Skip to content

🙌 OpenHands: AI-Driven Development

docs(verify-openhands): map maintenance 2026-10-14 (#18437)

Engel Nyste4775b4ba61a
diff --git a/.agents/skills/verify-openhands/references/feature-map/F02-app-shell.md b/.agents/skills/verify-openhands/references/feature-map/F02-app-shell.md
index 108664a5cc..6b1aa024b9 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F02-app-shell.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F02-app-shell.md
@@ -33,6 +33,7 @@ Source: `src/components/features/sidebar/`, `src/components/features/command-men
- `F02.checklist-progress`: items get a check and a strikethrough when done. The LLM item completes when a profile is configured, Start your first chat when a conversation exists, Customize your agent after a visit to `/settings/agents`, and Join Slack on click.
- `F02.checklist-preview`: hovering an item shows a preview with a description, a **Documentation** link and an action button that navigates.
- `F02.checklist-minimize`: the card header minimizes and expands the card, and the state survives a reload.
+- `F02.checklist-dismiss`: hovering or focusing the card (expanded or minimized) reveals an X (`Dismiss checklist`) in its header; touch devices always show it. Clicking it hides the card for good (it stays hidden after a reload) and turns off the **Show Getting Started checklist** switch in Settings → Application, which brings the card back in its previous expanded or minimized state.
- `F02.checklist-hide`: the **Show Getting Started checklist** switch in Settings → Application hides the card (it stays hidden after a reload) and can show it again.
- `F02.checklist-all-complete`: the card disappears once all six items are complete. Completion is recomputed from current data, so deleting the automation or the MCP server brings the card back.
- `F02.super-admin-setup-guide`: on an OpenHands Enterprise (cloud) backend with `ENABLE_SUPER_ADMIN`, the first Super Admin (`/me` permissions include `manage_super_admins`) sees a floating **Setup guide** panel in the lower-right corner with `n/4` progress, the steps Add an LLM, Choose an automation template, Add an integration, Invite users and Optional: SAML / instance, a `Next: <step>` status and a **Start** link for that step (the positive Enterprise path remains unverified here). Close collapses it to a pill that reopens it. While the guide loads or shows, the **Getting started** card is hidden. Local backends never show it and make no `/api/admin/setup-state` request (#17969).
@@ -49,7 +50,7 @@ Source: `src/components/features/sidebar/`, `src/components/features/command-men
- Command menu: press `Control+k` or `Meta+k` anywhere, or click **Search commands** (`command-menu-trigger`) in the rail or drawer. Type a page or setting name, then press Enter or click the row.
- Hover a sidebar row to reveal its pin. Hover the collapsed rail's logo to reveal the expand chevron.
- Settings → Application (`/settings/app`): the **Show Getting Started checklist** switch.
-- The **Getting started** card: click an item row to go to its page, or hover it for the preview.
+- The **Getting started** card: click an item row to go to its page, or hover it for the preview; hover the card for the X that dismisses it.
- Keyboard: `Control+k` also works while the composer has focus; Tab moves from the search input into the option list.
## Driving it with control-openhands
@@ -82,12 +83,13 @@ Preconditions:
- **Progress (`F02.checklist-progress`).** Each completed item renders a struck-through label, so `control-openhands browser count 'testid=sidebar-onboarding-checklist-item-<id> >> css=span.line-through'` is `1`. Run `control-openhands browser goto /settings/agents`, then come back; `customize-agent` is done. Run `control-openhands browser click 'testid=sidebar-onboarding-checklist-item-join-slack'`; `join-slack` is done and `browser tabs` shows an external Slack tab, which you close with `control-openhands browser close-tab 1`. After `llm preset deepseek`, run `control-openhands browser reload` (the open page does not refetch settings written through the API), then `control-openhands browser wait 'testid=sidebar-onboarding-checklist-item-configure-llm >> css=span.line-through' --timeout 15000`. After `conversation start`, `start-conversation` is done. The header count rises each time (`4 complete` after these four).
- **Item preview (`F02.checklist-preview`).** Run `control-openhands browser hover 'testid=sidebar-onboarding-checklist-item-join-slack'`, `control-openhands browser hover 'testid=sidebar-onboarding-checklist-item-connect-mcp'` and `control-openhands browser wait 'testid=sidebar-onboarding-checklist-preview-connect-mcp' --timeout 5000`. Then run `control-openhands browser text 'testid=sidebar-onboarding-checklist-preview-connect-mcp'`; it reads the title, the description, `Documentation` and `Connect`. `control-openhands browser attr 'testid=sidebar-onboarding-checklist-preview-docs-connect-mcp' href` is `https://docs.openhands.dev/overview/model-context-protocol`. Take `control-openhands browser screenshot --feature F02.checklist-preview --name connect-mcp`, then run `control-openhands browser click 'testid=sidebar-onboarding-checklist-preview-action-connect-mcp'` and `control-openhands browser wait-url '/mcp$'`.
- **Minimize (`F02.checklist-minimize`).** Run `control-openhands browser click 'testid=sidebar-onboarding-checklist-toggle'`, `control-openhands browser attr 'testid=sidebar-onboarding-checklist' data-minimized` (`true`) and `browser count 'testid=sidebar-onboarding-checklist >> role=listitem'` (`0`). After `control-openhands browser reload` the attribute is still `true` and the toggle's `aria-label` is `Expand checklist`. Click the toggle again to restore `false`.
+- **Dismiss (`F02.checklist-dismiss`).** On `/conversations`, `control-openhands browser count 'role=button[name="Dismiss checklist"]'` is `0` (the X is `opacity-0` and `invisible` until the card is hovered). Run `control-openhands browser hover 'testid=command-menu-trigger'`, `control-openhands browser hover 'testid=sidebar-onboarding-checklist-toggle'` and `control-openhands browser wait 'role=button[name="Dismiss checklist"]' --state visible --timeout 5000` (`browser screenshot 'aside[data-collapsed]' --feature F02.checklist-dismiss --name hover-close`). Run `control-openhands browser click 'role=button[name="Dismiss checklist"]'`: `browser count 'testid=sidebar-onboarding-checklist'` is `0`, still `0` after `control-openhands browser reload`, and `browser storage` lists `openhands-sidebar-onboarding-checklist-dismissed`. On `/settings/app`, `control-openhands browser eval "document.querySelector('[data-testid=show-getting-started-checklist-switch]').checked"` is `false`; `control-openhands browser click 'text=Show Getting Started checklist'` brings the card back. Minimized: `control-openhands browser click 'testid=sidebar-onboarding-checklist-toggle'` (`data-minimized` `true`), hover as above, click the X (count `0`), click the switch text again: the card returns with `data-minimized` `true`; click the toggle to expand it. At phone width the X shows in the drawer's card the same way.
- **Hide from settings (`F02.checklist-hide`).** Run `control-openhands browser goto /settings/app` and `control-openhands browser click 'text=Show Getting Started checklist'`; `browser count 'testid=sidebar-onboarding-checklist'` is `0`. After `control-openhands browser reload` the count is still `0`, and `control-openhands browser eval "document.querySelector('[data-testid=show-getting-started-checklist-switch]').checked"` is `false`. Click the same text again; the count is `1`.
- **All complete (`F02.checklist-all-complete`).** Start with the four items above done (`5 complete` would mean one fixture is left over). Arrange two fixtures: a minimal stdio MCP server, `control-openhands fixture file --name qa-f02-mcp.py --content "$(printf '%s\n' 'import sys,json' 'for line in sys.stdin:' ' m=json.loads(line)' ' if "id" not in m: continue' ' r={}' ' if m.get("method")=="initialize": r={"protocolVersion":m["params"].get("protocolVersion","2024-11-05"),"capabilities":{"tools":{}},"serverInfo":{"name":"qa","version":"1"}}' ' elif m.get("method")=="tools/list": r={"tools":[]}' ' sys.stdout.write(json.dumps({"jsonrpc":"2.0","id":m["id"],"result":r})+"\n"); sys.stdout.flush()')"`, and `control-openhands fixture file --name qa-f02.automation.json --content '{"version":1,"kind":"automation","spec":{"name":"QA_F02 Done","trigger":{"type":"cron","schedule":"0 0 1 1 *","timezone":"UTC"},"enabled":false,"prompt":"Reply with the single word: pong. Do not run any tools."}}'`. Add the server in the UI: `control-openhands browser goto /mcp`, `control-openhands browser click 'testid=mcp-add-custom-server'`, `control-openhands browser click 'testid=server-type-dropdown'`, `control-openhands browser click 'role=option[name="STDIO"]'`, `control-openhands browser fill 'testid=name-input' qa_f02_mcp`, `control-openhands browser fill 'testid=command-input' python3`, `control-openhands browser fill 'testid=args-input' "$OH_VERIFY_RUN/evidence/_fixtures/qa-f02-mcp.py"`, `control-openhands browser click 'testid=mcp-custom-editor >> testid=submit-button'` and `control-openhands browser wait 'testid=mcp-custom-editor' --state detached`. The header reads `5 complete`. Import the automation: `control-openhands browser goto /automations`, `control-openhands browser click 'testid=automations-add-automation'`, `control-openhands browser click 'testid=automations-import-automation'`, `control-openhands browser upload 'testid=automations-import-file' "$OH_VERIFY_RUN/evidence/_fixtures/qa-f02.automation.json"`, `control-openhands browser click 'testid=import-automation-confirm'` and `control-openhands browser wait 'testid=import-automation-modal' --state detached`. Now `browser count 'testid=sidebar-onboarding-checklist'` is `0`, still `0` after `control-openhands browser reload`, and `control-openhands browser screenshot 'aside[data-collapsed]' --feature F02.checklist-all-complete --name rail` shows the rail without the card. Restore and prove it comes back: open the card's kebab with `control-openhands browser click '[data-testid^="automation-card-"] >> has-text=QA_F02 Done >> role=button[name="Automation actions"]'`, then `control-openhands browser click 'role=list >> has-text=Export >> role=button[name="Delete"]'`, `control-openhands browser click 'role=heading[name="Delete automation"] >> xpath=.. >> role=button[name="Delete"]'` and `control-openhands browser reload`; the card count is `1` again (`5 complete`). Then `control-openhands browser goto /mcp`, `control-openhands browser click 'testid=mcp-server-detail-qa_f02_mcp'`, `control-openhands browser click 'testid=mcp-custom-editor-delete'`, `control-openhands browser click 'testid=confirmation-modal >> testid=confirm-button'` and `control-openhands browser reload`; `testid=mcp-installed-empty` is back and the card reads `4 complete`.
- **Setup guide (`F02.super-admin-setup-guide`).** Local: `control-openhands browser count 'testid=super-admin-setup-guide'` is `0` at `desktop` and `phone`, `browser count 'testid=sidebar-onboarding-checklist'` is `1`, and `control-openhands browser network --filter setup-state` reports `total` `0`. The positive path is blocked locally: it needs an OHE backend with `ENABLE_SUPER_ADMIN` whose `/api/admin/setup-state` returns a `guide_org_id`, `guide_dismissed` `false` and at least one unfinished required step, signed in as the first Super Admin.
- **Setup tour (`F02.super-admin-setup-tour`).** Local, at `desktop` and `phone`: run `control-openhands browser network --clear`, `control-openhands browser goto '/?setup_tour=first-automation'`, wait about 6 s, then `control-openhands browser count '.oh-setup-tour-popover'` and `browser count 'testid=super-admin-setup-guide'` are `0`, `browser count 'testid=home-screen'` is `1`, `browser url` still ends in `?setup_tour=first-automation`, and `browser network --filter setup-state` reports `total` `0`. The tour itself is blocked with the same Enterprise prerequisite as the guide.
- **Update tile (`F02.update-tile`).** The tile is blocked in an up-to-date checkout or without registry access. Check with `control-openhands browser count 'testid=agent-canvas-version-tile'`. When the count is `1`, click it; the modal has `agent-canvas-update-command-npm`, `agent-canvas-update-command-docker` and `agent-canvas-update-check-button`. The manual check runs either way: on `/settings/app` (`control-openhands browser goto /settings/app`; the previous bullet ends on `/mcp`), `testid=agent-canvas-update-badge` reads `Up to date` (or `New version`), `control-openhands browser click 'testid=agent-canvas-update-toggle'` opens `agent-canvas-update-modal`, and `control-openhands browser click 'testid=agent-canvas-update-check-button'` reports the result in `agent-canvas-update-status` (`You're running the latest version. What's new` when current). Close it with `control-openhands browser click 'testid=close-agent-canvas-update-modal'`.
-- **Error toast (`F02.error-toasts`).** Run `control-openhands browser goto /conversations/00000000-0000-0000-0000-000000000001` and `control-openhands browser wait-text 'This conversation does not exist'`. The URL becomes `/conversations`, and `control-openhands browser screenshot --feature F02.error-toasts --name missing-conversation` shows a top-right error toast. It is the only toast: `control-openhands browser toasts` lists just the "This conversation does not exist, or you do not have permission to access it…" text, and `control-openhands browser toasts --history` has no entry containing `this UI does not understand`. `control-openhands browser wait 'text=This conversation does not exist' --state hidden` passes within 30 s. Light theme: arrange `control-openhands browser eval "localStorage.setItem('openhands-color-theme','light-plus')"`, repeat the `goto`, and read `control-openhands browser eval "(()=>{const t=[...document.querySelectorAll('[role=status]')].find(e=>e.textContent.includes('does not exist'));return getComputedStyle(t).color+' on '+getComputedStyle(t.closest('div[style]')).backgroundColor})()"`: `rgb(97, 97, 97) on rgb(229, 229, 229)`, readable. A success toast is not: save any change on `/settings/app` and read the `Settings saved` toast the same way. Expected: dark text. Known failure (reproduced 2026-10-13, #18387): `rgb(255, 255, 255) on rgb(229, 229, 229)` (about 1.3:1), because the shared toast style hardcodes white text (`src/utils/custom-toast-handlers.tsx`); Solarized Light is the same. Restore with `browser eval "localStorage.removeItem('openhands-color-theme')"`.
+- **Error toast (`F02.error-toasts`).** Run `control-openhands browser goto /conversations/00000000-0000-0000-0000-000000000001` and `control-openhands browser wait-text 'This conversation does not exist'`. The URL becomes `/conversations`, and `control-openhands browser screenshot --feature F02.error-toasts --name missing-conversation` shows a top-right error toast. It is the only toast: `control-openhands browser toasts` lists just the "This conversation does not exist, or you do not have permission to access it…" text, and `control-openhands browser toasts --history` has no entry containing `this UI does not understand`. `control-openhands browser wait 'text=This conversation does not exist' --state hidden` passes within 30 s. Light theme: arrange `control-openhands browser eval "localStorage.setItem('openhands-color-theme','light-plus')"`, repeat the `goto`, and read `control-openhands browser eval "(()=>{const t=[...document.querySelectorAll('[role=status]')].find(e=>e.textContent.includes('does not exist'));return getComputedStyle(t).color+' on '+getComputedStyle(t.closest('div[style]')).backgroundColor})()"`: `rgb(31, 31, 31) on rgb(229, 229, 229)`: every toast uses the theme's high-emphasis ink (`--oh-contrast`) on its surface (the default theme reads `rgb(255, 255, 255) on rgb(49, 49, 49)`). A success toast reads the same: save any change on `/settings/app` (for example `control-openhands browser click 'testid=app-settings-screen >> text=Sound Notifications'`, `control-openhands browser click 'testid=submit-button'`, then toggle it back and save again) and read the `Settings saved` toast the same way: `rgb(31, 31, 31) on rgb(229, 229, 229)`. Solarized Light (`solarized-light`) gives `rgb(0, 43, 54) on rgb(230, 223, 204)` for both toasts. Restore with `browser eval "localStorage.removeItem('openhands-color-theme')"`.
- **Check for errors.** Run `control-openhands browser errors --app-only` after each group; it reports `pageErrors` `0` and `appErrors` `0`. External `registry.npmjs.org` and `cdn.simpleicons.org` failures are sandbox noise.
- **Restore.** Leave the sidebar expanded, unpinned, with the checklist shown and expanded, at `browser viewport desktop`.
@@ -106,6 +108,6 @@ Preconditions:
- **Toggle sidebar** at phone width flips the stored desktop state without any visible change. Expand the rail again after returning to desktop.
- Typing does not reset the command menu's selection. The active option keeps its index, cut down to the shorter list, from the last ArrowUp/ArrowDown or hover, and a pointer resting where the list opens hovers the option under it. A term with several matches can then run a later one on Enter: after two `ArrowDown` presses, typing `LLM profiles` (LLM profiles and Model Router) selects Model Router. Before opening the menu for an Enter check, rest the pointer with `browser hover 'testid=command-menu-trigger'`, and read `aria-activedescendant` before Enter when a term matches several items. Repro candidate, not yet filed.
- The command menu's **LLM profiles** item uses a magnifier icon (`lucide-search`, the same icon as the menu's search field).
-- The checklist card has no dismiss button. Hide it with the App settings switch or by completing every item. Clicking **Join the OpenHands Slack** counts as joining and opens an external tab you must close.
+- Dismissing the checklist with its X and the App settings switch write the same browser-local flag, so either one undoes the other. Clicking **Join the OpenHands Slack** counts as joining and opens an external tab you must close.
- Registry access depends on the sandbox. Where the browser cannot reach `registry.npmjs.org` (`ERR_CERT_AUTHORITY_INVALID`), the update check reports `Couldn't check for updates`; that is the environment, not the product. Where it can, the badge reads `Up to date` and the check says `You're running the latest version.`
- Adding a custom MCP server tests the connection first and refuses an unreachable one (`Could not reach the server...`, or `Connection failed: McpError: Connection closed` for `uvx mcp-server-time` here), so the all-complete recipe uses the local stdio fixture.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F03-home.md b/.agents/skills/verify-openhands/references/feature-map/F03-home.md
index a23cacbeed..a4a91c70df 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F03-home.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F03-home.md
@@ -10,7 +10,7 @@ warning banner replaces a silent failure when no LLM is configured. Below the
composer, home offers recommended automations, a pinned-automations grid and an
activity list of enabled automations with per-item actions.
-Source: `src/routes/home.tsx`, `src/routes/index-home.tsx`, `src/components/features/home/`, `src/components/features/plugins/plugin-picker-modal.tsx`, `src/components/features/automations/recommended-automations-launcher.tsx`, `src/hooks/use-home-pinned-automations.ts`, `src/hooks/use-home-automation-actions.ts`, `src/utils/recommended-automation-rail.ts`, `src/components/features/automations/responder-deployment-modal.tsx`.
+Source: `src/routes/home.tsx`, `src/routes/index-home.tsx`, `src/components/features/home/`, `src/components/features/plugins/plugin-picker-modal.tsx`, `src/components/features/automations/recommended-automations-launcher.tsx`, `src/hooks/use-home-pinned-automations.ts`, `src/fixtures/home-automations-demo.ts`, `src/hooks/use-home-automation-actions.ts`, `src/utils/recommended-automation-rail.ts`, `src/components/features/automations/responder-deployment-modal.tsx`.
## Sub-features
@@ -93,7 +93,7 @@ Starting work:
- **Entry points (`F03.home-entry-points`).** `control-openhands browser goto /conversations` and `control-openhands browser count 'testid=home-chat-launcher'` is `1` (same for `/`). From `/settings/secrets` (`control-openhands browser goto /settings/secrets`), `control-openhands browser press Control+k`, `control-openhands browser type 'testid=command-menu >> role=combobox' 'New chat'`, `control-openhands browser snapshot 'testid=command-menu'` (one option `New chat Start a fresh OpenHands conversation. Go` [selected]), `control-openhands browser press Enter`, `control-openhands browser url`: `/conversations`. From `/settings/secrets` (`control-openhands browser goto /settings/secrets` again), `control-openhands browser click 'testid=sidebar-conversations-link' --expect-url '/conversations(\?|$)'` and `browser count 'testid=home-chat-launcher'` is `1`.
- **`/model` on home (`F03.home-model-command`).** `control-openhands api GET /api/profiles` shows `active_profile` `deepseek-flash`. Run `control-openhands browser goto /` (Entry points ended on `/conversations`). The draft from the earlier bullets is still in the composer: clear it with `control-openhands browser fill 'testid=chat-input' ''` (typing would append). Run `control-openhands browser type 'testid=chat-input' '/model deepseek-pro'`, `control-openhands browser press Escape --selector 'testid=chat-input'` (closes the slash menu), `control-openhands browser press Enter --selector 'testid=chat-input'`. The URL stays `/`, `chat-input` is empty, the composer's model pill (`browser text 'testid=chat-input-llm-profile'`) reads `deepseek-pro`, `api GET /api/profiles` shows `active_profile` `deepseek-pro`, and `api GET '/api/conversations/search?limit=20'` lists no new conversation. Restore with the same steps and `/model deepseek-flash`.
