci: the agent pane page ships the Inspector's ContextSection.css (base red since #23789) #23935's build-script patch for the base ContextSection.css red). Queued on the fleet; results to be posted as a comment.
• PR reds that come from base: cmux-next crash-class rules (an objc-selector probe on CEFFileDragRelay.swift, which cmux NEXT browser: files dragged in from Finder drop into the Chromium page #23790 added) and the Checks job structure step of cmux-next checks. This PR touches neither.
• Not run: a live capture of a window with a browser and a terminal side by side; no test covers nested Chromium child windows.
• Follow-ups from the review, not in this PR: a resize during the capture can shift the masks; the Passwords page (sites and usernames) is not masked; the web-view scan starts at contentView.
• No capture: this changes what automation receives, not the app's UI.

Agent session: cosmic-gray

🤖 Generated with Claude Code

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 91e026a: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr23983 -f ref=cmux/window-screenshot-agents -f platform=macos -f backend_mode=local

red: cmux-next checks (god files, concurrency, crash safety, l10n); NEW: red check: Checks job structure is not red on the base 76db10b; do not merge until green; see lane-rules

[teamleaderleo]

Owned by cmux-mcp-surface on cosmic-gray · session

[blacksmith-sh]

Found 16 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmp0f_o65wo/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmpvcbfbxir/summary-missing.json (live):/summary-missing.json (live):	View Logs
[new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows/ [new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows	View Logs
[new_workspaces] the window draws no frames/[new_workspaces] the window draws no frames	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
cmux-next-ios-kit.yml runs macOS jobs on pull requests but a new push does not cancel t he previous run; key the concurrency group on the pull request and set cancel-in-progre ss for its pull request events/ cmux-next-ios-kit.yml runs macOS jobs on pull requests but a new push does not cancel t he previous run; key the concurrency group on the pull request and set cancel-in-progre ss for its pull request events	View Logs
convention violations found (ERROR lines above)./ convention violations found (ERROR lines above).	View Logs
nightly cmux-tui resolve/nightly cmux-tui resolve	View Logs
test_every_submodule_free_checkout_on_an_owned_runner_resets_submodules (main.Reuse dWorkspaceSubmodules.test_every_submodule_free_checkout_on_an_owned_runner_resets_submo dules)/ test_every_submodule_free_checkout_on_an_owned_runner_resets_submodules (main.Reuse dWorkspaceSubmodules.test_every_submodule_free_checkout_on_an_owned_runner_resets_submo dules)	View Logs

...and 6 more test failures. View all on Blacksmith: run 39711313842, run 39711313898

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[teamleaderleo]

cmux-next crash-class rules / cmux-next crash-class rules is red on 11 of 11 fresh feat-cmux-next PR heads in the last 2:00:00, so it is most likely a base red, not this PR's. The base-reds lane has it; a base merge picks up the fix once it lands.

[teamleaderleo]

Review (subagent, correctness first): SHIP at 91e026a

• Full re-review of 238d980: SHIP. The three earlier blockers are fixed: an app's user-origin run, nested page windows, and masks measured before the capture. It found no path that gives a non-person caller an unmasked image.
• Delta review of 91e026a (a masked capture whose masks cannot be painted throws WindowScreenshotFailure.blank instead of returning the unmasked image): SHIP.
• Fleet app compile at 91e026a: passed (job 7a50e9fb5e1e, 86 s).
• Both reviews were by reading; nothing was built or run.

[teamleaderleo]

Bead cx-2dfdf created for this PR by the cmux coordinator.

[teamleaderleo]

Red/green on the fleet (scripts/ci/package-test-lane.sh suite Packages/macOS/CmuxNext …; both scratch refs add the #23935 base-red patch so the package builds):

• Red: the failing test at f65a9ec, job 6171d5b254cad839d43ce7bf, ScreenshotPrivacyMaskTests: onlyAPersonsOwnRunIsUnmasked fails with 3 issues (a .user run carrying callerApp, callerAgentSession or callerTerminal came out unmasked). The other 4 tests pass.
• Green: the fix at 65b122a (the 238d980 tree), job 83aa0d75a687e7081c19f2e8, ActionOriginTests,ScreenshotPrivacyMaskTests: 11/11 and 5/5 pass.

Observed / Not run

• The head 91e026a adds only the fail-closed throw when a masked composite can't be painted. A fleet app compile passed on the head (job 7a50e9fb5e1e, 86 s). No test reaches that branch: the suites above ran on the 238d980 tree, not the head.
• I didn't do a live capture in a running app.