[teamleaderleo]

Summary

Stacked on #20514.

The summary card's Subagents title now opens a Subagents tab beside the card's Summary tab, like the right pane in the ChatGPT desktop app (bead cx-x62ev):

• The tab lists Active · N (or No active subagents) and Done · N, newest first, nested subagents included. Waiting subagents count as active, and failed or cancelled ones count as done.
• Each row is the transcript's own SubagentListRow (avatar with status dot, name, kind, state or current action, elapsed time). In the card it uses a compact 40px row with no border.
• Clicking a row shows that subagent through the card's existing onOpen. In the app that is chat.subagentPane (opens beside the chat). When the host opens no pane, the subagent shows in this pane.
• The tab's × goes back to the summary and puts focus back on the Subagents title. The Summary and Subagents tabs switch between the two views.
• The tabs are aria-pressed toggle buttons, like the gallery filters, because ui-rules bans raw tablist/tab roles in page code.

Reuse: I did not add a second row component. The tab builds SUBAGENT_ROW rows from subagents/subagentRows.ts and renders them with SubagentListRow from subagents/SubagentGroup.tsx, through SubagentPaneContext. #20718 extended that same row with progress and a stop button. #20718 is merged on feat-cmux-next, but #20514's head does not include it yet. When #20514 takes a base merge, this tab picks up #20718's progress and stop button with no extra change. The section title still uses subagentsIn and SubagentStatusList.

Also: one-line lower of scripts/cmux-next/layers-baseline.tsv. #20514's reduced-transparency fallback leaves summary.css with 0 bare backdrop-filter, so check-layers asked for the lower baseline. Without it, bun run check fails.

Not in this PR:

• The bead's "Outputs + with the tooltip Create a file or site". There is no create action behind it yet, and a + that does nothing would be dead UI.
• Generated colourful avatars. The tab uses the same avatar as the transcript's subagent stack (the harness mark), and changing that look for everyone needs a call.

In-app stills (side by side with ChatGPT) will follow in a comment.

Testing

• Red, at db86bd4 (test only): CMUX_BROWSER_TESTS=1 bun test test/agent-pane-subagents-tab.test.ts gave 0 pass / 2 fail. The heading "Subagents · 1 running" is there, but there is no button to open the tab.
• Green at head:
  • agent-pane-subagents-tab.test.ts 2 pass
  • agent-pane-summary-panel.test.ts 7 pass
  • popup-surface 10 pass
  • agent-pane-background-tasks 2 pass
  • sources-changes 1 pass. Its first run timed out on a cold vite server and passed on rerun.
  • ui-rules 4 pass
  • subtle-elevation and popup-one-hairline 42 pass
• vp fmt, vp check and bun run check all pass.
• gen-strings.mjs --check passes, and check-l10n.sh reports 0 errors.
• Localization: 4 new keys (summary.subagents.active, .ended, .noneActive, .closeTab) in all 21 locales. The tab labels reuse summary.title and summary.subagents.

Changelog

• Added: The agent chat's summary card has a Subagents tab listing active and finished subagents; click one to open it.

🤖 Generated with Claude Code

---

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

---

Summary by cubic

Adds a Subagents tab to the agent chat's summary card so active and finished subagents can be reviewed without leaving the summary.

The Subagents section's title now opens a tab beside the card's Summary tab, mirroring the ChatGPT desktop app's right pane. The tab lists Active · N (or "No active subagents") and Done · N, newest first, nested subagents included; waiting subagents count as active, failed and cancelled as done. Rows reuse the transcript's SubagentListRow, compacted to 40px, and clicking a row opens that subagent via the card's existing onOpen—the chat's subagent pane, or this tab when the host opens none. Opening the tab drops the card's head counts since the tab lists them itself (they were cut off next to the two tabs). The close button returns to the summary and refocuses the Subagents title. Tabs are aria-pressed toggles because ui-rules bans raw tablist/tab roles in page code. Four new locale strings ship in all 21 locales, and the layers baseline is lowered one line since summary.css no longer has a bare backdrop-filter. SubagentsTab is allowlisted in gallery coverage until it gets an entry.