- **Launch without a folder (`F03.launch-without-workspace`, `F03.draft-persistence`).** Run `control-openhands browser goto /conversations` (no workspace is picked yet; the pill reads Open Workspace), then fill `chat-input` with `Reply with only the word ok.` and click `testid=home-chat-launcher >> testid=submit-button` with `--expect-url '/conversations/[^/]+$'`. `control-openhands conversation wait <id> --timeout 180` reports `workspace` `<run>/private/home/workspace/project/<id without dashes>`. This is the first successful launch: `control-openhands browser click 'testid=sidebar-conversations-link' --expect-url '/conversations$'` shows an empty `chat-input` and `browser storage --session` no longer lists `oh:home-prompt-draft`.
-- **Code / Automate toggle (`F03.launcher-mode-toggle`).** Run `control-openhands browser goto /`, `control-openhands browser fill 'testid=chat-input' ''` and `control-openhands browser snapshot 'testid=home-launcher-mode-toggle'`: `radiogroup "Choose launcher mode"` with `radio "Code" [checked]` and `radio "Automate"`; `control-openhands browser eval "document.querySelector('[data-testid=chat-input]').getAttribute('data-placeholder')"` is `What do you want to build?`. Run `control-openhands browser click 'testid=home-launcher-mode-automate'`: Automate is checked, the placeholder is `What work do you want to automate?`, `control-openhands browser attr 'testid=home-composer-actions' aria-disabled` is `true`, and `control-openhands browser enabled 'testid=open-workspace-button'` and `control-openhands browser enabled 'testid=open-plugin-picker'` are `false` (`browser screenshot --feature F03.launcher-mode-toggle --name automate`). `control-openhands browser reload` brings Code back (`browser attr 'testid=home-launcher-mode-code' aria-checked` is `true`). Light theme: arrange `control-openhands browser eval "localStorage.setItem('openhands-color-theme','light-plus')"` and `browser reload`, then `control-openhands browser eval "getComputedStyle(document.querySelector('[data-testid=home-launcher-mode-code]')).color"` and `browser screenshot 'testid=home-launcher-mode-toggle' --feature F03.launcher-mode-toggle --name light-plus-code`. Expected: the checked label is dark on the light pill. Known failure (reproduced 2026-10-13, #18386): it is `rgb(255, 255, 255)` on the `rgb(229, 229, 229)` pill (about 1.3:1), so **Code** is nearly invisible (Solarized Light: white on `rgb(230, 223, 204)`, also at phone width); a hovered inactive option turns white too (`text-white`, `hover:text-white` in `home-launcher-mode-toggle.tsx`, #17199). Restore with `browser eval "localStorage.removeItem('openhands-color-theme')"` and `browser reload`.
+- **Code / Automate toggle (`F03.launcher-mode-toggle`).** Run `control-openhands browser goto /`, `control-openhands browser fill 'testid=chat-input' ''` and `control-openhands browser snapshot 'testid=home-launcher-mode-toggle'`: `radiogroup "Choose launcher mode"` with `radio "Code" [checked]` and `radio "Automate"`; `control-openhands browser eval "document.querySelector('[data-testid=chat-input]').getAttribute('data-placeholder')"` is `What do you want to build?`. Run `control-openhands browser click 'testid=home-launcher-mode-automate'`: Automate is checked, the placeholder is `What work do you want to automate?`, `control-openhands browser attr 'testid=home-composer-actions' aria-disabled` is `true`, and `control-openhands browser enabled 'testid=open-workspace-button'` and `control-openhands browser enabled 'testid=open-plugin-picker'` are `false` (`browser screenshot --feature F03.launcher-mode-toggle --name automate`). `control-openhands browser reload` brings Code back (`browser attr 'testid=home-launcher-mode-code' aria-checked` is `true`). Light theme: arrange `control-openhands browser eval "localStorage.setItem('openhands-color-theme','light-plus')"` and `browser reload`, then `control-openhands browser eval "getComputedStyle(document.querySelector('[data-testid=home-launcher-mode-code]')).color"` and `browser screenshot 'testid=home-launcher-mode-toggle' --feature F03.launcher-mode-toggle --name light-plus-code`. The checked label uses the theme's high-emphasis ink (`text-contrast`): `rgb(31, 31, 31)` on the light pill, and a hovered inactive option (`control-openhands browser hover 'testid=home-launcher-mode-automate'`) turns the same dark ink; Solarized Light reads `rgb(0, 43, 54)`, also at phone width. Restore with `browser eval "localStorage.removeItem('openhands-color-theme')"` and `browser reload`.
- **Automate launch (`F03.launch-automate-mode`).** On `/` click `testid=home-launcher-mode-automate`, fill `chat-input` with `Set up a disabled automation named QA_F03 Auto that runs every day at 09:00 UTC with the prompt: Reply with the single word pong. Only fill the automation form; do not run any other tools and do not create the automation.` and run `control-openhands browser click 'testid=home-chat-launcher >> testid=submit-button' --expect-url '/conversations/[^/]+'`. The page shows `automation-setup-panel` beside the chat (`browser testids --filter automation-setup` lists `automation-setup-form`, `automation-setup-create`, `automation-setup-save-draft`). After `control-openhands conversation wait <id> --timeout 240` is `finished`, `control-openhands browser value 'testid=automation-setup-name'` is `QA_F03 Auto` and the filled fields carry `✦ Filled by OpenHands` (`browser screenshot --feature F03.launch-automate-mode --name setup-split`), while `control-openhands api GET /api/automation/v1 --pick total` is still `0`. The sidebar row is tagged `Automation setup: Draft`.
- **Plugins (`F03.plugin-picker`).** Run `control-openhands browser goto /conversations` (a fresh page), `control-openhands browser click 'testid=open-plugin-picker'` and `control-openhands browser wait 'testid=plugin-picker-loading' --state hidden --timeout 60000`; cards `plugin-picker-card-<name>` are listed. `control-openhands browser fill 'testid=plugin-picker-search-input' weather` leaves one card (`control-openhands browser count '[data-testid^="plugin-picker-card-"]'` is `1`); `zzz-none` makes `browser text 'testid=plugin-picker'` read `No plugins available.` With `weather` again, `control-openhands browser click 'testid=plugin-picker-toggle-city-weather' --hover-first 150`: the footer reads `1 selected` and `control-openhands browser attr 'testid=plugin-picker-toggle-city-weather' aria-label` is `Remove`. `control-openhands browser click 'testid=plugin-picker-done'`, then `control-openhands browser text 'testid=plugin-picker-count'` is `1`; reopening keeps the switch checked, and `testid=plugin-picker-close` also closes it. On hover-swapping cards, use `--hover-first 150` for subsequent Add/Remove clicks too and assert both the label and `aria-checked`; the shared index gotcha explains why a plain mouse click can be swallowed (including at phone width).
- **Plugin reaches the agent (`F03.launch-with-plugin`).** With `city-weather` attached, start the no-folder launch above without its `goto` (a page load drops the attached plugin), then `control-openhands api GET '/api/conversations/<id>/events/search?limit=20'`: the system prompt contains `city-weather:now` (a conversation started without the plugin has none). Back on `/conversations` (`control-openhands browser goto /conversations`), `control-openhands browser count 'testid=plugin-picker-count'` is `0`.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F04-conversation-list.md b/.agents/skills/verify-openhands/references/feature-map/F04-conversation-list.md
index f1eb702289..a27d790ade 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F04-conversation-list.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F04-conversation-list.md
@@ -107,7 +107,7 @@ Preconditions:
- **Remove tags (`F04.edit-tags-remove`).** Reopen Edit tags on that row and run `control-openhands browser click 'testid=remove-tag-qaregion'` (likewise `remove-tag-qastage`, `remove-tag-qaowner`); `control-openhands browser count 'testid=edit-tags-rows >> [data-testid^=edit-tag-row-]'` is `2`. Save with `testid=edit-conversation-tags-modal >> testid=confirm-button` (`Tags updated`); after `browser reload` the `+3` chip is gone and `control-openhands api GET '/api/conversations/search?limit=1' --pick items.0.tags` is `clientsource`, `qabare`, `qaenv` only.
- **Tag filter (`F04.tag-filter`).** After Edit tags (above) gave one row `qaenv=one`, run `control-openhands browser click 'testid=conversation-layouts-toggle'`, `control-openhands browser click 'testid=tag-filters-section'`, `control-openhands browser click 'testid=tag-facet-row-qaenv=one'` (`aria-checked` `true`; the menu stays open), `control-openhands browser press Escape`, then `control-openhands browser count 'testid=conversation-panel-list-scroll >> testid=conversation-card'`: only the tagged row plus pinned rows. The count is the same after `browser reload`.
- **Filter chips (`F04.filter-chips`).** With the facet active, `control-openhands browser text 'testid=conversation-active-tag-filters'` is `qaenv=one` and `Clear all`. `control-openhands browser click 'testid=active-tag-filter-qaenv=one'` removes it (`browser count 'testid=conversation-active-tag-filters'` is `0`). Select `tag-facet-row-qaenv=one` and `tag-facet-row-qabare=` again, then `control-openhands browser click 'testid=clear-tag-filters'`: the strip disappears and the full list returns.
-- **Automation setup tag (`F04.automation-setup-tag`).** Arrange one setup conversation through the setup page (F22): `control-openhands browser goto /automations`, `control-openhands browser click 'testid=automations-add-automation'`, `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`, then `control-openhands conversation start --stay --prompt 'Reply with only the word ok. Do not run any tools.' --wait --timeout 180` (one tiny model call; note `<setup-id>`). `control-openhands api GET /api/conversations/<setup-id> --pick tags` has `automationsetup: draft`. In the sidebar, `control-openhands browser text 'testid=conversation-card >> has-text=Reply >> testid=conversation-card-tag-chip'` is `Automation setup: Draft` (the first chip; the internal draft-id keys stay hidden). Run `control-openhands browser click 'testid=conversation-layouts-toggle'`, `control-openhands browser click 'testid=tag-filters-section'`: the facet `testid=tag-facet-row-automationsetup=draft` reads `Automation setup: Draft`; click it, `control-openhands browser press Escape`, and `control-openhands browser text 'testid=conversation-active-tag-filters'` is `Automation setup: Draft` and `Clear all`, with only setup conversations (and pinned rows) left; `control-openhands browser click 'testid=active-tag-filter-automationsetup=draft'` removes it. After `control-openhands browser mouse-click 900 900`, `control-openhands browser tooltip 'testid=conversation-card >> has-text=Reply'` lists `Automation setup` and the raw value `draft` (`browser screenshot --feature F04.automation-setup-tag --name facet`). End with `control-openhands browser goto /` and clear the Home draft the setup send left behind (`control-openhands browser fill 'testid=chat-input' ''`, see F05 Home draft): opening another conversation from the sidebar while the setup conversation is open re-tags it (see Gotchas).
+- **Automation setup tag (`F04.automation-setup-tag`).** Arrange one setup conversation through the setup page (F22): `control-openhands browser goto /automations`, `control-openhands browser click 'testid=automations-add-automation'`, `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`, then `control-openhands conversation start --stay --prompt 'Reply with only the word ok. Do not run any tools.' --wait --timeout 180` (one tiny model call; note `<setup-id>`). `control-openhands api GET /api/conversations/<setup-id> --pick tags` has `automationsetup: draft`. In the sidebar, `control-openhands browser text 'testid=conversation-card >> has-text=Reply >> testid=conversation-card-tag-chip'` is `Automation setup: Draft` (the first chip; the internal draft-id keys stay hidden). Run `control-openhands browser click 'testid=conversation-layouts-toggle'`, `control-openhands browser click 'testid=tag-filters-section'`: the facet `testid=tag-facet-row-automationsetup=draft` reads `Automation setup: Draft`; click it, `control-openhands browser press Escape`, and `control-openhands browser text 'testid=conversation-active-tag-filters'` is `Automation setup: Draft` and `Clear all`, with only setup conversations (and pinned rows) left; `control-openhands browser click 'testid=active-tag-filter-automationsetup=draft'` removes it. After `control-openhands browser mouse-click 900 900`, `control-openhands browser tooltip 'testid=conversation-card >> has-text=Reply'` lists `Automation setup` and the raw value `draft` (`browser screenshot --feature F04.automation-setup-tag --name facet`). End with `control-openhands browser goto /`; the setup send leaves no Home draft, and opening another conversation from the sidebar leaves that conversation's tags alone (see Gotchas).
- **Download (`F04.download`).** Open ⋮ and run `control-openhands browser click 'testid=context-menu >> testid=download-trajectory-button'`, then `control-openhands browser downloads`. A file `…-conversation_<id>.zip` lands in `<run>/private/downloads`; `control-openhands browser downloads --last 1 --inspect` lists `<id-hex>/base_state.json` and `meta.json` for any row, plus event folders such as `bash_events/` for a conversation that ran (the `pong` row; it is the oldest, so Load more first).
- **Download failure (`F04.download-error`).** With ⋮ already open on a row, run `control-openhands browser errors --clear` and `control-openhands service stop agent-server`, click `testid=context-menu >> testid=download-trajectory-button`, then `control-openhands browser toasts --history` and `control-openhands browser count 'testid=context-menu'`. There is exactly one toast, `Couldn't download the conversation data: An error occurred`, and the menu is closed (`0`). The download request returned a plain-text ingress 502; `getApiErrorMessage` intentionally omits raw transport text and falls back to this generic message when the response body has no usable error field. `control-openhands browser errors --app-only` has `pageErrors` `0`; its 502 request and console error are the outage itself. Run `control-openhands restart` and `control-openhands doctor` afterwards.
- **Archive (`F04.archive`).** Open a conversation from its row, open ⋮ and run `control-openhands browser click 'testid=context-menu >> testid=archive-button'`; `control-openhands browser snapshot 'role=dialog'` reads `Archive conversation` and `Archive "<title>"? The conversation stays available with its full history and can be restored from Show archived.` `control-openhands browser click 'role=dialog >> testid=cancel-button'` keeps the row. Repeat and `control-openhands browser click 'role=dialog >> testid=confirm-button' --expect-url '/conversations(\?|$)'`; the row count is `0`, also after `browser reload`. Archiving a pinned row removes the Pinned section.
@@ -138,5 +138,5 @@ Preconditions:
- `service stop agent-server` needs `control-openhands restart` and a `browser reload` before anything else works.
- The list refreshes on an interval: a row keeps `conversation-status-working` and `Conversation <id5>` for up to about 40 s after the conversation finished. Use `browser wait ... --timeout 60000`, not an immediate `count`.
- Model-written titles vary from run to run; match finished conversations on a word from the prompt (`has-text=pong`), never on the whole title.
-- Expected: opening a conversation from the sidebar never changes its tags. Known failure (reproduced 2026-10-13, #18378): a client-side switch from an automation setup conversation (one tagged `automationsetup: draft`) to any other conversation re-tags that conversation `automationsetup: draft` on the server, so it gains the `Automation setup: Draft` chip and opens as a setup page (F07 `F07.open-by-url`). Remove the tag with Edit tags (`remove-tag-automationsetup`) to restore it. Use `browser goto` between setup and other conversations.
+- Opening a conversation from the sidebar never changes its tags, also right after an automation setup conversation (one tagged `automationsetup: draft`): the other conversation keeps its tags and opens as a normal chat (recipe in F07 `F07.open-by-url`). A conversation that did get the tag loses it with Edit tags (`remove-tag-automationsetup`).
- Known issue #17919: Delete all deletes only the loaded pages, and its dialog states that smaller count (`F04.delete-all`).
diff --git a/.agents/skills/verify-openhands/references/feature-map/F05-composer.md b/.agents/skills/verify-openhands/references/feature-map/F05-composer.md
index 9619330946..fdbd026f1e 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F05-composer.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F05-composer.md
@@ -59,12 +59,32 @@ Preconditions:
- **Type and send (`F05.send-message`).** On `/conversations/<id>` run `control-openhands browser attr 'testid=submit-button' aria-label` (`Send message`), `control-openhands browser enabled 'testid=submit-button'` (`false`), `control-openhands browser type 'testid=chat-input' 'Reply with only:'`, `control-openhands browser enabled 'testid=submit-button'` (`true`), `control-openhands browser press Shift+Enter`, `control-openhands browser type 'testid=chat-input' 'ok2'` and `control-openhands browser eval "document.querySelector('[data-testid=chat-input]').innerText"`: `Reply with only:\nok2`. Run `control-openhands browser press Enter --selector 'testid=chat-input'`; the field is empty and `control-openhands browser count 'testid=stop-button'` is `1` while the agent runs. Then `control-openhands conversation wait <id> --fresh --timeout 180` (a fast model can finish before a separate `--until running` wait starts, which then times out on `finished`), `control-openhands browser reload` and `control-openhands browser count 'testid=user-message >> has-text=ok2'`: `1`; `control-openhands browser text 'testid=agent-message >> nth=-1'` is `ok2`.
- **Draft survives navigation (`F05.draft-persistence`).** Run `control-openhands browser type 'testid=chat-input' 'QA draft text'` and wait out the 500 ms save: `control-openhands browser eval "new Promise(r=>setTimeout(()=>r(JSON.parse(localStorage.getItem('conversation-state-<id>')).draftMessage),1500))"` returns `QA draft text`. Then `control-openhands browser click 'testid=backend-selector-settings-link' --expect-url settings`, `control-openhands browser back` and the `innerText` eval above: `QA draft text`. `control-openhands browser reload` restores it again. A draft typed just before leaving is flushed, not lost (#17926, fixed by #17968): `control-openhands browser type 'testid=chat-input' ' plus fast'` and at once `control-openhands browser click 'testid=backend-selector-settings-link' --expect-url settings`, `control-openhands browser back`: the eval reads `QA draft text plus fast`. The same holds for an immediate sidebar switch to another conversation and back (that other conversation's composer stays empty). Clear with `control-openhands browser fill 'testid=chat-input' ''` and wait a second before navigating.
-- **Home draft (`F05.draft-persistence`).** Run `control-openhands browser goto /`, `control-openhands browser type 'testid=chat-input' 'QA home draft'`, `control-openhands browser goto /settings/secrets`, `control-openhands browser goto /`, then the `innerText` eval: `QA home draft`; `control-openhands browser storage --session` lists `oh:home-prompt-draft`. Clear with `control-openhands browser fill 'testid=chat-input' ''`. Expected: a prompt sent from the automation setup page's docked composer (`/automations/setup`, F22) does not stay behind as the Home draft. Known failure (reproduced 2026-10-13, #18379): after `browser goto /automations`, `browser click 'testid=automations-add-automation'`, `browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`, `browser type 'testid=chat-input' 'Reply with only the word ok2. Do not run any tools.'` and `browser click 'testid=submit-button' --expect-url '/conversations/'`, `browser storage --session --values` still holds that text under `oh:home-prompt-draft`, and New Chat shows it in the Home composer (a Home send clears it). Clear it before the next bullet.
+- **Home draft (`F05.draft-persistence`).** Run `control-openhands browser goto /`, `control-openhands browser type 'testid=chat-input' 'QA home draft'`, `control-openhands browser goto /settings/secrets`, `control-openhands browser goto /`, then the `innerText` eval: `QA home draft`; `control-openhands browser storage --session` lists `oh:home-prompt-draft`. The automation setup page's docked composer (`/automations/setup`, F22) keeps no draft and never touches the Home one. With `QA home draft` still in the Home composer, run `control-openhands browser goto /automations`, `control-openhands browser click 'testid=automations-add-automation'`, `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'` and `control-openhands browser wait 'testid=chat-input'`: the `innerText` eval is empty. `control-openhands browser type 'testid=chat-input' 'Reply with only the word ok2. Do not run any tools.'`, then `control-openhands browser eval "new Promise(r=>setTimeout(()=>r(sessionStorage.getItem('oh:home-prompt-draft')),1500))"` still returns `QA home draft`. `control-openhands browser click 'testid=submit-button' --expect-url '/conversations/'` and `control-openhands browser click 'testid=sidebar-conversations-link' --expect-url '/conversations(\?|$)'`: the Home composer shows `QA home draft`. Clear it (`fill 'testid=chat-input' ''`, wait a second) and repeat the setup send: New Chat is empty and `browser storage --session --values` has no `oh:home-prompt-draft`. Unsent text in the setup composer is lost when the user leaves that page (by design).
