[teamleaderleo]

Leo-ask: ask-12231fbe5e

Leo: "idk where things are man i dont see anything other than the current workspace which is you and you have a cmux next github brwoser tab open (we need to have cmux expose this harder if it isn't already/hve better instructions)"

What happened: an agent needed Leo to confirm GitHub sudo in a browser tab it had opened (in a workspace 42 rows down the sidebar) and told him a raw ws_… id he could not see or search.

This is slice A of the "needs you here" surface: the daemon state, CLI and MCP path. The app styling, notification with Show, the "Go to what needs me" palette command and the agent instructions follow in their own PRs.

What changes

• tab.update takes attention: a one-line reason (1-200 characters, no control characters) or null to clear. The daemon stores {reason, at_ms} on the tab record (tab_state.attention_json, an additive column), and every client sees it as the tab's extra.attention. Capability tab-attention-v1.
• CLI: cmux attention request --tab <tab_…> --reason <text> and cmux attention clear --tab <tab_…>. Both send tab.update {attention}, so CLI, MCP (tab_update with attention, generated from the catalog) and the app share one mutation path.
• It never moves focus. A test checks that the active surface stays the same.
• Clearing:
  • the agent's clear;
  • a browser tab whose page goes to a different URL, for both frontend-rendered (WebKit/CEF) and CDP browsers; a title-only change keeps the mark;
  • the tab's close. The mark is dropped when the tab row is tombstoned, so a reopened tab (which keeps its id) does not ask again.
  • Terminal tabs keep the mark until the agent clears it or the tab closes.
• The agent guide (cmux docs, "Show your work…") tells agents to mark the tab and to name workspaces and tabs by their titles, never by a ws_ or tab_ id.

Tests

• state/tab_attention_tests.rs:
  • mark, list, keep across another field, clear;
  • bad values are rejected;
  • focus does not move;
  • navigation clears the mark, a title change or the same URL does not;
  • close plus reopen leaves no mark.
• tab_state_store unit tests: migration, clear on close, reason validation.
• CLI: attention_request_and_clear_send_tab_update, plus the transport-path coverage cases (which check the plan against the catalog).

Not in this PR

App styling (the existing blue needs-you treatment on the tab and its workspace row), the Needs you list entry, the system notification with Show, the palette command, and the MCP server instructions text.

No capture: daemon/CLI only; no native UI changes in this PR.

Agent session: cosmic-gray

🤖 Generated with Claude Code

---

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

---

Summary by cubic

Agents can now mark the tab where the person must act (confirm, log in, approve, pick), instead of pointing them at an opaque ws_/tab_ id for a tab far down the sidebar.

• tab.update takes attention, a one-line reason or null to clear; the daemon stores it on the tab record and every client sees it as extra.attention under tab-attention-v1. Reasons with control characters, U+2028, or U+2029 are refused.
• New CLI commands cmux attention request --tab <tab_…> --reason <text> and cmux attention clear --tab <tab_…> both send tab.update — the same mutation path CLI, MCP (tab_update), and the app share.
• The mark never moves focus.
• A mark records the page it was set on, so a navigation clears only a mark set on another page, and the clear is decided in its own commit: a mark an agent sets on the new page survives a late or repeated clear for an older page. It clears on the agent's clear, a browser tab's navigation to another URL (a title-only change keeps it), or the tab's close; reopening a closed tab does not re-ask. The CDP event thread detects URL changes over target-info, frame, and same-document navigation alike and always asks on its first event.
• cmux docs now tells agents to mark the tab and to name tabs and workspaces by their titles, never by id.

App styling, the needs-you list entry, system notification with Show, and the palette command follow in later PRs.

^{Written for commit 12f9a15. Summary will update on new commits.}

[github-actions]

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

Review 12f9a15: 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=pr24286 -f ref=sp-attn-a -f platform=macos -f backend_mode=local

red: cmux-next checks (god files, concurrency, crash safety, l10n); no finished base run on 44fc3d3 to compare; do not merge until green; see lane-rules

[teamleaderleo]

Bead cx-rfo6q created for this PR by the cmux coordinator.