^{Written for commit 3d01e89. Summary will update on new commits.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Repository: manaflow-ai/cmux/.coderabbit.yaml
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: ddf8682b-37e7-48a3-b64e-32cd9c2b971d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[blacksmith-sh]

Found 5 test failures on Blacksmith runners:

Failures

Test	View Logs
GitHub-hosted runner label: .github/workflows/ intent-familiar-review.yml:21: runs-on: ubuntu-24.04	View Logs
GitHub-hosted runner label: .github/workflows/ pr-red-checks-comment.yml:16: runs-on: ubuntu-24.04	View Logs
test_scans_only_test_sources (main.CommandLine.test_scans_only_test_sources)/ test_scans_only_test_sources (main.CommandLine.test_scans_only_test_sources)	View Logs
test_scans_the_ui_test_sources (main.CommandLine.test_scans_the_ui_test_sources) (p ath='ios/cmuxUITests/ProbeTests.swift')	View Logs
two failed fetches then a success did not pass/ two failed fetches then a success did not pass	View Logs

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

[github-actions]

Probably yours: Fast static checks failed: a static check failed, FAILED localization (0.93s).

CI failed on 3d01e89ea8 (run 38982116420 attempt 1): 2 code, 6 unknown.

Job	Verdict	Why
Fast static checks	code	a static check failed
CI fast guards	unknown	no known signature; failed step: Run the fast guard tests
guards / workflow-guard-history	unknown	no known signature; failed step: Validate SwiftPM lockfile policy
guards / workflow-guard-tests / app-host-cache	unknown	no known signature; failed step: Validate iOS video recording failures
guards / workflow-guard-tests / preflight	unknown	no known signature; failed step: Validate macOS runner guards
guards / workflow-guard-tests / quality-determinism	code	a Python test failed
guards / workflow-guard-tests / release-tooling	unknown	no known signature; failed step: Validate retried Ghostty Zig package fetches
guards / workflow-guard-tests / release-ios	unknown	no known signature; failed step: Validate iOS simulator build-once contract

Matched log lines

Fast static checks: FAILED localization (0.93s)
guards / workflow-guard-tests / quality-determinism: FAIL: test_scans_only_test_sources (__main__.CommandLine.test_scans_only_test_sources)

Not re-run automatically: Fast static checks, CI fast guards, guards / workflow-guard-history, guards / workflow-guard-tests / app-host-cache, guards / workflow-guard-tests / preflight are not machine failures.

Written by scripts/ci/classify_failures.py (ci-failure-attribution.yml); signatures are its SIGNATURES table. A machine verdict is the runner's fault, not this PR's; yours means the failing file is one this PR changes, also red on main that main's latest full suite fails the same way, seen on other PRs that it failed on another pull request's run lately.

[github-actions]

CI fast guards failed on 3d01e89ea8 (https://github.com/manaflow-ai/cmux/actions/runs/38982116420). It does not block the merge; a red guard merged into main breaks it for every open PR.
The log named no failed step; see the run.

Agents: python3 scripts/ci/guard_attribution.py fix applies the mechanical fixes locally. This comment is updated in place on each push.

[teamleaderleo]

Page renders (the PR's browser test fixture in headless Chromium, dark; not yet in the app: the in-app mock chat has no subagents):

Summary tab	Subagents tab

Follow-ups before Leo's look: light-theme and in-app captures with a session that has subagents; the card head's counts truncate next to the tabs ("1 running · 1 f…"), so the tab row likely should drop them while Subagents shows (the tab lists the same counts). The 24h durations come from the fixture's start time.

[teamleaderleo]

Superseded by #21676, the Subagents tab recut onto the current summary card (#21564's inline column and popover). This branch sat on #20514's card.