- **Auto-resize (`F05.composer-resize`).** On `/conversations/<id>` (`control-openhands browser goto /conversations/<id>`; Home draft left `/`) type `line`, then eight times `control-openhands browser press Shift+Enter` and `control-openhands browser type 'testid=chat-input' 'lineN'` (`line2`…`line9`): `control-openhands browser bbox 'testid=chat-input'` grows by 20 px per line, from `height` `20` to `180`; after `control-openhands browser fill 'testid=chat-input' ''` it is `20`. `control-openhands browser count '#resize-grip'` is `1`. `control-openhands browser drag '#resize-grip' --by 0,-200` makes the empty field `220` high and `control-openhands browser drag '#resize-grip' --by 0,150` brings it to `70`; after `control-openhands browser reload` it is `20` again (the dragged height is not kept).
- **Tools menu and macros (`F05.plus-menu`).** On `/conversations/<id>` run `control-openhands browser click 'testid=chat-plus-button'` and `control-openhands browser testids 'testid=tools-context-menu'`: `macros-button`, `show-skills-button`, `show-hooks-button`, `show-agent-tools-button`, `add-files-and-images-button` (no `git-tools-button` outside a repository). `control-openhands browser hover 'testid=macros-button'` shows `macros-submenu` with `increase-test-coverage-button`, `fix-readme-button`, `auto-merge-prs-button`, `clean-dependencies-button`. `control-openhands browser click 'testid=fix-readme-button'` closes the menu and fills the composer with `Please look at the README and make the following improvements…`; the `testid=user-message` count is unchanged. Clear with `control-openhands browser fill 'testid=chat-input' ''`. Open the menu again and `control-openhands browser press Escape`: `browser count 'testid=tools-context-menu'` is `0` and `control-openhands browser eval "document.activeElement.dataset.testid"` is `chat-plus-button`.
- **Show dialogs (`F05.plus-menu`).** For each of `show-skills-button`, `show-hooks-button`, `show-agent-tools-button`: `control-openhands browser click 'testid=chat-plus-button'`, `control-openhands browser click 'testid=show-hooks-button'` (etc.), `control-openhands browser text 'role=dialog'` (starts `Available Skills`, `Available Hooks`, `Agent Tools & Metadata`), then `control-openhands browser press Escape` and `control-openhands browser count 'role=dialog'`: `0`.
- **Agent profile before start (`F05.agent-profile-switch`).** Run `control-openhands browser goto /`, `control-openhands browser click 'testid=chat-plus-button'`, `control-openhands browser hover 'testid=switch-agent-profile-button'` and `control-openhands browser text 'testid=agent-profile-submenu'`: `AVAILABLE PROFILES`, `default` (checked; `chat-input-agent-profile-option-default`), `Manage agent profiles`. `control-openhands browser click 'testid=agent-profile-submenu >> text=Manage agent profiles' --expect-url 'settings/agents'` lands on `/settings/agents`. In a started conversation (`control-openhands browser goto /conversations/<id>`) `control-openhands browser click 'testid=chat-plus-button'` then `control-openhands browser count 'testid=switch-agent-profile-button'` is `0`.
-- **Attach a file and an image (`F05.attach-files`).** On `/conversations/<id>` run `control-openhands browser upload 'testid=upload-image-input' "$OH_VERIFY_RUN/evidence/_fixtures/qa-image.png" "$OH_VERIFY_RUN/evidence/_fixtures/qa-note.txt"` and `control-openhands browser screenshot 'testid=interactive-chat-box' --feature F05.attach-files --name attached`: a `qa-note.txt TXT` chip and a thumbnail. `control-openhands browser count 'testid=interactive-chat-box >> alt=qa-image.png'` is `1`. `control-openhands browser click 'role=button[name="Upload as file"]'` turns the image toggle into `Do not upload as file` (`control-openhands browser attr 'role=button[name="Do not upload as file"]' aria-pressed` is `true`); click it again to restore. The remove buttons are named: `control-openhands browser snapshot 'testid=interactive-chat-box'` lists `button "Remove qa-note.txt"` and `button "Remove image"`. Remove the image with `control-openhands browser click 'testid=interactive-chat-box >> role=button[name="Remove image"]'`; the `alt=qa-image.png` count is `0`. Then `control-openhands browser type 'testid=chat-input' 'Reply with only the content of the attached file.'`, `control-openhands browser press Enter --selector 'testid=chat-input'`, wait as in Type and send, and `control-openhands browser text 'testid=user-message >> nth=-1'`: it ends `NEW FILES ADDED: qa-note.txt`. `control-openhands conversation events <id> --kinds MessageEvent --last 3` shows the reply `QA attachment 7731`, and `qa-note.txt` sits in the conversation `workspace` printed by `control-openhands conversation status <id>`.
+- **Attach a file and an image (`F05.attach-files`).** On `/conversations/<id>` run `control-openhands browser upload 'testid=upload-image-input' "$OH_VERIFY_RUN/evidence/_fixtures/qa-image.png" "$OH_VERIFY_RUN/evidence/_fixtures/qa-note.txt"` and `control-openhands browser screenshot 'testid=interactive-chat-box' --feature F05.attach-files --name attached`: a `qa-note.txt TXT` chip and a thumbnail. `control-openhands browser count 'testid=interactive-chat-box >> alt=qa-image.png'` is `1`. `control-openhands browser click 'role=button[name="Upload as file"]'` turns the image toggle into `Do not upload as file` (`control-openhands browser attr 'role=button[name="Do not upload as file"]' aria-pressed` is `true`); click it again to restore. The remove buttons are named: `control-openhands browser snapshot 'testid=interactive-chat-box'` lists `button "Remove qa-note.txt"` and `button "Remove image"`. Remove the image with `control-openhands browser click 'testid=interactive-chat-box >> role=button[name="Remove image"]'`; the `alt=qa-image.png` count is `0`. Then `control-openhands browser type 'testid=chat-input' 'Reply with only the content of the attached file.'`, `control-openhands browser press Enter --selector 'testid=chat-input'`, wait as in Type and send, and `control-openhands browser text 'testid=user-message >> nth=-1'`: it ends `NEW FILES ADDED: qa-note.txt`. `control-openhands conversation events <id> --kinds MessageEvent --last 3` shows the reply `QA attachment 7731`, and `qa-note.txt` sits in the conversation `workspace` printed by `control-openhands conversation status <id>`. Attachments added on Home before the conversation exists upload after it is created: with no workspace or a Local Repo workspace they land in it (`browser network --filter upload` shows `200`). Expected: the same in New Worktree mode. Known failure (reproduced 2026-10-14, #18421): run `control-openhands fixture git-repo --name qa-repo`, `control-openhands browser goto /`, `control-openhands workspace open qa-repo`, `control-openhands browser click 'testid=workspace-mode-selector'`, `control-openhands browser click 'testid=workspace-mode-selector-option-new_worktree'`, `control-openhands browser click 'testid=chat-plus-button'`, `control-openhands browser upload-via 'testid=add-files-and-images-button' "$OH_VERIFY_RUN/evidence/_fixtures/qa-note.txt"`, type `Reply with only the word received. Do not run any tools.` and `control-openhands browser click 'testid=home-chat-launcher >> testid=submit-button' --expect-url '/conversations/[^/]+$'`: `control-openhands browser toasts --history` shows `HTTP request failed (422 Unprocessable Entity): {"detail":"path must be inside the conversation workspace"}` and `browser network --filter upload` a `422` for `?path=<run>/workspace/qa-repo/qa-note.txt`, while `conversation status <id>` reports the worktree `/tmp/conversation-worktrees/<id>/qa-repo`; the message is sent without the file. The same file attached inside that conversation afterwards uploads to the worktree (`200`). Restore with `testid=workspace-mode-selector-option-local_repo`.
+ ```sh known-bug F05.attach-files home-new-worktree
+ control-openhands browser viewport desktop
+ control-openhands fixture git-repo --name qa-repo
+ control-openhands fixture file --name qa-note.txt --content "QA attachment 7731"
+ control-openhands browser goto /
+ control-openhands browser fill 'testid=chat-input' ''
+ control-openhands workspace open qa-repo
+ control-openhands browser click 'testid=workspace-mode-selector'
+ control-openhands browser click 'testid=workspace-mode-selector-option-new_worktree'
+ control-openhands assert --field text --matches 'New Worktree' -- browser text 'testid=home-git-control-bar-preview'
+ control-openhands browser click 'testid=chat-plus-button'
+ control-openhands browser upload-via 'testid=add-files-and-images-button' "$OH_VERIFY_RUN/evidence/_fixtures/qa-note.txt"
+ control-openhands browser type 'testid=chat-input' 'Reply with only the word received. Do not run any tools.'
+ control-openhands browser network --clear
+ control-openhands browser click 'testid=home-chat-launcher >> testid=submit-button' --expect-url '/conversations/[^/]+$'
+ control-openhands browser wait 'testid=user-message' --timeout 30000
+ control-openhands browser eval "new Promise(r=>setTimeout(r,5000))"
+ control-openhands browser screenshot --feature F05.attach-files --name home-new-worktree
+ control-openhands assert --field recent.0.status --equals 200 --bug-shows 422 -- browser network --filter file/upload # bug
+ ```
- **Size limit (`F05.attach-size-limit`).** Arrange `head -c 4194304 /dev/zero > "$OH_VERIFY_RUN/evidence/_fixtures/qa-big.bin"` (and `qa-two-a.bin`, `qa-two-b.bin` with `head -c 2097152`). Run `control-openhands browser upload 'testid=upload-image-input' "$OH_VERIFY_RUN/evidence/_fixtures/qa-big.bin"` and `control-openhands browser toasts`: `Error: Files exceeding 3MB are not allowed: qa-big.bin`. Uploading `qa-two-a.bin` and `qa-two-b.bin` together toasts `Error: Total file size would be 4.0MB, exceeding the 3MB limit. Please select fewer or smaller files.` `control-openhands browser count 'testid=interactive-chat-box >> text=qa-big.bin'` is `0`.
- **Image alone (`F05.image-only-send`).** From the home composer: `control-openhands browser goto /`, `control-openhands browser enabled 'testid=submit-button'` (`false`, empty field), `control-openhands browser click 'testid=chat-plus-button'`, `control-openhands browser upload-via 'testid=add-files-and-images-button' "$OH_VERIFY_RUN/evidence/_fixtures/qa-image.png"`, then `control-openhands browser count 'testid=home-chat-launcher >> alt=qa-image.png'` (`1`) and `control-openhands browser enabled 'testid=submit-button'`: `true` while `control-openhands browser eval "document.querySelector('[data-testid=chat-input]').innerText"` is still empty (`control-openhands browser screenshot 'testid=home-chat-launcher' --feature F05.image-only-send --name image-alone-send-enabled`). Run `control-openhands browser click 'testid=submit-button' --expect-url '/conversations/'` and read `<image-id>` from `control-openhands browser url`. After `control-openhands browser wait 'testid=user-message'` and `control-openhands browser wait 'testid=chat-message-sending' --state detached --timeout 5000` (the pending bubble carries `testid=user-message` too, so wait for its `Sending...` to go, not only for a bubble; the wait returns `"state": "detached"` at once when the indicator has already gone, the usual case on a local stack), `control-openhands browser count 'testid=user-message >> testid=image-preview'` is `1`, and `control-openhands browser count 'testid=chat-message-sending'` and `control-openhands browser count 'testid=chat-message-retry'` are `0`. Server side, `control-openhands conversation events <image-id> --kinds MessageEvent --grep 'data:image/png;base64'` matches the one user row (`count` `1`; its `text` is a blank `" "`, the row carries `images: 1`, and the excerpt shows the `"type":"image"` block with `image_urls`), and `control-openhands conversation events <image-id> --kinds MessageEvent` (default `--last`) lists the same row with `images: 1` and no excerpt. `--kinds` filters before `--last` counts, so `--kinds MessageEvent --last 1` is that same row. To check that the model is handed the picture, send the image from Home together with `Which two colours are the bars in this image? Answer with the colour names only. Do not run any tools.` and read `control-openhands conversation events <id> --kinds MessageEvent`. Expect `orange` and `blue` in the agent row and `--kinds ActionEvent` `count` `0` (`fixture image` draws those bars). Known failure (reproduced 2026-10-08, agent-server 1.53.0, the pinned version): on a `deepseek/*` profile the user row still carries `images: 1`, but the agent answers that no image was attached. The model gets text only. The cause is in the SDK, not Canvas: 1.53.0 string-serializes every model id containing `deepseek`, which drops image parts. OpenHands/software-agent-sdk#5460 and #5467 fixed it on 2026-10-08, after the 1.53.0 release. On a stack launched with `--sdk-path <software-agent-sdk main>` the same send answers `Blue and orange.` with no tool call. An image-only send there reaches the model too: its first reasoning describes four blue and orange stripes. On an affected stack, the image-only send leaves the model with an empty task, and deepseek-flash then explores the disk (see Gotchas). Return with `control-openhands browser goto /conversations/<id>`.
- **Paste text (`F05.paste`).** Run `control-openhands browser type 'testid=chat-input' 'QA paste text'`, `control-openhands browser press Control+a`, `control-openhands browser press Control+c`, `control-openhands browser press Delete`, `control-openhands browser press Control+v`, then `control-openhands browser eval "document.querySelector('[data-testid=chat-input]').innerHTML"`: `QA paste text`, no markup. On a cleared field `control-openhands browser paste 'testid=chat-input' --file "$OH_VERIFY_RUN/evidence/_fixtures/qa-image.png"` attaches the image (`control-openhands browser count 'testid=interactive-chat-box >> alt=qa-image.png'` is `1`, with an `Upload as file` toggle) and `control-openhands browser paste 'testid=chat-input' --text 'QA pasted plain'` inserts plain text (`innerHTML` is `QA pasted plain`). `control-openhands browser drop-files 'testid=chat-input' "$OH_VERIFY_RUN/evidence/_fixtures/qa-note.txt" --stage over` makes `control-openhands browser text 'testid=interactive-chat-box'` start `Drop your files here`; the same command without `--stage` drops it: a `qa-note.txt` `TXT` chip appears. Remove the chip with `control-openhands browser click 'testid=interactive-chat-box >> role=button[name="Remove qa-note.txt"]'` and the image as in Attach; clear the field.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F06-agent-activity.md b/.agents/skills/verify-openhands/references/feature-map/F06-agent-activity.md
index d382252ae4..3061edb486 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F06-agent-activity.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F06-agent-activity.md
@@ -39,7 +39,7 @@ Source: `src/components/features/chat/chat-interface.tsx`, `src/components/conve
- `F06.reconnect`: when the Agent Server goes away the composer shows Disconnected; after it returns the page reconnects without a reload.
- `F06.error-banner`: losing the server shows "Unable to connect to server" with Retry, Copy and Close; Retry reconnects once the server is back.
- `F06.error-events`: a conversation error (ConversationErrorEvent) surfaces as a warning banner and the composer status Error.
-- `F06.confirmation-mode`: with confirmation mode on, a lock chip sits above the composer and each action waits for "Do you want to continue with this action?" with Cancel and Continue (hints `⇧⌘⌫` / `⌘↩` on macOS, `Ctrl+⇧+⌫` / `Ctrl+↩` elsewhere); a cancelled action is marked Rejected and counts as resolved in its group.
+- `F06.confirmation-mode`: with confirmation mode on, a lock chip sits above the composer and each action waits for "Do you want to continue with this action?" with Cancel and Continue (hints `⇧⌘⌫` / `⌘↩` on macOS, `Ctrl+⇧+⌫` / `Ctrl+↩` elsewhere, hidden below 640 px); a cancelled action is marked Rejected and counts as resolved in its group.
- `F06.confirmation-shortcuts`: Cmd+Enter or Ctrl+Enter continues and Shift+Cmd+Backspace or Shift+Ctrl+Backspace cancels a pending action.
- `F06.branch-from-here`: hovering a message shows Branch from here; on an agent message it forks inclusively; on a user message it forks without that message and pre-fills its text in the new composer, ready to send. The branch is titled "<title> (branch)" and keeps working.
- `F06.image-attachments`: images sent with a message show as thumbnails in the bubble; expand opens a lightbox that closes with X or Escape.
@@ -77,7 +77,7 @@ Preconditions:
- **Markdown file preview (`F06.markdown-file-preview`).** Run `control-openhands conversation start --prompt "Use the file_editor tool to create the file qa_notes.md containing a heading '# QA Notes' and one bullet '- first'. Do nothing else and reply DONE." --wait --timeout 150`. Without any click, `control-openhands browser text 'testid=markdown-file-preview'` reads `QA Notes first qa_notes.md View` (rendered heading and bullet, then the file name) and `count 'testid=event-group'` is `0`. `control-openhands browser count 'testid=file-quick-row-item-qa_notes.md'` is `0`; `control-openhands browser click 'testid=markdown-file-preview-view'` makes it `1` and `control-openhands browser text 'testid=file-content-viewer-markdown'` reads `QA Notes first`. Then `control-openhands browser goto /conversations/<stream id>` to return to the stream conversation for the next bullets.
- **Timestamps (`F06.timestamps`).** Run `control-openhands browser tooltip 'testid=user-message'` and `control-openhands browser tooltip 'testid=event-group-toggle'`. Both return the local date and time, e.g. `Oct 4, 2026, 11:08 PM`.
- **Copy a message and a code block (`F06.message-copy`, `F06.code-block-copy`).** On the same page run `control-openhands browser clipboard --write qa-empty`, `control-openhands browser hover 'testid=agent-message >> nth=0'` and `control-openhands browser click 'testid=agent-message >> nth=0 >> testid=copy-to-clipboard >> nth=0'` (the bubble's button; the code block has a second, hidden `copy-to-clipboard`). `control-openhands browser attr 'testid=agent-message >> nth=0 >> testid=copy-to-clipboard >> nth=0' aria-label` is `Copied to clipboard`, and `Copy to clipboard` again about 2 s later; `control-openhands browser clipboard` starts with `| Step | Result |` and ends with the fenced `print(1)` block. Then `control-openhands browser hover 'testid=agent-message >> pre'`, `control-openhands browser click 'testid=agent-message >> pre >> testid=copy-to-clipboard'` and `control-openhands browser clipboard` is exactly `print(1)`.
-- **Long table cell (`F06.markdown-rendering`).** On the stream conversation write a prompt file whose first line is `Reply with only the word OK. Do not run any tools.`, then a blank line and the table `| Name | Description | Status |`, `|---|---|---|`, `| Alpha | This is a really long sentence in a single cell that goes on and on to explain a very involved aspect of the system, which nobody could read on one line without scrolling forever to the right. | ok |`, `| Beta | brief | fail |`. Run `control-openhands browser fill 'testid=chat-input' x --value-file <file>`, `control-openhands browser click 'testid=submit-button'` and `control-openhands conversation wait <id> --fresh --timeout 120`; the user bubble renders the table. Expected: the long cell wraps at a bounded width and the `Status` column stays visible. Known failure (reproduced 2026-10-13, #18356): every cell is `white-space: nowrap`, so `control-openhands browser bbox 'testid=user-message >> nth=-1 >> testid=markdown-table-scroll'` has `scrollWidth` `1394` against `clientWidth` `704` at desktop with the drawer open and `Status` is scrolled off (`browser screenshot 'testid=chat-interface' --feature F06.markdown-rendering --name long-cell-table`); an agent reply with such a table behaves the same.
+- **Long table cell (`F06.markdown-rendering`).** On the stream conversation write a prompt file whose first line is `Reply with only the word OK. Do not run any tools.`, then a blank line and the table `| Name | Description | Status |`, `|---|---|---|`, `| Alpha | This is a really long sentence in a single cell that goes on and on to explain a very involved aspect of the system, which nobody could read on one line without scrolling forever to the right. | ok |`, `| Beta | brief | fail |`. Run `control-openhands browser fill 'testid=chat-input' x --value-file <file>`, `control-openhands browser click 'testid=submit-button'` and `control-openhands conversation wait <id> --fresh --timeout 120`; the user bubble renders the table. Cells wrap once they reach 28rem (`max-w-md`): `control-openhands browser eval "getComputedStyle(document.querySelector('[data-testid=markdown-table-scroll] td')).whiteSpace + ' ' + getComputedStyle(document.querySelector('[data-testid=markdown-table-scroll] td')).maxWidth"` is `normal 448px`. At desktop with the right panel closed, `control-openhands browser bbox 'testid=user-message >> nth=-1 >> testid=markdown-table-scroll'` has `scrollWidth` `582` equal to `clientWidth` and `Status` is visible (`browser screenshot 'testid=chat-interface' --feature F06.markdown-rendering --name long-cell-table`). A table still wider than the conversation keeps the horizontal scroll with its edge fade: with the panel open (`control-openhands browser click 'testid=right-panel-toggle'`) `clientWidth` is `474`, and at phone width `326`, so `Status` is scrolled off there while the long cell stays wrapped. An agent reply with such a table behaves the same.