[blacksmith-sh]

Found 8 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmp8flduogh/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmpn1vv9q4z/summary-missing.json (live):/summary-missing.json (live):	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	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
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
test_base_passes (main.NoticesPairTests.test_base_passes)/ test_base_passes (main.NoticesPairTests.test_base_passes)	View Logs
test_targets_outside_the_app_go_to_the_mac_canary (main.SwiftCanaryFleetCoverage.te st_targets_outside_the_app_go_to_the_mac_canary)/ test_targets_outside_the_app_go_to_the_mac_canary (main.SwiftCanaryFleetCoverage.te st_targets_outside_the_app_go_to_the_mac_canary)	View Logs

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

[teamleaderleo]

Review (claude sonnet, correctness first): SHIP at a6e4474

Independent re-review by a different model (Claude Sonnet; the lane runs Opus), after the BLOCK at 78947df. Read-only checks: git diff, grep, rustfmt --check, sync-resource-api.py --check (exit 0).

Previous BLOCK items, all fixed:

1. cli/command/state/tests.rs:642 asserts .contains("not an id"), the message Selector::parse("ws_1") returns.
2. rustfmt passes for every changed .rs file, state/tabs.rs included.
3. clippy cloned_ref_to_slice_refs: fresh_upserts(..., std::slice::from_ref(&tab)).
4. CDP clear on every URL change: clear_attention_if_page_changed runs after TargetInfoChanged, FrameNavigated and NavigatedWithinDocument, comparing browser.url() with the URL the event thread last saw. A navigate-hold change made on another thread is picked up at the next of those events. New test in browser/tests/navigation_barriers.rs.
5. The bad-value test uses short idempotency keys (attention-bad-{index}).
6. The reason limit is 200 UTF-8 bytes in the store's check, the catalog (catalog_validation.rs:347 counts bytes), the spec and the help.
7. The help no longer names the palette UI, which this PR does not ship.
8. All 7 SDK descriptors carry the same catalog hash.

worker.rs: the per-thread seen_url borrows cleanly. The state and registry locks are taken one after the other, never nested, after the URL and title guards are released, so there is no lock-order risk. Screencast frames never call it. The DB read happens only when the URL differs, and the commit only when the tab holds a mark.

Non-blocking:

• The CDP test passes Weak::new(). It proves the URL tracking, not the daemon clear. The clear is covered end to end on the frontend-browser path (tab_attention_tests.rs).
• If an agent navigates a tab and marks it before the CDP navigation event lands, that event clears the mark. Agents should mark after the page loads; the skills PR (agents: name tabs by title, mark the tab the person must act in (skills, MCP instructions) #24299) says to mark the page the person must act on.
• The god-file ratchet files grow by a few lines.

[teamleaderleo]

cmux-next guards / nightly-guards / workflow-guard-tests / release-notary is red on 14 of 14 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.

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[teamleaderleo]

rust workspaces / plan (changed Cargo workspaces) is red on 14 of 15 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]

rust workspaces / rust workspaces is red on 14 of 15 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 (codex xhigh, correctness first): SHIP at 12f9a15

No findings.

Independent review by a different model (Codex gpt-6.1-sol at xhigh on cmux16, diff-only, read-only sandbox). It checked that each finding of the earlier BLOCK at a3d0d8c is fixed: a stale navigation clear now checks the current page, the SDK refuses a multi-line reason, and the worker test records each clear.

Checks at this head:

• Fleet: cmux-tui fmt, clippy and the attention tests pass (cmux-ci step a15ee8ac00f420e8883e1b99).
• Reds that fail the same way on feat-cmux-next: rust workspaces (Every Cargo workspace is listed / Every planned workspace passed), Guard status, release-notary, and cmux-next checks. The SDK rust package job fails only on rustdoc for [::1] in bindings/rust/src/generated/types.rs:2612, which is also on base; its fmt, tests and clippy pass.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.3993797697001 (7e7162114e)
• cmux NEXT nightly 1.0.0-nightly.3994065441401 (50826464b6)