- **Non-zero exit badge (`F06.tool-visualizers`).** In an open conversation run `control-openhands browser fill 'testid=chat-input' "Run exactly this terminal command, verbatim, with nothing appended: ls /qa_missing_dir2 . Then reply OK."`, `control-openhands browser click 'testid=submit-button'` and `control-openhands conversation wait <id> --fresh --timeout 150` (without `--fresh` it returns the previous run's `finished` at once). If the reply's actions render as a single row, run `control-openhands browser click 'testid=chat-scroll-container >> role=button[name="Expand"][exact] >> nth=-1'`; if they render as a group (`browser count 'testid=event-group'` grew, e.g. `2 actions completed` after a terminal reset and rerun), run `control-openhands browser click 'testid=event-group-toggle >> nth=-1'` and `control-openhands browser click 'testid=event-group-content >> nth=-1 >> role=button[name="Expand"][exact] >> nth=-1'`. Then `control-openhands browser count 'testid=chat-scroll-container >> text="exit 2"'` is `1`; the badge sits above the `No such file or directory` output (`browser screenshot 'testid=event-group >> nth=-1' --feature F06.tool-visualizers --name exit-code`). Repeating the check in the same conversation needs a new directory name and a count that grows, because an earlier expanded badge stays on the page.
- **Task list (`F06.task-list`, `F06.status-indicator`).** From `/` run `control-openhands browser goto /`, `control-openhands browser fill 'testid=chat-input' "Use your task_tracker tool to plan exactly two tasks titled 'qa step one' and 'qa step two'; mark 'qa step one' done and 'qa step two' todo. Do nothing else and reply DONE."` and `control-openhands browser click 'testid=submit-button' --expect-url '/conversations/' --observe '[data-testid=chat-status-indicator]' --observe-ms 8000`. `observed` contains `Starting` for a moment (the pill). Then `control-openhands conversation wait <id> --timeout 150` and `control-openhands browser text 'testid=chat-scroll-container'`: it contains `Tasks`, `qa step one`, `qa step two`; the screenshot shows a check icon and muted text for the done task, an empty circle for the todo one.
- **Empty state (`F06.empty-state-suggestions`).** Run `control-openhands browser click 'testid=conversation-panel-new-thread-picker'`, `control-openhands browser click 'testid=launch-no-workspace' --expect-url '/conversations/'`, then `control-openhands browser text 'testid=chat-suggestions'`: `Let's start building!` plus `Increase test coverage`, `Auto-merge PRs`, `Fix README`, `Clean dependencies`. Run `control-openhands browser click 'testid=chat-suggestions >> text=Fix README'` and `control-openhands browser text 'testid=chat-input'`: the composer holds the README-improvement prompt.
@@ -90,7 +90,7 @@ Preconditions:
- **Reload mid-run (`F06.reload-mid-run`).** Run `control-openhands conversation start --prompt "Run exactly one terminal command: sleep 15 && echo qa-reloaded. Then reply with only its output."`, wait for `testid=live-activity-chip`, then `control-openhands browser reload`. The composer still reads `Running` and the chip is back. After `control-openhands conversation wait <id> --timeout 120`: `testid=agent-message` is `qa-reloaded`, `count 'testid=user-message'` is `1`, `control-openhands browser eval "document.querySelectorAll('[data-testid=chat-scroll-container] [data-testid=generic-event-message-title]').length"` is `1`.
- **Older history (`F06.load-older-history`).** Run `control-openhands conversation start --prompt "Run these 16 terminal commands strictly one per tool call, sequentially (never combine them): echo qa1, echo qa2, echo qa3, echo qa4, echo qa5, echo qa6, echo qa7, echo qa8, echo qa9, echo qa10, echo qa11, echo qa12, echo qa13, echo qa14, echo qa15, echo qa16. Then reply DONE." --wait --timeout 400` (note `<history-id>`); `control-openhands conversation events <history-id> --last 500` reports a count above 50. Run `control-openhands browser network --clear`, `control-openhands browser reload`, then `control-openhands browser network --last 300`: two `GET /api/conversations/<history-id>/events/search` requests (newest page, then the older page). `control-openhands browser text 'testid=chat-scroll-container'` starts with the prompt and shows `16 actions completed` and `DONE`. The other trigger is the user scrolling to the top, driven without a model on the dummy profile from Preconditions (`qa-zeta` active; in a keyed run create and activate it with that same `llm set` command for this bullet, then re-run `control-openhands llm preset deepseek` and remove it with `control-openhands api DELETE /api/profiles/qa-zeta --write`, so later families still count two profiles; arrange, not proof). Run `control-openhands conversation start --prompt "qa older 1"` (no `--wait`; note `<older-id>`), then `for N in $(seq 2 30); do control-openhands conversation send <older-id> --prompt "qa older $N"; done`: every send is accepted while the agent sits in error, about 1.5 s apiece (42 s for the 29 here), and each refused message adds five events (a `MessageEvent`, three `ConversationStateUpdateEvent`s, a `ConversationErrorEvent`), so the conversation ends with about 150 events (152 here). `control-openhands conversation events <older-id> --last 500` reports `"total": 152`, `"pages": 2`, `"more": false` (both server pages read, see Gotchas); its first row is the `SystemPromptEvent` and the first `MessageEvent` is `qa older 1`, while `--last 25` reports `"pages": 1`, `"more": true` and starts at `qa older 26`. Run `control-openhands browser network --clear`, `control-openhands browser reload`, `control-openhands browser wait 'testid=chat-input'`, then `control-openhands browser network --filter 'events/search' --last 10`: two requests, `?limit=50&sort_order=TIMESTAMP_DESC` and one with `timestamp__lt=<ts of qa older 21>`, because the newest 50 events hold only ten bubbles, shorter than the viewport, so the automatic fetch ran once and then stopped. `control-openhands browser count 'testid=user-message'` is `20`, `control-openhands browser text 'testid=chat-scroll-container'` starts `qa older 11` and holds none of `qa older 1`…`qa older 10`, and `control-openhands browser bbox 'testid=chat-scroll-container'` has `scrollHeight` `1576` against `clientHeight` `814`: taller than the viewport, so the third page waits for the user. Run `control-openhands browser scroll 'testid=chat-scroll-container' --by -100000`: `control-openhands browser network --filter 'events/search' --last 5` gains a request with `timestamp__lt=<ts of qa older 11>`, `control-openhands browser count 'testid=user-message'` is `29` and the text starts `qa older 2` (that page held messages 2–10 plus the tail of the first message's events; not seen in two runs: if the first refusal ever logs six events instead of seven, the arithmetic gives count `30`, text `qa older 1` and a last page of only the `SystemPromptEvent`), and the view stays on `qa older 11`: `control-openhands browser eval "document.querySelector('[data-testid=chat-scroll-container]').scrollTop"` is `684`, the height of the prepended rows. Scroll to the top again the same way: a fourth request (`timestamp__lt=` the oldest loaded event), the count is `30`, the text starts `qa older 1` and `scrollTop` is `76` (`control-openhands browser screenshot 'testid=chat-interface' --feature F06.load-older-history --name scroll-top-loaded-qa-older-1`). A third scroll adds no request: the short page ended the paging. `control-openhands browser count 'testid=loading-older-events'` is `0` after each scroll because the `Fetching older messages…` row lasts a few tens of milliseconds on a local stack; to see it, scroll with `control-openhands browser scroll 'testid=chat-scroll-container' --by -100000 --observe 'testid=loading-older-events' --observe-ms 1500` instead (both scrolls were driven this way here, with the same counts), whose `observed` goes `<absent>` at `0` ms, `Fetching older messages…` at about 20 ms and `<absent>` again within about 40 to 120 ms (`116` ms on the first scroll, `39` on the second; `observedBy` `mutation`); the `timestamp__lt=` network rows remain the durable proof. Return to the 16-command conversation for the next bullet: `control-openhands browser goto /conversations/<history-id>` and `control-openhands browser wait 'testid=chat-input'`.
- **Scroll to bottom (`F06.scroll-to-bottom`).** On the 16-command conversation (`<history-id>`; in a model-free run this bullet is blocked with the automatic-fetch half, prerequisite `DEEPSEEK_API_KEY`), `control-openhands browser click 'testid=event-group-toggle'` makes the transcript taller than the viewport while the view stays where it was. Run `control-openhands browser scroll 'testid=chat-scroll-container' --by 2000` first (the button only reacts to scroll events, so a page that is already at the top never shows it): `count 'testid=scroll-to-bottom'` is `0`. Run `control-openhands browser scroll 'testid=chat-scroll-container' --by -2000`; the count is `1`. `control-openhands browser click 'testid=scroll-to-bottom'`, then the count is `0` and `control-openhands browser eval "(()=>{const e=document.querySelector('[data-testid=chat-scroll-container]');return Math.round(e.scrollHeight-e.scrollTop-e.clientHeight)})()"` is `0`.
-- **Confirmation mode on (`F06.confirmation-mode`).** Arrange through the settings UI, in this order (the Security Analyzer field appears only once Confirmation Mode is on): `control-openhands browser goto /settings/verification`, `control-openhands browser click 'testid=verification-settings-screen >> text=Confirmation Mode'`, `control-openhands browser click 'testid=sdk-section-all-toggle'`, `control-openhands browser click 'testid=verification-settings-screen >> role=button[name="Show suggestions"]'`, `control-openhands browser click 'role=listbox >> role=option[name="None"]'`, `control-openhands browser value 'testid=verification-settings-screen >> role=combobox[name="Security Analyzer"]'` (`None`), `control-openhands browser click 'testid=verification-settings-screen >> testid=save-button'`; `control-openhands api GET /api/settings` shows `"confirmation_mode": true` and `"security_analyzer": "none"` (None means every action is confirmed; the default LLM analyzer only stops HIGH-risk actions, so `echo` would run unprompted). Run `control-openhands conversation start --prompt "Run the terminal command 'echo qa-confirm-ok' and reply with its output." --wait --timeout 180`: it returns `"status": "waiting_for_confirmation"`. `control-openhands browser text 'testid=chat-scroll-container'` ends `Do you want to continue with this action? Cancel Ctrl+⇧+⌫ Continue Ctrl+↩` on Linux (`Cancel ⇧⌘⌫ Continue ⌘↩` on macOS, #18127), `control-openhands browser tooltip 'testid=action-confirm-button'` is `Confirm the requested action`, and a lock icon sits above the composer (screenshot). The tooltip stays open over the Cancel button, so move the pointer first: `control-openhands browser hover 'testid=chat-input'`. Run `control-openhands browser click 'testid=action-reject-button'`: status `idle`, the buttons are gone, `conversation events <id> --last 6` ends with `UserRejectObservation`, and the action's row reads `Rejected` (`control-openhands browser count 'testid=rejected-indicator'` is `1`) with no spinner. Send `Run the terminal command 'echo qa-confirm-two' and reply with its output.` through `testid=chat-input` / `testid=submit-button`, `control-openhands browser wait 'testid=action-confirm-button' --timeout 20000`, `control-openhands browser click 'testid=action-confirm-button'`; about 15 s later the status is `finished` and the events contain the observation `qa-confirm-two`. A rejection inside a group resolves it too (#17920, fixed by #17995): send `Run the terminal command 'echo qa-g1'. After it finishes, run the terminal command 'echo qa-g2' as a separate tool call. Then reply DONE.`, continue the first action (`browser wait 'testid=action-confirm-button'`, then `control-openhands browser press Control+Enter`, which is also the next bullet's shortcut) and cancel the second (`browser wait 'testid=action-confirm-button' --timeout 60000`, `control-openhands browser press Control+Shift+Backspace`): status `idle`, `control-openhands browser text 'testid=event-group >> nth=-1'` ends `2 actions completed`, `control-openhands browser count 'testid=event-group >> nth=-1 >> .animate-spin'` is `0`, the same after `browser reload`, and the expanded group (`control-openhands browser click 'testid=event-group >> nth=-1 >> button >> nth=0'`) shows `Rejected` on the `qa-g2` row (`browser screenshot 'testid=chat-interface' --feature F06.confirmation-mode --name group-after-reload`). At phone width (`browser viewport phone`) the hints should fit their buttons. Known failure (reproduced 2026-10-13, #18395): the labels wrap inside the 26 px buttons (`scrollHeight` 40), so Cancel's `Ctrl+⇧+⌫` line falls below the button, invisible, and Continue's hint spills below it; the same wrap shows on desktop with the side panel open, and at `553d4519` with the ⌘ hints (screenshot the awaiting prompt at phone width).
+- **Confirmation mode on (`F06.confirmation-mode`).** Arrange through the settings UI, in this order (the Security Analyzer field appears only once Confirmation Mode is on): `control-openhands browser goto /settings/verification`, `control-openhands browser click 'testid=verification-settings-screen >> text=Confirmation Mode'`, `control-openhands browser click 'testid=sdk-section-all-toggle'`, `control-openhands browser click 'testid=verification-settings-screen >> role=button[name="Show suggestions"]'`, `control-openhands browser click 'role=listbox >> role=option[name="None"]'`, `control-openhands browser value 'testid=verification-settings-screen >> role=combobox[name="Security Analyzer"]'` (`None`), `control-openhands browser click 'testid=verification-settings-screen >> testid=save-button'`; `control-openhands api GET /api/settings` shows `"confirmation_mode": true` and `"security_analyzer": "none"` (None means every action is confirmed; the default LLM analyzer only stops HIGH-risk actions, so `echo` would run unprompted). Run `control-openhands conversation start --prompt "Run the terminal command 'echo qa-confirm-ok' and reply with its output." --wait --timeout 180`: it returns `"status": "waiting_for_confirmation"`. `control-openhands browser text 'testid=chat-scroll-container'` ends `Do you want to continue with this action? Cancel Ctrl+⇧+⌫ Continue Ctrl+↩` on Linux (`Cancel ⇧⌘⌫ Continue ⌘↩` on macOS, #18127), `control-openhands browser tooltip 'testid=action-confirm-button'` is `Confirm the requested action`, and a lock icon sits above the composer (screenshot). The tooltip stays open over the Cancel button, so move the pointer first: `control-openhands browser hover 'testid=chat-input'`. Run `control-openhands browser click 'testid=action-reject-button'`: status `idle`, the buttons are gone, `conversation events <id> --last 6` ends with `UserRejectObservation`, and the action's row reads `Rejected` (`control-openhands browser count 'testid=rejected-indicator'` is `1`) with no spinner. Send `Run the terminal command 'echo qa-confirm-two' and reply with its output.` through `testid=chat-input` / `testid=submit-button`, `control-openhands browser wait 'testid=action-confirm-button' --timeout 20000`, `control-openhands browser click 'testid=action-confirm-button'`; about 15 s later the status is `finished` and the events contain the observation `qa-confirm-two`. A rejection inside a group resolves it too (#17920, fixed by #17995): send `Run the terminal command 'echo qa-g1'. After it finishes, run the terminal command 'echo qa-g2' as a separate tool call. Then reply DONE.`, continue the first action (`browser wait 'testid=action-confirm-button'`, then `control-openhands browser press Control+Enter`, which is also the next bullet's shortcut) and cancel the second (`browser wait 'testid=action-confirm-button' --timeout 60000`, `control-openhands browser press Control+Shift+Backspace`): status `idle`, `control-openhands browser text 'testid=event-group >> nth=-1'` ends `2 actions completed`, `control-openhands browser count 'testid=event-group >> nth=-1 >> .animate-spin'` is `0`, the same after `browser reload`, and the expanded group (`control-openhands browser click 'testid=event-group >> nth=-1 >> button >> nth=0'`) shows `Rejected` on the `qa-g2` row (`browser screenshot 'testid=chat-interface' --feature F06.confirmation-mode --name group-after-reload`). Each button keeps its label and hint on one line: while the prompt is waiting, `control-openhands browser eval "[...document.querySelectorAll('[data-testid=action-reject-button],[data-testid=action-confirm-button]')].map(b=>b.offsetHeight+'/'+b.scrollHeight).join(' ')"` is `26/26 26/26` at desktop, also with the side panel open (`testid=right-panel-toggle`). At phone width (`browser viewport phone`) the shortcut hints are hidden (no hardware keyboard), the buttons read `Cancel` and `Continue`, and `browser text 'testid=chat-scroll-container'` ends `Do you want to continue with this action? Cancel Continue` (screenshot the awaiting prompt at phone width).
- **Confirmation shortcuts (`F06.confirmation-shortcuts`).** The previous bullet resolved its action, so arrange a new one in the same conversation: send `Run the terminal command 'echo qa-confirm-three' and reply with its output.` through `testid=chat-input` / `testid=submit-button` and `control-openhands browser wait 'testid=action-confirm-button' --timeout 20000`. `control-openhands browser text 'testid=action-confirm-button'` is `Continue Ctrl+↩` off macOS. Ctrl works off macOS (#17928, fixed by #18127): the previous bullet's `Control+Enter` continued and `Control+Shift+Backspace` cancelled. `control-openhands browser press Meta+Enter` continues too: `control-openhands conversation wait <id> --fresh --timeout 120` ends `finished` and `control-openhands conversation events <id> --kinds ObservationEvent` contains `qa-confirm-three`. Restore afterwards: `browser goto /settings/verification`, click `testid=sdk-section-all-toggle`, `Show suggestions`, `role=listbox >> role=option[name="LLM"]`, then `testid=verification-settings-screen >> text=Confirmation Mode`, then `save-button`; `api GET /api/settings` shows `"confirmation_mode": false`, `"security_analyzer": "llm"`.
- **Branch from here (`F06.branch-from-here`).** Open a conversation with at least two exchanges (the OK/PONG one above): `control-openhands browser goto /conversations/<id>` with its id (the previous bullet left `/settings/verification`). Agent message: `control-openhands browser hover 'testid=agent-message >> nth=0'`, `control-openhands browser click 'testid=agent-message >> nth=0 >> role=button[name="Branch from here"]' --expect-url '/conversations/(?!<id>)'`; after `control-openhands browser wait 'testid=user-message'` (the history mounts a moment after the URL changes, so an immediate count reads `0`) the new conversation has `count 'testid=user-message'` `1`, the agent message `OK`, and `browser eval "document.title"` contains `(branch)`. User message (edit): back on the OK/PONG conversation (`control-openhands browser goto /conversations/<id>`; the branch above has one user message), `control-openhands browser hover 'testid=user-message >> nth=1'` and `control-openhands browser click 'testid=user-message >> nth=1 >> role=button[name="Branch from here"]' --observe '[data-testid=chat-input]' --observe-ms 4000`: `observed` shows the composer change from empty to `Reply with only the word PONG.` and stay. The new conversation (`<branch-id>` from `browser url`) omits that message, `control-openhands browser text 'testid=chat-input'` is `Reply with only the word PONG.`, and nothing was sent: `control-openhands conversation events <branch-id> --kinds MessageEvent` lists only the copied first exchange (count `2`: the long message that starts `Reply with only the word OK.`, then `OK`; no `PONG`). `control-openhands browser enabled 'testid=submit-button'` should be `true` with the prefilled text; today it is `false` until the text is edited or the page is reloaded (see Gotchas). Send it with `control-openhands browser press Enter --selector 'testid=chat-input'` (Enter sends whatever the Send button's state is) and `control-openhands conversation wait <branch-id> --fresh --timeout 120`: expected `finished` with a reply; today it ends `error` (Agent Server bug, see Gotchas).
- **Images (`F06.image-attachments`).** Run `control-openhands fixture image --name qa-f06-img` (prints the PNG path), `control-openhands browser upload 'testid=upload-image-input' <png path>` (a thumbnail appears in the composer), fill `testid=chat-input` with `Reply with only the word SEEN.` and click `testid=submit-button`. The user bubble contains `image-carousel` / `image-preview`. Run `control-openhands browser hover 'testid=image-preview'`, `control-openhands browser click 'testid=expand-image-button'` (`count 'testid=image-lightbox'` is `1`), `control-openhands browser press Escape` (`0`), click `expand-image-button` again and `control-openhands browser click 'testid=image-lightbox-close'` (`0`).
diff --git a/.agents/skills/verify-openhands/references/feature-map/F07-conversation-page.md b/.agents/skills/verify-openhands/references/feature-map/F07-conversation-page.md
index c7a07cc069..992bf86aa5 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F07-conversation-page.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F07-conversation-page.md
@@ -59,7 +59,7 @@ Preconditions:
- Desktop viewport unless a bullet says otherwise.
- **Open by URL (`F07.open-by-url`).** Run `control-openhands browser goto /conversations/<id>`, `control-openhands browser wait 'testid=chat-pane-header'`, `control-openhands browser text 'testid=conversation-name-title'` and `control-openhands browser count 'testid=chat-interface >> testid=agent-message'`. The title is the generated one and the count is `1`. `control-openhands browser eval "localStorage.getItem('openhands-last-conversation-by-backend')"` reads `{"default-local":"<id>"}`.
-- **Leaving an automation setup conversation (`F07.open-by-url`).** A conversation tagged `automationsetup: draft` (started from `/automations/setup`, see `F04.automation-setup-tag`) opens as a setup page (`testid=automation-setup-topbar`, no `chat-pane-header`). Make an empty conversation: `control-openhands browser goto /`, `control-openhands browser click 'testid=conversation-panel-new-thread-picker'`, `control-openhands browser click 'testid=launch-no-workspace' --expect-url '/conversations/'` (note `<plain-id>`; `control-openhands api GET /api/conversations/<plain-id> --pick tags` is only `clientsource`). Open the setup conversation from the sidebar (`control-openhands browser click 'a[href*="/conversations/<setup-id>"] >> nth=0' --expect-url '/conversations/<setup-id>'`), then go back the same way (`'a[href*="/conversations/<plain-id>"] >> nth=0'`). Expected: `<plain-id>` opens as a normal conversation (`browser count 'testid=chat-pane-header'` is `1`) and its tags are unchanged. Known failure (reproduced 2026-10-13, 3 of 3 at desktop, #18378): the setup panel stays mounted for the first render of the new id and tags it, so `api GET /api/conversations/<plain-id> --pick tags` gains `automationsetup: draft`, `browser count 'testid=automation-setup-topbar'` is `1` and the sidebar row gains the `Automation setup: Draft` chip, also after a reload (`browser screenshot --feature F07.open-by-url --name plain-became-setup`). It triggers when the setup conversation is in this tab's `openhands-automation-setup-handoffs` sessionStorage entry (`control-openhands browser storage --session --values`) and the other conversation is already loaded in the page; it did not reproduce at phone width (0 of 2). Remove the tag with F04's Edit tags (`remove-tag-automationsetup`) to restore the conversation.
+- **Leaving an automation setup conversation (`F07.open-by-url`).** A conversation tagged `automationsetup: draft` (started from `/automations/setup`, see `F04.automation-setup-tag`) opens as a setup page (`testid=automation-setup-topbar`, no `chat-pane-header`). Make an empty conversation: `control-openhands browser goto /`, `control-openhands browser click 'testid=conversation-panel-new-thread-picker'`, `control-openhands browser click 'testid=launch-no-workspace' --expect-url '/conversations/'` (note `<plain-id>`; `control-openhands api GET /api/conversations/<plain-id> --pick tags` is only `clientsource`). Open the setup conversation from the sidebar (`control-openhands browser click 'a[href*="/conversations/<setup-id>"] >> nth=0' --expect-url '/conversations/<setup-id>'`), then go back the same way (`'a[href*="/conversations/<plain-id>"] >> nth=0'`). Expected: `<plain-id>` opens as a normal conversation: `control-openhands browser wait 'testid=chat-pane-header'`, `browser count 'testid=chat-pane-header'` is `1`, `browser count 'testid=automation-setup-topbar'` is `0`, and `api GET /api/conversations/<plain-id> --pick tags` is still only `clientsource`, also after `browser reload` and after repeating the round trip; the sidebar row of `<plain-id>` has no `Automation setup: Draft` chip (`browser screenshot --feature F07.open-by-url --name plain-stays-plain`). The setup conversation is in this tab's `openhands-automation-setup-handoffs` sessionStorage entry during the check (`control-openhands browser storage --session --values`). The same holds at phone width through the sidebar drawer (`sidebar-mobile-menu-toggle`, then the rows inside `testid=sidebar-mobile-drawer`).
- **Unknown id (`F07.missing-redirect`).** Run `control-openhands browser goto /conversations/00000000-0000-0000-0000-000000000000`, `control-openhands browser wait-url '/conversations$'` and `control-openhands browser toasts`. A toast reads `This conversation does not exist, or you do not have permission to access it. If this is your conversation, try switching to the workspace where it was created.` and the last-conversation slot is `{}`. It is the only toast: `control-openhands browser toasts --history` has no entry with `does not understand`.
- **Rename inline (`F07.rename-inline`).** On `/conversations/<id>` (`control-openhands browser goto /conversations/<id>`; the unknown id redirected to `/conversations`) run `control-openhands browser dblclick 'testid=conversation-name-title'`, `control-openhands browser fill 'testid=conversation-name-input' 'QA_F07 renamed'`, `control-openhands browser press Enter`, `control-openhands browser toasts` (`Conversation title updated successfully`), then `control-openhands browser reload` and `control-openhands browser text 'testid=conversation-name-title'`. It reads `QA_F07 renamed`, and so does `control-openhands browser text 'testid=conversation-card >> testid=conversation-card-title'`.
- **Rename from the menu, empty value (`F07.rename-inline`).** Run `control-openhands browser click 'testid=chat-pane-header >> testid=ellipsis-button'`, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=rename-button'`, `control-openhands browser fill 'testid=conversation-name-input' ''` and `control-openhands browser press Enter`. The title stays `QA_F07 renamed` and no toast appears.
@@ -73,7 +73,7 @@ Preconditions:
- **Hooks (`F07.hooks-modal`).** On `/conversations/<running-id>` (no workspace attached; Paused conversation left the browser there) open the "..." menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-hooks-button'` and `control-openhands browser text 'testid=hooks-modal'`. It reads `Available Hooks`, the workspace explanation and `No hooks configured for this conversation.` (the backend workspace root has no `hooks.json`). Run `control-openhands browser network --clear` and `control-openhands browser click 'testid=hooks-modal >> testid=refresh-hooks'`: the text stays and `control-openhands browser network` shows `POST /api/hooks`. `control-openhands browser click 'testid=close-hooks-modal'` closes it.
- **Skills (`F07.skills-modal`).** Open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-skills-button'`, `control-openhands browser snapshot 'role=dialog'` (`Public skills 12` and one button per skill), `control-openhands browser click 'testid=skills-modal >> role=button[name="github Knowledge"]'`; the row expands to `Triggers` and `Content`. `control-openhands browser press Escape` closes it (`browser count 'testid=skills-modal'` is `0`).
- **Project skills and configured hooks (`F07.skills-modal-project`, `F07.hooks-modal`).** Arrange (outside the UI) a project skill and a hook in the git fixture: `mkdir -p <run>/workspace/qa-f07-repo/.agents/skills/qa-f07-skill <run>/workspace/qa-f07-repo/.openhands`, write `<run>/workspace/qa-f07-repo/.agents/skills/qa-f07-skill/SKILL.md` with front matter `name: qa-f07-skill`, `description: QA fixture skill for F07.`, `triggers: [qaf07trigger]` and the body `QA_F07 skill body.`, and write `<run>/workspace/qa-f07-repo/.openhands/hooks.json` as `{"pre_tool_use":[{"matcher":"terminal","hooks":[{"type":"command","command":"true","timeout":10}]}]}`. Then `control-openhands conversation start --workspace qa-f07-repo --prompt "Reply with the single word: hi" --wait --timeout 240` prints `<hooks-id>`; `control-openhands api GET /api/conversations/<hooks-id>` shows that `hook_config.pre_tool_use` holds the `terminal` matcher. On `/conversations/<hooks-id>` open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-skills-button'` and `control-openhands browser snapshot 'role=dialog'`: `Project skills 1` with `button "qa-f07-skill AgentSkills"` above `Public skills 12`. `control-openhands browser click 'testid=skills-modal >> role=button[name="qa-f07-skill AgentSkills"]'` expands it to `Triggers qaf07trigger Content QA_F07 skill body.`; `control-openhands browser click 'testid=skills-modal >> testid=refresh-skills'` re-reads the workspace (`browser network` shows `POST /api/skills`). Close with `control-openhands browser press Escape`. Hooks: open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-hooks-button'` and `control-openhands browser snapshot 'testid=hooks-modal'`: one collapsed event, `button "Pre Tool Use 1 hook"`. `control-openhands browser click 'testid=hooks-modal >> role=button[name="Pre Tool Use 1 hook"]'` expands it to `Matcher terminal Commands true Type: command Timeout: 10s` (`control-openhands browser screenshot --feature F07.hooks-modal --name workspace-hooks`). Close it with `control-openhands browser click 'testid=close-hooks-modal'` and write `<run>/workspace/qa-f07-repo/.openhands/hooks.json` as `{"pre_tool_use":[{"matcher":"terminal","hooks":[{"type":"command","command":"true","timeout":10}]}],"stop":[{"hooks":[{"type":"command","command":"echo qa-f07-stop"}]}]}`. A reopen keeps the list it read: run `control-openhands browser network --clear`, open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-hooks-button'` again, then `control-openhands browser network --filter /api/hooks` (no request) and `control-openhands browser snapshot 'testid=hooks-modal'` (still only `button "Pre Tool Use 1 hook"`). Refresh re-reads the file: `control-openhands browser click 'testid=hooks-modal >> testid=refresh-hooks'`, then `control-openhands browser network --filter /api/hooks` shows `POST /api/hooks` and the snapshot adds `button "Stop 1 hook"`; `control-openhands browser click 'testid=hooks-modal >> role=button[name="Stop 1 hook"]'` expands it to `Matcher * Commands echo qa-f07-stop Type: command Timeout: 60s` (the SDK defaults). Close with `control-openhands browser click 'testid=close-hooks-modal'`.
-- **Agent tools (`F07.agent-tools-modal`).** Open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-agent-tools-button'`; the dialog `testid=system-message-modal` has tabs `System Message` (selected, the system prompt) and `Available Tools`. `control-openhands browser click 'role=dialog >> role=tab[name="Available Tools"]'`, `control-openhands browser count 'testid=system-message-modal >> testid=toggle-button'` (backend-provided count; `9` on Agent Server 1.54.0: terminal, file_editor, task_tracker, canvas_ui_control, launch_child_conversation, finish, think, switch_llm, invoke_skill), `control-openhands browser click 'testid=system-message-modal >> testid=toggle-button >> nth=0'` expands `terminal` with its description. Close with `control-openhands browser click 'testid=close-system-message-modal'`.
+- **Agent tools (`F07.agent-tools-modal`).** Open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-agent-tools-button'`; the dialog `testid=system-message-modal` has tabs `System Message` (selected, the system prompt) and `Available Tools`. `control-openhands browser click 'role=dialog >> role=tab[name="Available Tools"]'`, `control-openhands browser count 'testid=system-message-modal >> testid=toggle-button'` (backend-provided count; `10` on Agent Server 1.54.0: terminal, file_editor, task_tracker, canvas_ui_control, launch_child_conversation, request_secret (local backends, see `F14.agent-request`), finish, think, switch_llm, invoke_skill), `control-openhands browser click 'testid=system-message-modal >> testid=toggle-button >> nth=0'` expands `terminal` with its description. Close with `control-openhands browser click 'testid=close-system-message-modal'`.
- **Export transcript (`F07.export-transcript`).** Open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=export-transcript-button'`; `control-openhands browser snapshot 'role=dialog'` shows `Export conversation`, radio `Markdown document (.md)` checked, `Standalone web page (.html)`, and checkboxes `Include tool call details` and `Include timestamps` checked. `control-openhands browser click 'testid=confirm-transcript-export'` then `control-openhands browser downloads --last 1 --inspect`: a `<ms>-conversation-<id>.md` (the harness prefixes a timestamp to the suggested `conversation-<id>.md`) lands in `<run>/private/downloads` starting `# QA_F07 renamed`, `**Model:** deepseek/deepseek-flash`, `## User` with `<sub>` timestamps. Reopen, `control-openhands browser click 'role=dialog >> role=radio[name="Standalone web page (.html)"]'`, `control-openhands browser click 'role=dialog >> role=checkbox[name="Include timestamps"]'`, export: the `.html` has `<title>QA_F07 renamed</title>` and no timestamps. Reopen once more: `control-openhands browser click 'testid=cancel-transcript-export'` closes it without a download (the file count in `<run>/private/downloads` is unchanged); `browser press Escape` and `testid=close-transcript-export-modal` close it too.
- **Download data (`F07.download-zip`).** Open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=download-trajectory-button'`, then `control-openhands browser downloads`. A `<ms>-conversation_<id>.zip` lands in `<run>/private/downloads`; `browser downloads --last 1 --inspect` lists `<id-without-dashes>/base_state.json` and `bash_events/`, and every LLM `api_key` in `base_state.json` is masked as asterisks (`**********`).
- **Usage (`F07.display-cost`).** Open the menu, `control-openhands browser click 'testid=conversation-name-context-menu >> testid=display-cost-button'`, `control-openhands browser wait 'testid=usage-panel'` and `control-openhands browser attr 'testid=right-panel-toggle' aria-pressed` (`true`). The panel shows Context Window, Token Usage and Total Cost (contents are F27).
@@ -105,7 +105,8 @@ Preconditions:
- The Open Workspace folder browser has no path field: walk it with `folder-browser-up` and `folder-browser-entry-<name>`. `fixture git-repo` puts the repo in `<run>/workspace`, three levels above the `workspace` favorite.
- The repo, branch and Pull/Push chips come from `git remote get-url origin` in the workspace, read when the page loads: reload after changing the remote.
- The Usage tab logs `GET /api/llm/balance` 404 on a local backend (F27); count it in `browser errors --app-only` but it is not an F07 failure.
-- The overview's Automations section has no opener in the UI (the overview menu offers Workspace and Git sections only).
+- The Overview drawer has no Automations, Skills, MCP or Secrets sections. The Overview menu offers Workspace and Git rows only, and its drawer opens only Pull Requests (`F27.overview-drawer`).
+- Dialogs move focus inside when they open and keep Tab inside: the Skills and Hooks dialogs start on their close button, Stop and Delete on `cancel-button`, Export on `close-transcript-export-modal`, Agent Tools on `close-system-message-modal`. In the Hooks dialog, focusing Refresh with Tab opens its tooltip, which takes the first Escape; a second Escape closes the dialog.
- The Hooks dialog shows the workspace's `hooks.json` as of its last read (its first open for the conversation after a page load, or Refresh), not the hooks the conversation started with: after the hooks steps above, `browser goto /conversations/<git-id>` (started before the file existed; its `hook_config` is `null`) and opening the dialog lists both events too. A reopen within 5 minutes of that read shows the same list and sends no request: click Refresh after editing `hooks.json`.
- A prompt- or agent-type hook shows its text under Commands (the `prompt`, or an agent hook's `system_prompt`). With the Hooks dialog closed (the hooks steps above end that way; otherwise `control-openhands browser click 'testid=close-hooks-modal'`), write `<run>/workspace/qa-f07-repo/.openhands/hooks.json` as `{"stop":[{"hooks":[{"type":"prompt","prompt":"QA_F07 prompt hook"}]}],"session_start":[{"hooks":[{"type":"agent","system_prompt":"QA_F07 agent hook"}]}]}`, run `control-openhands browser goto /conversations/<hooks-id>`, open the menu and `control-openhands browser click 'testid=conversation-name-context-menu >> testid=show-hooks-button'`, click `testid=hooks-modal >> testid=refresh-hooks`, then `role=button[name="Session Start 1 hook"]` and `role=button[name="Stop 1 hook"]` inside `testid=hooks-modal` (an event expanded before a Refresh stays expanded, and a second click collapses it, so start from a reopened dialog). In `control-openhands browser snapshot 'testid=hooks-modal'` they expand, in that order, to `Matcher * Commands QA_F07 agent hook Type: agent Timeout: 60s` and `Matcher * Commands QA_F07 prompt hook Type: prompt Timeout: 60s`, and `browser text 'testid=hooks-modal >> pre'` returns `QA_F07 agent hook` and `QA_F07 prompt hook` (`control-openhands browser screenshot --feature F07.hooks-modal --name prompt-agent-hooks`). An agent hook without `system_prompt` shows no box: with `{"session_start":[{"hooks":[{"type":"agent"}]}]}`, close, reopen and Refresh, `Session Start 1 hook` expands to `Matcher * Commands Type: agent Timeout: 60s` and `browser count 'testid=hooks-modal >> pre'` is `0`.
- An agent hook shows its `system_prompt`, the text it runs with, even when it also carries a `prompt`, and an agent hook with only a `prompt` shows no box (#18206, fixed by #18215). With `{"session_start":[{"hooks":[{"type":"agent","prompt":"QA_F07 stray prompt","system_prompt":"QA_F07 agent system prompt"}]}],"stop":[{"hooks":[{"type":"agent","prompt":"QA_F07 prompt only"}]}]}` in `hooks.json`, close, reopen and Refresh the dialog, then expand both events: `Session Start 1 hook` reads `Matcher * Commands QA_F07 agent system prompt Type: agent Timeout: 60s`, `Stop 1 hook` reads `Matcher * Commands Type: agent Timeout: 60s`, and `control-openhands browser count 'text=QA_F07 stray prompt'` and `browser count 'text=QA_F07 prompt only'` are `0` (`browser screenshot --feature F07.hooks-modal --name agent-hook-system-prompt`). Agent hooks in `hooks.json` really run on the next turn in that workspace (the session-start one shows a `Hook: SessionStart failed` row), so write the recipe's two-event `hooks.json` (`pre_tool_use` and `stop`) back and Refresh afterwards.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F08-workspace-files-and-changes.md b/.agents/skills/verify-openhands/references/feature-map/F08-workspace-files-and-changes.md
index f4dcea6c51..c91b49c5d2 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F08-workspace-files-and-changes.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F08-workspace-files-and-changes.md
@@ -81,7 +81,7 @@ Preconditions:
- **Tree (`F08.files-tree`).** Run `control-openhands browser click 'testid=files-tab-refresh'` first (see Gotchas: files the agent made with bash are missing until a refresh). Then `control-openhands browser attr 'testid=file-tree-dir-src' aria-expanded` (`false`), `control-openhands browser click 'testid=file-tree-dir-src'` (now `true`) and `control-openhands browser click 'testid=file-tree-file-src/calc.py'`. `control-openhands browser text 'testid=file-content-viewer-highlighted'` shows the 4 lines `def add`…`return a - b`, and `testid=file-quick-row-item-src/calc.py` exists.
- **Tree toggle (`F08.files-tree-toggle`).** Run `control-openhands browser bbox 'testid=files-tab-tree'` (width `224`). Then `control-openhands browser click 'testid=file-quick-row-tree-toggle'`, `control-openhands browser attr 'testid=file-quick-row-tree-toggle' aria-label` (`Show file tree`) and `control-openhands browser count 'testid=files-tab-tree'` (`0`). Reload and reopen the drawer: the count is still `0`. Click the toggle again to restore (count `1`).
- **Tree resize (`F08.files-tree-resize`).** Run `control-openhands browser drag 'testid=files-tab-tree-resize-handle >> .cursor-ew-resize' --by 100,0`; `browser bbox 'testid=files-tab-tree'` is `324` wide and `control-openhands browser eval "localStorage.getItem('files-tab-tree-width')"` is `"324"`. `--by 500,0` clamps at `480`, `--by -600,0` at `160`, and the width survives `browser reload` plus the toggle. Restore with `--by 64,0` (`224`).
-- **File discovery (`F08.files-discovery-settings`).** With the drawer open on Files (open it with `right-panel-toggle` and click `testid=conversation-tab-files` only if Files is not already the active tab: clicking the active tab closes the drawer), run `control-openhands browser click 'testid=file-discovery-settings'` and `control-openhands browser snapshot 'role=dialog[name="File discovery"]'`: `Applies to this workspace on this backend, including other conversations using it.`, textbox `Excluded directories` (`.git`, `node_modules`, `.venv`, … `target`), spinbutton `Maximum files (0 for unlimited)` `2000`, checkbox `Include symbolic links to files`, buttons `Restore defaults`, `Cancel`, `Save`. With `D='role=dialog[name="File discovery"]'`: `control-openhands browser fill "$D >> role=spinbutton" '-1'` makes `control-openhands browser enabled "$D >> role=button[name=\"Save\"]"` `false`; `2000` makes it `true`. Run `control-openhands browser fill "$D >> role=textbox" "$(printf '.git\nnode_modules\nsrc')"` and click Save: the dialog closes, `control-openhands api GET /api/settings` holds `workspace_file_discovery` keyed by the workspace path with `excludedPatterns` `.git`, `node_modules`, `src`, and after `control-openhands browser click 'testid=files-tab-refresh'` `control-openhands browser count 'testid=file-tree-dir-src'` is `0`, also after `browser reload` and reopening the drawer (`browser screenshot --feature F08.files-discovery-settings --name dialog` before saving). Restore: reopen the dialog, click `Restore defaults` (the textbox returns to the default list; the dialog stays open), then `Save`; after Refresh the `file-tree-dir-src` count is `1`. `control-openhands browser press Escape` closes the dialog without saving. At phone width the dialog should fit the viewport. Known failure (reproduced 2026-10-13, #18396): on the 390 px `/panel` page `browser bbox 'role=dialog[name="File discovery"]'` is 520 px wide (x -65 to 455), so the title, the textbox and Save are cut off; `workspace-file-discovery-settings.tsx` (lines 55-58) uses `ModalBody width="md"` without `max-w-[90vw]`.
+- **File discovery (`F08.files-discovery-settings`).** With the drawer open on Files (open it with `right-panel-toggle` and click `testid=conversation-tab-files` only if Files is not already the active tab: clicking the active tab closes the drawer), run `control-openhands browser click 'testid=file-discovery-settings'` and `control-openhands browser snapshot 'role=dialog[name="File discovery"]'`: `Applies to this workspace on this backend, including other conversations using it.`, textbox `Excluded directories` (`.git`, `node_modules`, `.venv`, … `target`), spinbutton `Maximum files (0 for unlimited)` `2000`, checkbox `Include symbolic links to files`, buttons `Restore defaults`, `Cancel`, `Save`. With `D='role=dialog[name="File discovery"]'`: `control-openhands browser fill "$D >> role=spinbutton" '-1'` makes `control-openhands browser enabled "$D >> role=button[name=\"Save\"]"` `false`; `2000` makes it `true`. Run `control-openhands browser fill "$D >> role=textbox" "$(printf '.git\nnode_modules\nsrc')"` and click Save: the dialog closes, `control-openhands api GET /api/settings` holds `workspace_file_discovery` keyed by the workspace path with `excludedPatterns` `.git`, `node_modules`, `src`, and after `control-openhands browser click 'testid=files-tab-refresh'` `control-openhands browser count 'testid=file-tree-dir-src'` is `0`, also after `browser reload` and reopening the drawer (`browser screenshot --feature F08.files-discovery-settings --name dialog` before saving). Restore: reopen the dialog, click `Restore defaults` (the textbox returns to the default list; the dialog stays open), then `Save`; after Refresh the `file-tree-dir-src` count is `1`. `control-openhands browser press Escape` closes the dialog without saving. At phone width the dialog fits the viewport: on the 390 px `/panel` page (`control-openhands browser viewport phone`, `browser goto /conversations/<id>/panel`, `browser click 'testid=conversation-tab-files'`, then `file-discovery-settings`) `control-openhands browser bbox 'testid=file-discovery-settings-modal'` is about 351 px wide at x about 20 with `insideViewport` `true` (520 px at desktop), and `browser screenshot --feature F08.files-discovery-settings --name phone-fits` shows the title, the textbox and `Restore defaults`, `Cancel`, `Save` whole. Measure the `file-discovery-settings-modal` test id: `role=dialog` is the full-screen backdrop. Focus starts in the dialog (the Excluded directories textbox) and Tab cycles its fields and buttons without leaving it.
- **Rich/Plain (`F08.files-rich-plain`).** Click `testid=file-tree-file-qa-notes.md`; `testid=file-content-viewer-markdown` exists. Click `testid=file-tree-file-qa-page.html`, then run `control-openhands browser attr 'testid=file-content-viewer-iframe' sandbox` (`allow-same-origin`) and `control-openhands browser eval "document.querySelector('[data-testid=file-content-viewer-iframe]').contentDocument.body.innerText"` (`QA page`). Run `control-openhands browser click 'testid=files-tab-content-mode-toggle-option-plain'`; `control-openhands browser text 'testid=files-tab-content >> testid=file-content-viewer-highlighted'` reads `1<h1>QA page</h1>`. After `browser reload` and reopening, `control-openhands browser attr 'testid=files-tab-content-mode-toggle-option-plain' aria-checked` is `true`. Switch back with `testid=files-tab-content-mode-toggle-option-rich`.
- **Open in new window (`F08.files-open-new-window`).** With `qa-page.html` selected, run `control-openhands browser click 'testid=files-tab-open-in-new-window'` and `control-openhands browser tabs`. Page 1's URL is `/api/conversations/<id>/workspace/qa-page.html?v=<n>`. Run `control-openhands browser tab 1`, `control-openhands browser text 'body'` (`QA page`) and `control-openhands browser close-tab 1`.
- **Open-file strip (`F08.files-open-tabs`).** With several files opened, run `control-openhands browser attr 'testid=file-quick-row-item-qa-page.html' title` (the path). Select `qa-notes.md` with `control-openhands browser click 'testid=file-quick-row-item-qa-notes.md'`, then close it with `control-openhands browser click 'testid=file-quick-row-close-qa-notes.md'`. `control-openhands browser eval "[...document.querySelectorAll('[data-testid^=file-quick-row-item-]')].filter(e=>e.getAttribute('aria-selected')==='true').map(e=>e.dataset.testid)"` names the neighbour. Run `browser reload`, reopen the drawer and `control-openhands browser text 'testid=file-quick-row'`: the same files remain. Opening a different conversation (see Commits empty state) shows no quick-row items: `browser text 'testid=file-quick-row'` is empty. `browser testids` omits strip tabs scrolled out of view; count them with `browser eval "[...document.querySelectorAll('[data-testid^=file-quick-row-item-]')].length"`.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F10-llm-profiles.md b/.agents/skills/verify-openhands/references/feature-map/F10-llm-profiles.md
index a756799372..7a055960ae 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F10-llm-profiles.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F10-llm-profiles.md
@@ -75,7 +75,7 @@ Preconditions:
- **Cloud provider pagination (`F10.cloud-provider-pagination`).** Blocked: needs a Cloud organization backend where the user can manage profiles (local backends serve the whole list in one `GET /api/llm/providers`). There, the Basic provider list comes from `providers/search` one page at a time (`control-openhands browser network --filter providers/search` lists one request per page, the later ones with a `page_id`), and `xai` and `openrouter`, which sort past the first page, are offered and pick like any other provider. Record `control-openhands evidence add --feature F10.cloud-provider-pagination --result blocked ...` naming the Cloud account.
- **Create the first profile (`F10.create`, `F10.name-autofill`).** In a fresh editor pick deepseek as above, then `control-openhands browser click 'testid=llm-model-input'` and `control-openhands browser click 'testid=model-item-deepseek-flash'`. `control-openhands browser value 'testid=profile-name-input'` now follows the model: `deepseek-flash` (right after the provider pick, while the model is still empty, it still reads `gpt-5.6-sol`). Run `control-openhands browser fill 'testid=profile-name-input' QA_flash`, `control-openhands browser fill 'testid=llm-api-key-input' --value-file <key file>` (or `--value-env DEEPSEEK_API_KEY`), `control-openhands browser click 'testid=save-profile-btn' --observe 'testid=save-profile-btn' --observe-ms 4000` and `control-openhands browser wait-text 'Profile "QA_flash" created' --timeout 60000`. The observed button states are `Save`, `Validating...`, `Saving...`, then the editor closes. After `control-openhands browser reload`, `control-openhands browser text '[data-testid=profile-row]:has([title="QA_flash"])'` reads `QA_flash`, `deepseek/deepseek-flash`, `Default`, and `control-openhands api GET /api/profiles` has `active_profile` `QA_flash`.
- **A typed name sticks; a cleared one follows again (`F10.name-autofill`).** In a fresh editor (`control-openhands browser click 'testid=add-llm-profile'`) run `control-openhands browser fill 'testid=profile-name-input' QA_typed`, then pick deepseek as above, `control-openhands browser click 'testid=llm-model-input'` and `control-openhands browser click 'role=option[name="deepseek-v4-pro"][exact]'`; `browser value 'testid=profile-name-input'` stays `QA_typed`. Run `control-openhands browser fill 'testid=profile-name-input' ''`, then `browser click 'testid=llm-model-input'` and `browser click 'testid=model-item-deepseek-flash'`: the name is `deepseek-flash`; pick `role=option[name="deepseek-v4-pro"][exact]` again and it is `deepseek-v4-pro`.
-- **Name rules (`F10.name-validation`).** In the editor run `control-openhands browser fill 'testid=profile-name-input' 'QA bad'`; `control-openhands browser enabled 'testid=save-profile-btn'` is `false` and `control-openhands browser attr 'testid=profile-name-input' aria-invalid` is `true` (the same for `-QA` and a 65-character name), and the rule text under the field turns to the theme's danger colour: `control-openhands browser eval "(()=>{const i=document.querySelector('[data-testid=profile-name-input]');return getComputedStyle(document.getElementById(i.getAttribute('aria-describedby'))).color+' '+getComputedStyle(document.body).getPropertyValue('--oh-color-danger')})()"` reads `rgb(231, 106, 94) #e76a5e` on OpenHands-Neutral and `rgb(209, 52, 56) #D13438` on Light+ (`F16.color-theme`) (#18199); a valid name turns it back to the muted grey. An existing name (`QA_flash`) also gives `enabled` `false` and `aria-invalid` `true`, with a role=alert message under the field in the danger colour while the rule text stays grey. Expected: the message says an LLM profile with that name exists. Known failure (reproduced 2026-10-13, #18383): it reads `A meta-profile with this name already exists. Choose a different name, or edit the existing one.`, the Model Router's wording reused by #18144 (`browser eval "[...document.querySelectorAll('[role=alert]')].map(a=>a.innerText+' | '+getComputedStyle(a).color)"`; screenshot `--feature F10.name-validation --name duplicate-neutral`). `QA_flash2` gives `true` and no message.
+- **Name rules (`F10.name-validation`).** In the editor run `control-openhands browser fill 'testid=profile-name-input' 'QA bad'`; `control-openhands browser enabled 'testid=save-profile-btn'` is `false` and `control-openhands browser attr 'testid=profile-name-input' aria-invalid` is `true` (the same for `-QA` and a 65-character name), and the rule text under the field turns to the theme's danger colour: `control-openhands browser eval "(()=>{const i=document.querySelector('[data-testid=profile-name-input]');return getComputedStyle(document.getElementById(i.getAttribute('aria-describedby'))).color+' '+getComputedStyle(document.body).getPropertyValue('--oh-color-danger')})()"` reads `rgb(231, 106, 94) #e76a5e` on OpenHands-Neutral and `rgb(209, 52, 56) #D13438` on Light+ (`F16.color-theme`) (#18199); a valid name turns it back to the muted grey. An existing name (`QA_flash`) also gives `enabled` `false` and `aria-invalid` `true`, with a role=alert message under the field in the danger colour while the rule text stays grey. The message reads `An LLM profile with this name already exists. Choose a different name, or edit the existing one.` (`browser eval "[...document.querySelectorAll('[role=alert]')].map(a=>a.innerText+' | '+getComputedStyle(a).color)"`; screenshot `--feature F10.name-validation --name duplicate-neutral`). `QA_flash2` gives `true` and no message.
- **Validation failure (`F10.create-validation`).** In a fresh editor choose deepseek / `model-item-deepseek-flash`, fill `profile-name-input` with `QA_nokey`, leave the API key empty and run `control-openhands browser click 'testid=save-profile-btn' --observe 'testid=save-profile-btn' --observe-ms 3000` (states `Save`, `Validating...`, `Save`). `control-openhands browser wait-text 'Authentication Fails'` finds the toast `litellm.AuthenticationError: AuthenticationError: DeepseekException - Authentication Fails (...)`; `browser text 'testid=profile-editor-title'` is still `Add LLM Profile` and `control-openhands api GET /api/profiles` has no `QA_nokey`.
- **Custom model (`F10.custom-model`).** In that editor run `control-openhands browser click 'testid=sdk-section-advanced-toggle'`. `browser value 'testid=llm-custom-model-input'` carries the Basic choice (`deepseek/deepseek-flash`; placeholder `openai/gpt-5.6-sol`), and `browser attr 'testid=base-url-input' placeholder` is `https://api.openai.com`. Run `control-openhands browser fill 'testid=llm-custom-model-input' ''`, `control-openhands browser fill 'testid=profile-name-input' QA_adv`, `control-openhands browser click 'testid=save-profile-btn'` and `control-openhands browser wait-text 'Model is required'`. Then `control-openhands browser fill 'testid=llm-custom-model-input' deepseek/deepseek-flash`, `control-openhands browser fill 'testid=base-url-input' https://api.deepseek.com`, fill the key as in Create, Save, and `control-openhands browser wait-text 'Profile "QA_adv" created' --timeout 60000`. `api GET /api/profiles` shows `QA_adv` with `base_url` `https://api.deepseek.com`.
- **Second profile from a verified model.** Repeat Create choosing `control-openhands browser click 'role=option[name="deepseek-v4-pro"][exact]'` (verified models have no `model-item-*` test ID) and the name `QA_pro`. The toast reads `Profile "QA_pro" created`; the row `QA_pro deepseek/deepseek-v4-pro` has no badge.
@@ -88,7 +88,7 @@ Preconditions:
- **Basic re-save keeps a hidden Base URL (`F10.basic-save-keeps-base-url`).** `QA_adv` carries `base_url` `https://api.deepseek.com` (Custom model above). Open its row menu, click `testid=profile-actions-menu >> testid=profile-edit`, then `control-openhands browser click 'testid=sdk-section-advanced-toggle'`: `control-openhands browser value 'testid=base-url-input'` is `https://api.deepseek.com`. Run `control-openhands browser click 'testid=sdk-section-basic-toggle'`: `control-openhands browser count 'testid=base-url-input'` is `0` (the view hides it), `control-openhands browser value 'testid=llm-model-input'` is `deepseek-flash` and `control-openhands browser enabled 'testid=save-profile-btn'` is `false` (`control-openhands browser screenshot --feature F10.basic-save-keeps-base-url --name basic-view`). Without touching the model, run `control-openhands browser fill 'testid=llm-api-key-input' --value-env DEEPSEEK_API_KEY` (Save turns `true`), `control-openhands browser click 'testid=save-profile-btn' --observe 'testid=save-profile-btn' --observe-ms 4000` (`Save`, `Validating...`, `Saving...`) and `control-openhands browser wait-text 'Profile "QA_adv" updated' --timeout 60000`. Then `control-openhands api GET /api/profiles/QA_adv --pick config.base_url` is still `https://api.deepseek.com` and `control-openhands api GET /api/profiles/QA_adv --pick config.model` is `deepseek/deepseek-flash`; after `control-openhands browser reload` the same GETs agree. Without a key, arrange `QA_DUMMY_KEY=qa control-openhands llm set --profile QA_base --model deepseek/deepseek-chat --api-key-env QA_DUMMY_KEY --base-url https://api.deepseek.com --no-validate --no-activate` (it prints `activated` `false` and `control-openhands api GET /api/profiles --pick active_profile` is unchanged: the Default stays where it is and the editors below inherit it, as long as some profile is Default; on a run where `QA_base` is the only profile the page makes it Default at the next load, see Gotchas) and `control-openhands browser reload`, then drive the same steps on `QA_base`'s row with its own readbacks: its editor opens on Advanced (the Base URL is a non-default value) with `control-openhands browser value 'testid=base-url-input'` `https://api.deepseek.com`, Basic hides it (`count` `0`), `control-openhands browser value 'testid=llm-model-input'` is `deepseek-chat` and `control-openhands browser enabled 'testid=save-profile-btn'` is `false`; fill the key with `QA_DUMMY_KEY=qa-dummy-key control-openhands browser fill 'testid=llm-api-key-input' --value-env QA_DUMMY_KEY` (Save turns `true`) and Save: the observed states are `Save`, `Validating...` (`[disabled] [busy]`), `Save` within a second, `control-openhands browser toasts` has the provider's message `litellm.BadRequestError: DeepseekException - {"error":{"message":"Authentication Fails, Your api key: ****-key is invalid …`, `browser text 'testid=profile-editor-title'` is still `Edit LLM Profile`, `control-openhands api GET /api/profiles/QA_base --pick config.base_url` is still `https://api.deepseek.com` and `--pick config.model` is `deepseek/deepseek-chat`; record `blocked` with `DEEPSEEK_API_KEY` as the prerequisite, leave with `control-openhands browser click 'testid=cancel-profile-btn'` and remove the arranged profile with `control-openhands api DELETE /api/profiles/QA_base --write` (arrange, not proof).
- **Leave without saving (`F10.editor-discard`).** Edit `QA_adv`, `control-openhands browser click 'testid=sdk-section-all-toggle'`, `control-openhands browser fill 'testid=sdk-settings-llm.top_p' 0.5` (Save turns enabled), then `control-openhands browser click 'testid=back-to-profiles'`. The list returns at once (`browser count 'testid=add-llm-profile'` is `1`) with no prompt (`control-openhands browser events --kinds dialog` is empty), and reopening Edit → All shows `top_p` empty. `testid=cancel-profile-btn` behaves the same.
- **Schema field validation and re-apply (`F10.schema-validation`, `F10.edit-default-reapply`).** Edit the Default profile and run `control-openhands browser click 'testid=sdk-section-all-toggle'`. `control-openhands browser fill 'testid=sdk-settings-llm.temperature' 5`, Save, then `control-openhands browser wait-text 'must be at most 2'` (toast `Temperature must be at most 2`). Clear it, fill `testid=sdk-settings-llm.top_p` with `3`: `Top P must be at most 1`. Clear it, fill `testid=sdk-settings-llm.extra_headers` with `{bad`: `Invalid JSON for Extra Headers`. Clear that, fill temperature `0.3`, Save, and `wait-text 'Profile "<name>" updated'`. `control-openhands api GET /api/settings` then has `agent_settings.llm.temperature` `0.3`, and reopening Edit → All shows `0.3`.
-- **New profile prefill (`F10.create-prefill`).** With that Default (temperature `0.3`) run `control-openhands browser click 'testid=add-llm-profile'`. The editor opens on All (`browser eval "[...document.querySelectorAll('[role=tab]')].map(t=>t.innerText+':'+t.getAttribute('aria-selected')).join(',')"` is `Basic:false,Advanced:false,All:true`), `browser value 'testid=llm-custom-model-input'` is the backend default `openai/gpt-5.6-sol`, and `browser value 'testid=sdk-settings-llm.temperature'` is `0.3`, inherited from the Default. Run `control-openhands browser fill 'testid=llm-custom-model-input' deepseek/deepseek-flash`, `browser fill 'testid=profile-name-input' QA_inherit`, fill the key as in Create, Save and `wait-text 'Profile "QA_inherit" created' --timeout 60000`. Expected: the saved profile has the temperature the form showed. Known failure (reproduced 2026-10-13, #17955): `control-openhands api GET /api/profiles/QA_inherit` has `config.temperature` `null`, and after `browser reload` Edit `QA_inherit` → All shows the temperature empty (the create path merges only dirty fields over an empty base, so the shown inherited values are dropped). Screenshot `--feature F10.create-prefill --name edit-after-create`.
+- **New profile prefill (`F10.create-prefill`).** With that Default (temperature `0.3`) run `control-openhands browser click 'testid=add-llm-profile'`. The editor opens on All (`browser eval "[...document.querySelectorAll('[role=tab]')].map(t=>t.innerText+':'+t.getAttribute('aria-selected')).join(',')"` is `Basic:false,Advanced:false,All:true`), `browser value 'testid=llm-custom-model-input'` is the backend default `openai/gpt-5.6-sol`, and `control-openhands assert --field value --equals 0.3 -- browser value 'testid=sdk-settings-llm.temperature'` passes: the temperature is inherited from the Default. Run `control-openhands browser fill 'testid=llm-custom-model-input' deepseek/deepseek-flash`, `browser fill 'testid=profile-name-input' QA_inherit`, fill the key as in Create, `control-openhands browser scroll 'testid=sdk-settings-llm.temperature'`, `control-openhands browser screenshot --feature F10.create-prefill --name inherited-before-save`, Save and `wait-text 'Profile "QA_inherit" created' --timeout 60000`. The saved profile has the temperature the form showed: `control-openhands api GET /api/profiles/QA_inherit --pick config.temperature` is `0.3`. After `browser reload`, open `QA_inherit`'s row menu (`'[data-testid=profile-row]:has([title="QA_inherit"]) >> testid=profile-menu-trigger'`), click `testid=profile-actions-menu >> testid=profile-edit` and `testid=sdk-section-all-toggle`, `control-openhands browser scroll 'testid=sdk-settings-llm.temperature'` and `control-openhands assert --field value --equals 0.3 -- browser value 'testid=sdk-settings-llm.temperature'` passes (screenshot `--feature F10.create-prefill --name edit-after-create`); leave with `testid=cancel-profile-btn`. Repeat at `browser viewport phone` after `control-openhands api DELETE /api/profiles/QA_inherit --write` (the desktop pass already took the name); the same holds.
- **Link to a provider connection (`F10.connection-link`, `F10.list-grouped`).** Arrange a connection through the F11 form: `control-openhands browser click 'testid=add-provider-connection'`, `control-openhands browser fill 'testid=provider-connection-name-input' QA_ds_conn`, `control-openhands browser click 'testid=provider-connection-provider-input'`, `control-openhands browser fill 'testid=provider-connection-provider-input' deepseek`, `control-openhands browser click 'testid=provider-item-deepseek'`, `control-openhands browser fill 'testid=provider-connection-api-key-input' --value-file <key file>`, `control-openhands browser click 'testid=provider-connection-submit'` and `control-openhands browser wait-text 'Connection "QA_ds_conn" created'`. Edit `QA_adv`, click `testid=sdk-section-basic-toggle`, run `control-openhands browser click 'testid=llm-provider-connection-input'` (options `None`, `QA_ds_conn`) and `control-openhands browser click 'role=option[name="QA_ds_conn"]'`; `browser count 'testid=llm-api-key-input'` is `0`, and on the Advanced tab `browser count 'testid=base-url-input'` is `0`. Picking `role=option[name="None"]` brings the key field back (`1`). Re-select `QA_ds_conn`, then `control-openhands browser click 'testid=save-profile-btn' --observe 'testid=save-profile-btn' --observe-ms 3000`: states `Save`, `Saving...` (no `Validating...`), and `wait-text 'Profile "QA_adv" updated'`. After a reload `control-openhands browser eval "[...document.querySelectorAll('[data-testid=profile-group-header]')].map(e=>e.innerText).join(' | ')"` is `QA_DS_CONN | NOT LINKED` (CSS upper-case) and `control-openhands browser count 'div:has(> [data-testid=profile-group-header]:text-is("QA_ds_conn")) [data-testid=profile-row]:has([title="QA_adv"])'` is `1` (screenshot `--feature F10.list-grouped --name grouped`).
- **Add models from a provider (`F10.add-models`).** Run `control-openhands browser click 'testid=add-models-from-provider'`; `browser enabled 'testid=add-models-submit'` is `false`. Keyless: `control-openhands browser select 'testid=add-models-provider' moonshot` and `control-openhands browser wait 'testid=add-models-select-all'`; `browser text 'testid=add-models-keyless-note'` is `No connection — profiles will need a key added later`, rows start unchecked (`Add 0 profiles`), and `control-openhands browser check 'testid=add-models-select-all'` makes `browser text 'testid=add-models-submit'` read `Add 12 profiles`; uncheck it again. Linked: `control-openhands browser select 'testid=add-models-provider' deepseek`; the connection field (`testid=add-models-connection-field`, options `No connection (keyless)` and `QA_ds_conn`) is pre-bound to `QA_ds_conn`. Run `control-openhands browser check 'testid=add-models-check-deepseek/deepseek-chat'`, `control-openhands browser check 'testid=add-models-check-deepseek/deepseek-v3'`, `control-openhands browser click 'testid=add-models-submit'` and `control-openhands browser wait-text 'Added 2'` (toast `Added 2 profiles`). The modal closes; after a reload `deepseek-chat` and `deepseek-v3` sit under the `QA_DS_CONN` header and `api GET /api/profiles` shows both with a `provider_connection_id`. Preselect entry: `control-openhands browser click 'testid=provider-connection-row >> has-text=QA_ds_conn >> testid=provider-connection-menu-trigger'` (menu: Bulk add, Edit, Delete), `control-openhands browser click 'testid=provider-connection-actions-menu >> testid=provider-connection-add-models'`; provider `deepseek` and the connection are preset, every remaining row is checked (`Add 10 profiles`), and `browser count 'testid=add-models-row-deepseek/deepseek-chat'` is `0` (already added). Close with `control-openhands browser click 'role=dialog[name="Add models as profiles"] >> role=button[name="Cancel"]'`.
- **Broken link (`F10.broken-link`).** The local server refuses to delete a referenced connection (row menu → `testid=provider-connection-delete` → `testid=delete-provider-connection-confirm` toasts `Provider connection cannot be deleted while it is referenced by LLM profile(s): ...`), so arrange the state directly: `control-openhands api POST /api/profiles/QA_broken --write --data '{"llm":{"model":"deepseek/deepseek-chat","provider_connection_id":"00000000000000000000000000000000"},"include_secrets":false}'`. Then `control-openhands browser reload` and `control-openhands browser text '[data-testid=profile-row]:has([title="QA_broken"])'` reads `QA_broken deepseek/deepseek-chat Broken link`; `control-openhands browser attr '[data-testid=profile-row]:has([title="QA_broken"]) >> testid=profile-broken-connection-badge' title` is `The linked provider connection was deleted. Edit this profile to re-link it.` (screenshot `--feature F10.broken-link --name badge`). Edit it: `browser value 'testid=llm-provider-connection-input'` is the orphaned id and the listbox offers `None`, `QA_ds_conn` and that id. The row's group header shows the raw id, not a name (see Gotchas).
@@ -110,7 +110,7 @@ Preconditions:
- An open provider or model listbox intercepts clicks on the editor's buttons (`cancel-profile-btn`, `save-profile-btn` time out with "subtree intercepts pointer events"): `browser press Escape` first.
- Provider and model options render in a portal outside the editor and modals: select `testid=provider-item-<name>` unscoped. Only verified providers have `provider-item-*` IDs and only Other models have `model-item-*` IDs; use `role=option[name="..."][exact]` for the rest. `browser testids --filter model-item` lists only items scrolled into view; `browser count` finds the rest.
- The auto-filled Profile Name never reuses an existing profile's name: picking a model whose name is already a profile's (for example `deepseek-flash` after `llm preset deepseek`) empties the field, which disables Save. Recipes fill `profile-name-input` with a `QA_*` name anyway.
-- A duplicate profile name is flagged since #18144 (closed #17938; verified 2026-10-13 on desktop, phone, OpenHands-Neutral and Light+): `aria-invalid` `true` and a taken-name alert. That alert reuses the Model Router's `A meta-profile with this name already exists. …` (Known failure, #18383; see `F10.name-validation`). A checkout without #18144 disables Save silently.
+- A duplicate profile name is flagged: `aria-invalid` `true` and a taken-name alert. The alert has its own wording, `An LLM profile with this name already exists. …`.
- The Rename modal does not check duplicates client-side (Rename stays enabled); the server's 409 comes back as the toast `Profile '<name>' already exists`, and `browser errors --app-only` counts one `http-error` and one `console.error` for it.
- Set as default (and deleting the Default) rewrites the agent's active LLM settings (`control-openhands llm show`). Deleting every profile leaves the last applied model and key in agent settings.
- The app never leaves profiles without a Default: `src/routes/root-layout.tsx` runs `useEnsureActiveProfile` (`src/hooks/use-ensure-active-profile.ts`), which activates the first profile with a key (else the first profile) whenever profiles exist but none is Default, whether never activated or because the Default was deleted (this is what promotes a profile in `F10.delete-default`). So `llm set --no-activate` keeps the Default only while another profile is Default: on a run whose only profile is the arranged one, `control-openhands api GET /api/profiles --pick active_profile` flips from `null` to it at the next page load (`browser reload`) and the agent's LLM settings follow (`llm show`).
@@ -119,60 +119,5 @@ Preconditions:
- The All tab logs many `[i18n] Missing translation for key "SCHEMA$LLM$..."` console warnings; they are not page errors.
- The list follows the server's order, which sorts by profile file name: `QA_pro-copy-1` lists before `QA_pro-copy` (`-` sorts before `.json`'s `.`). Do not assert a strict alphabetical order on names that are prefixes of each other.
- Toasts last only a few seconds. A `browser screenshot` taken after a slow command can miss them, so assert with `browser wait-text` or `browser toasts`.
-- Known issue #17955: a new profile is saved without the inherited LLM options the editor shows, such as temperature (`F10.create-prefill`).
+- A new profile keeps the inherited LLM options the editor shows, such as temperature.
- `Subscription models are not loaded yet.` is translated in every language (`SETTINGS$SUBSCRIPTION_MODELS_NOT_LOADED`, #18308; source check 2026-10-13, not driven live). It shows when a subscription profile is saved before `GET /api/llm/subscription/openai/models` answers, in the settings save (`src/routes/llm-settings.tsx`) and in onboarding's LLM step. No recipe drives it, since it needs that request to be slow or to fail.
-
- ```sh known-bug F10.create-prefill desktop-inherited
- control-openhands llm set --profile QA_Inherit --model deepseek/deepseek-flash --api-key-env DEEPSEEK_API_KEY --no-activate --no-validate
- control-openhands api DELETE /api/profiles/QA_Inherit --write
- control-openhands api PATCH /api/settings --write --data '{"agent_settings_diff":{"llm":{"temperature":0.4}}}'
- control-openhands browser viewport desktop
- control-openhands browser goto /settings/llm
- control-openhands browser click 'testid=add-llm-profile'
- control-openhands browser click 'testid=sdk-section-all-toggle'
- control-openhands assert --field value --equals 0.4 -- browser value 'testid=sdk-settings-llm.temperature'
- control-openhands browser fill 'testid=llm-custom-model-input' deepseek/deepseek-flash
- control-openhands browser fill 'testid=profile-name-input' QA_Inherit
- control-openhands browser fill 'testid=llm-api-key-input' --value-env DEEPSEEK_API_KEY
- control-openhands browser record start --feature F10.create-prefill --name inherited-save
- control-openhands browser scroll 'testid=sdk-settings-llm.temperature'
- control-openhands browser screenshot --feature F10.create-prefill --name inherited-before-save
- control-openhands browser click 'testid=save-profile-btn'
- control-openhands browser wait-text 'created'
- control-openhands browser click '[data-testid=profile-row]:has([title="QA_Inherit"]) >> testid=profile-menu-trigger'
- control-openhands browser click 'testid=profile-actions-menu >> testid=profile-edit'
- control-openhands browser click 'testid=sdk-section-all-toggle'
- control-openhands browser scroll 'testid=sdk-settings-llm.temperature'
- control-openhands browser screenshot --feature F10.create-prefill --name inherited-after-save
- control-openhands browser record stop --gif
- control-openhands api GET /api/profiles/QA_Inherit --pick config.temperature
- control-openhands assert --field value --equals 0.4 --bug-shows '' -- browser value 'testid=sdk-settings-llm.temperature' # bug
- ```
-
- ```sh known-bug F10.create-prefill phone-inherited
- control-openhands llm set --profile QA_Inherit --model deepseek/deepseek-flash --api-key-env DEEPSEEK_API_KEY --no-activate --no-validate
- control-openhands api DELETE /api/profiles/QA_Inherit --write
- control-openhands api PATCH /api/settings --write --data '{"agent_settings_diff":{"llm":{"temperature":0.4}}}'
- control-openhands browser viewport phone
- control-openhands browser goto /settings/llm
- control-openhands browser click 'testid=add-llm-profile'
- control-openhands browser click 'testid=sdk-section-all-toggle'
- control-openhands assert --field value --equals 0.4 -- browser value 'testid=sdk-settings-llm.temperature'
- control-openhands browser fill 'testid=llm-custom-model-input' deepseek/deepseek-flash
- control-openhands browser fill 'testid=profile-name-input' QA_Inherit
- control-openhands browser fill 'testid=llm-api-key-input' --value-env DEEPSEEK_API_KEY
- control-openhands browser record start --feature F10.create-prefill --name phone-inherited-save
- control-openhands browser scroll 'testid=sdk-settings-llm.temperature'
- control-openhands browser screenshot --feature F10.create-prefill --name phone-inherited-before-save
- control-openhands browser click 'testid=save-profile-btn'
- control-openhands browser wait-text 'created'
- control-openhands browser click '[data-testid=profile-row]:has([title="QA_Inherit"]) >> testid=profile-menu-trigger'
- control-openhands browser click 'testid=profile-actions-menu >> testid=profile-edit'
- control-openhands browser click 'testid=sdk-section-all-toggle'
- control-openhands browser scroll 'testid=sdk-settings-llm.temperature'
- control-openhands browser screenshot --feature F10.create-prefill --name phone-inherited-after-save
- control-openhands browser record stop --gif
- control-openhands api GET /api/profiles/QA_Inherit --pick config.temperature
- control-openhands assert --field value --equals 0.4 --bug-shows '' -- browser value 'testid=sdk-settings-llm.temperature' # bug
- ```
-
diff --git a/.agents/skills/verify-openhands/references/feature-map/F12-model-router.md b/.agents/skills/verify-openhands/references/feature-map/F12-model-router.md
index a667bd6e71..43e977b2a3 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F12-model-router.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F12-model-router.md
@@ -87,7 +87,7 @@ Preconditions:
- Settings nav test ids embed the route: quote `'testid=sidebar-settings-/settings/meta-llm'` as written.
- The empty-state hint about LLM profiles only exists before any profile is saved: drive `F12.empty-states` first on a fresh run.
- The first router you create is activated and turns **Run on first message** on by itself; later routers do neither. Any recipe that creates a router on an empty list changes both.
-- `Run on first message` is a hidden `<input>` (#17900): click the label text inside `testid=meta-profile-run-at-conversation-start` and read `.checked`/`.disabled` with `browser eval`. The stored preference (`misc_settings.app_preferences.run_router_at_conversation_start`) is independent of what the disabled switch shows; with no active router it may still be `true`. Turn it off before deleting the last router, or a later "first router" will not re-enable it because it is already on.
+- `Run on first message` is an `sr-only` checkbox (keyboard reachable): click the label text inside `testid=meta-profile-run-at-conversation-start` and read `.checked`/`.disabled` with `browser eval`. The stored preference (`misc_settings.app_preferences.run_router_at_conversation_start`) is independent of what the disabled switch shows; with no active router it may still be `true`. Turn it off before deleting the last router, or a later "first router" will not re-enable it because it is already on.
- Router templates name 11 models (`gpt-5.4` … `claude-sonnet-4-5`) that will not exist in a sandbox. With a provider connection present, a template preselects it, and saving creates all 11 as LLM profiles linked to that connection unless you pick **Don't create profiles** first (`F12.create-more` does). If a run created them, remove them with `for p in gpt-5.4 gpt-5.5 minimax-m3 claude-opus-4-6 claude-opus-4-7 claude-opus-4-8 claude-fable-5 kimi-k2.5 kimi-k2.6 deepseek-v3.2-reasoner claude-sonnet-4-5; do control-openhands api DELETE /api/profiles/$p --write; done` (they live in the run's private state and only add rows to Settings → LLM).
- Profile matching is case-insensitive and exact by name: table rows are `- <profile name>: description`; existing `deepseek-flash`/`deepseek-pro` are not re-created, and the endpoint is derived as `<provider lowercased>/<name>`.
- The classifier and the model table refer to LLM *profile names*, not model ids. A router whose classifier profile does not exist (for example a template's `minimax-m3` without a connection) saves fine and only fails when a conversation tries to route.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F13-agent-profiles.md b/.agents/skills/verify-openhands/references/feature-map/F13-agent-profiles.md
index 189a6bb8d6..ff1fff2f5a 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F13-agent-profiles.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F13-agent-profiles.md
@@ -55,7 +55,7 @@ Preconditions:
- **List (`F13.list`).** From `/` run `control-openhands browser click 'testid=backend-selector-settings-link' --expect-url '/settings'`, then `control-openhands browser url` (`/settings/agents`) and `control-openhands browser text 'testid=agent-profile-row'`: `default`, its LLM profile `default`, `Default` on three lines. `control-openhands browser count 'testid=agent-profile-active-badge'` is `1`; `control-openhands browser screenshot --feature F13.list --name list`. Command menu: run `control-openhands browser goto /` (the browser is on `/settings/agents`, where the URL wait would pass at once), `control-openhands browser press Control+k`, `control-openhands browser type 'testid=command-menu >> role=combobox' 'Agent settings'`, `control-openhands browser press Enter`, `control-openhands browser wait-url '/settings/agents(\?|$)'`. Legacy URL: `control-openhands browser goto /settings/agent` returns `.../settings/agents`.
- **No LLM profile (`F13.editor-no-llm`).** Before the preset, run `control-openhands browser click 'testid=add-agent-profile'`, then `control-openhands browser text 'testid=agent-profile-no-llm'`: `Create an LLM profile first to define an OpenHands agent.` `control-openhands browser text 'testid=agent-profile-editor-title'` is `Add agent profile` and `control-openhands browser text 'testid=agent-profile-editor-description'` is `Configure the agent below, then click Save to create this profile.` (agent copy, not the LLM page's hint, since #18035). Run `control-openhands browser fill 'testid=agent-profile-name-input' QA_NoLlm`; Save turns enabled (`control-openhands browser enabled 'testid=save-agent-profile-btn'`). Click `testid=save-agent-profile-btn` and run `control-openhands browser toasts`: `Select an LLM profile for this agent.` `api GET /api/agent-profiles` lists no `QA_NoLlm`. Leave with `control-openhands browser click 'testid=cancel-agent-profile-btn'`.
-- **Name rules (`F13.name-validation`).** In the Add editor, fill `testid=agent-profile-name-input` with `''`, `default`, `'bad name!'` and `QA_ok` in turn; after each run `control-openhands browser attr 'testid=agent-profile-name-input' aria-invalid` and `control-openhands browser enabled 'testid=save-agent-profile-btn'`. Results: `''` → `true`/`false`; `default` (duplicate) → `true`/`false`; `'bad name!'` → `true`/`false`; `QA_ok` → `false`/`true`. The rule under the field is always shown (grey) and turns the theme danger colour (`text-danger`, #18199) while the name is invalid; it reads `Profile name must start with a letter or digit, and contain only letters, digits, dots, underscores, or hyphens (max 64 characters)`. For the duplicate, `control-openhands browser count 'role=alert >> text=already exists'` is `1`: the alert is the field's description, in the danger colour (`rgb(231, 106, 94)` on OpenHands-Neutral, `rgb(209, 52, 56)` on Light+), while the rule text stays grey. Expected: it says an agent profile with that name exists. Known failure (reproduced 2026-10-13, #18383): it reads `A meta-profile with this name already exists. Choose a different name, or edit the existing one.`, the Model Router's wording reused by #18144. Take `control-openhands browser screenshot --feature F13.name-validation --name duplicate` with `default` filled. In Edit the same message appears for another profile's name and not for the profile's own name (`F13.edit`).
+- **Name rules (`F13.name-validation`).** In the Add editor, fill `testid=agent-profile-name-input` with `''`, `default`, `'bad name!'` and `QA_ok` in turn; after each run `control-openhands browser attr 'testid=agent-profile-name-input' aria-invalid` and `control-openhands browser enabled 'testid=save-agent-profile-btn'`. Results: `''` → `true`/`false`; `default` (duplicate) → `true`/`false`; `'bad name!'` → `true`/`false`; `QA_ok` → `false`/`true`. The rule under the field is always shown (grey) and turns the theme danger colour (`text-danger`, #18199) while the name is invalid; it reads `Profile name must start with a letter or digit, and contain only letters, digits, dots, underscores, or hyphens (max 64 characters)`. For the duplicate, `control-openhands browser count 'role=alert >> text=already exists'` is `1`: the alert is the field's description, in the danger colour (`rgb(231, 106, 94)` on OpenHands-Neutral, `rgb(209, 52, 56)` on Light+), while the rule text stays grey. It reads `An agent profile with this name already exists. Choose a different name, or edit the existing one.` Take `control-openhands browser screenshot --feature F13.name-validation --name duplicate` with `default` filled. In Edit the same message appears for another profile's name and not for the profile's own name (`F13.edit`).
- **ACP form (`F13.acp-form`).** In the Add editor run `control-openhands browser click 'testid=agent-type-selector'` and `control-openhands browser click 'role=option[name="ACP (external subprocess)"]'`. `control-openhands browser value 'testid=agent-preset-selector'` is `Claude Code`, `browser value 'testid=agent-command-input'` is `npx -y --prefer-offline @agentclientprotocol/[email protected]` and `browser value 'testid=agent-model-selector'` is `Agent default`; `agent-profile-no-llm` is gone (ACP needs no LLM profile). The preset list (`control-openhands browser click 'div:has(> div > input[data-testid=agent-preset-selector]) >> role=button[name="Show suggestions"]'`, then `control-openhands browser snapshot 'role=listbox'`) is `Claude Code`, `Codex`, `Gemini CLI`, `Pi`, `OpenCode`, `Custom` (`browser press Escape`). Pick `Codex` and `Gemini CLI` from `testid=agent-preset-selector`: the command becomes `npx -y --prefer-offline @agentclientprotocol/[email protected]` / `npx -y --prefer-offline @google/[email protected] --acp` and the model `Agent default` / `gemini-2.5-pro` (the models come from the agent since #18131: `F13.acp-model-discovery`). Pick `Custom`: the command is empty and Save is disabled. `control-openhands browser fill 'testid=agent-command-input' 'npx -y qa-custom-acp --stdio'` enables Save. Filling the exact Codex command flips the preset to `Codex`; appending ` --verbose` flips it back to `Custom`. Model override: with a built-in preset, run `control-openhands browser choose 'testid=agent-model-selector' 'Custom'` (or open the model list with its toggle button and `control-openhands browser click 'role=option[name="Custom"][exact]'`); `control-openhands browser count 'testid=agent-model-input'` is `1` and empty. The **Custom** preset shows `agent-model-input` directly (no model list).
- **Create an ACP profile (`F13.create`).** With preset `Custom`, fill `testid=agent-profile-name-input` with `QA_acp_custom`, `testid=agent-command-input` with `'npx -y qa-custom-acp --stdio'` and `testid=agent-model-input` with `qa-model`, click `testid=save-agent-profile-btn`, then `control-openhands browser wait-text 'Profile "QA_acp_custom" created' --timeout 10000`, `control-openhands browser reload` and `control-openhands browser text 'testid=agent-profile-row >> has-text=QA_acp_custom'`: `QA_acp_custom` / `ACP (external subprocess)`. `control-openhands api GET /api/agent-profiles/QA_acp_custom` shows `acp_server` `custom`, the command and `acp_model` `qa-model`.
- **Pi and OpenCode presets (`F13.acp-local-presets`).** For each of `OpenCode` and `Pi`: run `control-openhands browser click 'testid=add-agent-profile'`, `control-openhands browser choose 'testid=agent-type-selector' 'ACP (external subprocess)'` and `control-openhands browser choose 'testid=agent-preset-selector' '<preset>'`. `browser value 'testid=agent-command-input'` is `npx -y --prefer-offline [email protected] acp` / `npx -y --prefer-offline [email protected] --package=@earendil-works/[email protected] pi-acp` (the versions follow `@openhands/typescript-client`). Fill `testid=agent-profile-name-input` with `QA_opencode` / `QA_pi`, run `control-openhands browser wait 'testid=agent-model-discovering' --state detached --timeout 120000`, click `testid=save-agent-profile-btn` and `control-openhands browser wait-text 'Profile "<name>" created' --timeout 10000`. `control-openhands api GET /api/agent-profiles/<name>` has `acp_server` `opencode` / `pi`, `acp_command` `null` and `acp_model` `null`, and after `browser reload` both rows read `<name>` / `ACP (external subprocess)`. Cloud hides both presets (blocked here: needs a Cloud backend).
@@ -82,10 +82,10 @@ Preconditions:
- The editor is not a route: `browser url` stays `/settings/agents` and `browser reload` drops unsaved edits back to the list. The settings page header (`settings-page-subtitle`) is hidden while it is open.
- Autocomplete dropdowns: clicking an input that already shows a value can open an empty listbox (the text filters it), and an option click then times out. The toggle-button opener does not always open the list either: on the Edit editor's `agent-profile-llm-selector` (bottom of the form) the option click timed out 5 of 5 times right after opening the editor, and the model list's `Custom` option once detached mid-click after a preset change. `browser choose` was reliable on every dropdown. A listbox also closes if you spend a command (snapshot, eval) in between, so click the option right after opening.
-- A taken name marks the field invalid with an alert since #18144 (closed #17938; verified 2026-10-13); the alert reuses the Model Router's `A meta-profile with this name already exists. …` (Known failure, #18383; see `F13.name-validation`). A checkout without #18144 disables Save silently.
+- A taken name marks the field invalid with an alert reading `An agent profile with this name already exists. …`.
- `has-text=default` also matches the active row's `Default` badge (Playwright's text match ignores case), so with several profiles `'testid=agent-profile-row >> has-text=default'` hits two rows; select the seeded profile with `'testid=agent-profile-row >> nth=-1'` (it sorts last) or by its badge.
- Model lookups (`F13.acp-model-discovery`) spawn the provider CLI on the agent-server machine through npx with the profile's allowed secrets; they send no prompt. A failed lookup is retried each time the preset is picked again; a successful one is cached for about five minutes per preset and secret set.
-- Switch rows are `<input>`s inside labels (#17900): click the label text (`'testid=agent-settings-tool-list >> text=<tool>'`, `'testid=agent-settings-mcp-list >> text=<name>'`) and read `.checked` with `browser eval`. A bare `label` selector segment is parsed as `label=`, not CSS.
+- Switch rows are `sr-only` checkboxes inside labels (keyboard reachable): click the label text (`'testid=agent-settings-tool-list >> text=<tool>'`, `'testid=agent-settings-mcp-list >> text=<name>'`) and read `.checked` with `browser eval`. A bare `label` selector segment is parsed as `label=`, not CSS.
- The Default badge, the **Set as active** item and the `Switched to profile` toast all refer to the same active pointer; the wording differs from the LLM page (**Set as default**).
- The OpenHands row's grey text is the LLM profile name, not the kind. A fresh run's `default` shows `default` there although no LLM profile of that name exists until it is edited (`F13.stale-llm-ref`).
- Home launches follow the LLM pill: a named profile pinned to another LLM profile launches through the legacy settings path, so its tool, concurrency and MCP settings do not apply (`F13.llm-pill-precedence`). To prove a profile's settings, pin it to the active LLM profile, and read `launched_agent_profile` on the conversation.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F14-secrets.md b/.agents/skills/verify-openhands/references/feature-map/F14-secrets.md
index cddd297102..dcd2126da0 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F14-secrets.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F14-secrets.md
@@ -4,7 +4,7 @@ Secrets are named values the agent can use during conversations. A user lists,
adds, edits and deletes them under Settings → Secrets; saved values are never
shown again, and every new conversation receives them as environment variables.
-Source: `src/routes/secrets-settings.tsx`, `src/components/features/settings/secrets-settings/`, `src/api/secrets-service.ts`.
+Source: `src/routes/secrets-settings.tsx`, `src/components/features/settings/secrets-settings/`, `src/api/secrets-service.ts`, `src/components/features/conversation/secret-request-modal.tsx`, `src/services/secret-request.ts`, `src/stores/secret-request-store.ts`, `src/constants/secret-request.ts`, `src/api/request-secret-client-tool.ts`.
## Sub-features
@@ -19,12 +19,14 @@ Source: `src/routes/secrets-settings.tsx`, `src/components/features/settings/sec
- `F14.rename`: the Edit form's name is editable; renaming with the value left blank keeps the stored value under the new name, and renaming onto an existing name is refused with `Secret already exists`.
- `F14.edit-value`: typing a value in the Edit form (labelled "Secret Value (leave blank to preserve the existing value)") replaces the stored value for new conversations.
- `F14.delete-escape`: Escape closes the delete confirmation without deleting.
+- `F14.agent-request`: in a local conversation the agent's `request_secret` tool opens **The agent needs a secret** over the chat with its reason and the name prefilled; Add secret saves it, attaches it to the running conversation and reports `saved` to the agent, Cancel reports `not_saved`, and a name that is already saved is attached without a dialog. The value never reaches the agent's events.
## How to get to it (user POV)
- Sidebar settings gear (`backend-selector-settings-link`), then **Secrets** in the settings navigation (`sidebar-settings-/settings/secrets`).
- Direct URL `/settings/secrets`.
- Command menu (`Control+k` or `Meta+k`, or `command-menu-trigger` in the sidebar): search "Secrets", choose **Secrets settings**.
+- The agent itself, in a conversation: its `request_secret` client tool opens the request dialog (`testid=secret-request-modal`) on `/conversations/<id>` (`F14.agent-request`).
## Driving it with control-openhands
@@ -32,7 +34,8 @@ Preconditions:
- Baseline state (launched, doctored, `onboard --skip` done).
- No secret named `QA_TMP_SECRET`, `QA_AGENT_SECRET`, `QA_RENAMED_SECRET` or `QA_BACK_SECRET` exists (`control-openhands api GET /api/settings/secrets`).
-- `F14.agent-access`, `F14.rename` and `F14.edit-value` need an active LLM profile (`control-openhands llm preset deepseek`).
+- `F14.agent-access`, `F14.rename`, `F14.edit-value` and `F14.agent-request` need an active LLM profile (`control-openhands llm preset deepseek`).
+- No secret named `QA_REQ_SECRET` or `QA_REQ_CANCEL` exists, for `F14.agent-request`.
- **Open the list (`F14.list`).** Navigate from the sidebar. Run `control-openhands browser click 'testid=backend-selector-settings-link'`, `control-openhands browser click 'testid=sidebar-settings-/settings/secrets'`, then `control-openhands browser count 'testid=secret-item >> has-text=OPENHANDS_AUTOMATION_API_KEY'`. The count is `1` and the URL is `/settings/secrets`.
- **Command-menu entry (`F14.list`).** Run `control-openhands browser goto /` (the previous bullet left `/settings/secrets`, which would make the URL check pass vacuously), then `control-openhands browser press Control+k`, `control-openhands browser type 'testid=command-menu >> role=combobox' Secrets`, `control-openhands browser press Enter`, then `control-openhands browser url`. The URL ends in `/settings/secrets`.
@@ -47,7 +50,21 @@ Preconditions:
- **Back discards the draft (`F14.form-back`).** On `/settings/secrets` run `control-openhands browser click 'testid=add-secret-button'` and `control-openhands browser text 'testid=secret-editor-title'` (`Add a Secret`). Fill `testid=add-secret-form >> testid=name-input` with `QA_BACK_SECRET` and `value-input` with `dummy`, then `control-openhands browser click 'testid=back-to-secrets'`. `control-openhands browser count 'testid=add-secret-form'` is `0`, the list is back (`browser count 'testid=secret-item'` > `0`) and `control-openhands api GET /api/settings/secrets` lists no `QA_BACK_SECRET`. Reopening Add shows an empty `name-input` (`browser value`); leave with `testid=back-to-secrets`. The Edit form shows `Edit a Secret` in the same `secret-editor-title`.
- **Rename keeps the value (`F14.rename`).** Needs `QA_AGENT_SECRET` from Agent receives the secret. Run `control-openhands browser goto /settings/secrets`, `control-openhands browser click 'testid=secret-item >> has-text=QA_AGENT_SECRET >> testid=edit-secret-button'`, fill `testid=edit-secret-form >> testid=name-input` with `OPENHANDS_AUTOMATION_API_KEY` and click `testid=edit-secret-form >> testid=submit-button`: `control-openhands browser snapshot 'testid=edit-secret-form'` shows the paragraph `Secret already exists` and the form stays open. Fill `name-input` with `QA_RENAMED_SECRET` (value left blank), click submit, `control-openhands browser reload`; `browser count 'testid=secret-item >> has-text=QA_RENAMED_SECRET'` is `1`, the `QA_AGENT_SECRET` count is `0`, and `api GET /api/settings/secrets` lists only the new name. Then run `control-openhands conversation start --prompt "Run exactly this command and reply with only its output: python3 -c \"import os; print(os.environ.get('QA_RENAMED_SECRET') == 'qa-dummy-4821', 'QA_AGENT_SECRET' in os.environ)\"" --wait --timeout 240` and `control-openhands conversation events <id> --kinds ObservationEvent`: the observation is `True False`.
- **Replace the value (`F14.edit-value`).** Run `control-openhands browser goto /settings/secrets`, `control-openhands browser click 'testid=secret-item >> has-text=QA_RENAMED_SECRET >> testid=edit-secret-button'` and `control-openhands browser snapshot 'testid=edit-secret-form'` (the value textbox is named `Secret Value (leave blank to preserve the existing value)`). Fill `testid=edit-secret-form >> testid=value-input` with `qa-dummy-7350`; `browser enabled 'testid=edit-secret-form >> testid=submit-button'` is `true`. Click submit; `browser count 'testid=edit-secret-form'` is `0`. Start a conversation with the prompt `Run exactly this command and reply with only its output: python3 -c "import os; print(os.environ.get('QA_RENAMED_SECRET') == 'qa-dummy-7350')"` (quoted as above) and read its `ObservationEvent`: `True`.
-- **Escape cancels delete (`F14.delete-escape`).** Run `control-openhands browser goto /settings/secrets`, `control-openhands browser click 'testid=secret-item >> has-text=QA_RENAMED_SECRET >> testid=delete-secret-button'`, `control-openhands browser press Escape`, then `control-openhands browser count 'testid=confirmation-modal'` (`0`) and `browser count 'testid=secret-item >> has-text=QA_RENAMED_SECRET'` (`1`). Clean up: delete `QA_RENAMED_SECRET` with `delete-secret-button` → `testid=confirmation-modal >> testid=confirm-button`, `browser reload`; `api GET /api/settings/secrets` lists only `OPENHANDS_AUTOMATION_API_KEY`.
+- **Agent requests a secret (`F14.agent-request`).** Run `control-openhands browser goto /`, then `control-openhands conversation start --prompt "Call the request_secret tool with name QA_REQ_SECRET and reason 'QA check: a dummy token for verification.' Do not run any other tools. After calling it, end your turn and wait." --wait --timeout 240` (prints `<req-id>`; it leaves the browser on the conversation). `control-openhands conversation events <req-id> --kinds ActionEvent,ObservationEvent` shows the `request_secret` action and the observation `Tool call dispatched to client.`; `control-openhands browser snapshot 'role=dialog'` reads `dialog "The agent needs a secret"`, the reason `QA check: a dummy token for verification.`, textbox `Name` holding `QA_REQ_SECRET`, `Value`, `Description Optional`, `Cancel` and `Add secret [disabled]` (`browser screenshot --feature F14.agent-request --name modal`). Focus starts in `name-input` and Tab cycles Name, Value, Description and Cancel without leaving the dialog (`control-openhands browser eval "document.activeElement.dataset.testid"` after each `browser press Tab`). `control-openhands browser press Escape` and `control-openhands browser mouse-click 30 500` leave it open (`browser count 'testid=secret-request-modal'` stays `1`), and so does `control-openhands browser reload`, with the name still prefilled. Run `control-openhands browser fill 'testid=secret-request-modal >> testid=value-input' qa-dummy-5190` and `control-openhands browser click 'testid=secret-request-modal >> testid=submit-button'`: the dialog closes, `control-openhands api GET /api/settings/secrets` lists `QA_REQ_SECRET`, and after `control-openhands conversation wait <req-id> --fresh --timeout 180` `control-openhands conversation events <req-id> --kinds MessageEvent` holds the user message `[secret-request] {"status":"saved","name":"QA_REQ_SECRET"}`, which the chat does not show (`browser count 'text=[secret-request]'` is `0`). The running conversation now has the value: `control-openhands conversation send <req-id> --prompt "Run exactly this command and reply with only its output: python3 -c \"import os; print(os.environ.get('QA_REQ_SECRET') == 'qa-dummy-5190')\"" --wait --timeout 240`, then the last terminal `ObservationEvent` is `True`. Cancel: `browser goto /`, start a second conversation with the same prompt naming `QA_REQ_CANCEL` and reason `QA cancel check.`, fill `value-input` with `should-not-save` and `control-openhands browser click 'testid=secret-request-modal >> testid=cancel-button'`: the dialog closes, `api GET /api/settings/secrets` lists no `QA_REQ_CANCEL`, the agent receives `[secret-request] {"status":"not_saved","name":"QA_REQ_CANCEL","guidance":...}`, and a reload does not bring the dialog back. Already saved: a third conversation asking for `QA_REQ_SECRET` again shows no dialog (`browser count 'testid=secret-request-modal'` `0`) and its events hold `[secret-request] {"status":"saved","name":"QA_REQ_SECRET"}`. Expected: at phone width the dialog fits the viewport (at most 90vw, 351 px at 390 px, `insideViewport` `true`) with the title, every field and both buttons visible. Known failure (reproduced 2026-10-14, #18432): at 390×844 `browser bbox 'testid=secret-request-modal'` is 520 px wide at x -65, `insideViewport` `false`, so the title, the field labels and Cancel are cut off on the left (and 65 px on the right); run the block below. Return with `browser viewport desktop`. `QA_REQ_SECRET` is deleted in the cleanup below.
+
+ ```sh known-bug F14.agent-request phone-width
+ control-openhands browser viewport desktop
+ control-openhands browser goto /
+ control-openhands conversation start --prompt "Call the request_secret tool with name QA_REQ_PHONE and reason 'QA phone check.' Do not run any other tools. After calling it, end your turn and wait." --wait --timeout 240
+ control-openhands browser wait 'testid=secret-request-modal'
+ control-openhands assert --field insideViewport --equals true -- browser bbox 'testid=secret-request-modal'
+ control-openhands browser viewport phone
+ control-openhands browser screenshot --feature F14.agent-request --name phone-width
+ control-openhands assert --field insideViewport --equals true --bug 'insideViewport eq false && box.width eq 520' -- browser bbox 'testid=secret-request-modal' # bug
+ control-openhands browser click 'testid=secret-request-modal >> role=button[name="Cancel"]'
+ control-openhands browser wait 'testid=secret-request-modal' --state detached
+ ```
+- **Escape cancels delete (`F14.delete-escape`).** Run `control-openhands browser goto /settings/secrets`, `control-openhands browser click 'testid=secret-item >> has-text=QA_RENAMED_SECRET >> testid=delete-secret-button'`, `control-openhands browser press Escape`, then `control-openhands browser count 'testid=confirmation-modal'` (`0`) and `browser count 'testid=secret-item >> has-text=QA_RENAMED_SECRET'` (`1`). Clean up: delete `QA_RENAMED_SECRET` and then `QA_REQ_SECRET` (from `F14.agent-request`), each with `delete-secret-button` → `testid=confirmation-modal >> testid=confirm-button`, `browser reload`; `api GET /api/settings/secrets` lists only `OPENHANDS_AUTOMATION_API_KEY`.
## Gotchas
@@ -59,6 +76,7 @@ Preconditions:
- The model is the weak link in the agent checks, not the product: asked to print a secret's length or last characters, deepseek-flash replies "I can't run that one" and runs nothing, and even for the equality check it may refuse to repeat `True` in its reply. Use the equality prompt and read the terminal `ObservationEvent`; `conversation events` truncates message text to about 160 characters.
- The conversation Overview has no Secrets entry point: its "..." menu (`conversation-overview-ellipsis`) offers Workspace and Git sections only.
- There is no empty state to reach safely: `secrets-empty` renders only with zero secrets, which means deleting the launcher-seeded key (see below).
-- Secrets reach only conversations started after the save. Reuse an older conversation and the variable is missing.
+- Secrets saved in Settings reach only conversations started after the save. Reuse an older conversation and the variable is missing. Only `request_secret` (`F14.agent-request`) attaches a secret to a running conversation, and only on a local backend.
+- `request_secret` is registered on every new OpenHands-agent conversation this frontend launches (not on ACP agents or older conversations) and appears in Show Agent Tools & Metadata (see `F07.agent-tools-modal`). The agent learns the outcome from a hidden `[secret-request] ...` user message; read it with `conversation events <id> --kinds MessageEvent`, not from the chat.
- At 390 px the name column truncates long names (`OPEN…`); use `browser text` on the row rather than the screenshot to read a name.
- `OPENHANDS_AUTOMATION_API_KEY` is seeded by the launcher for automations; deleting it breaks F18/F19 runs in the same stack.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F15-agent-behavior-settings.md b/.agents/skills/verify-openhands/references/feature-map/F15-agent-behavior-settings.md
index baa53b36db..234a3ec136 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F15-agent-behavior-settings.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F15-agent-behavior-settings.md
@@ -80,7 +80,7 @@ Preconditions:
## Gotchas
-- Switches are hidden `<input>`s (#17900): `browser testids` lists them only with `--hidden`. Toggle by clicking the label text scoped to the page, read `.checked` with `browser eval`.
+- Switches are `sr-only` checkboxes (keyboard reachable): toggle by clicking the label text scoped to the page, read `.checked` with `browser eval`.
- `text=Enable Critic` is a case-insensitive substring match and hits the help paragraph `Enable critic evaluation for the agent.` first, so the click does nothing. Quote it: `text="Enable Critic"` (same for `text="Enable Iterative Refinement"`).
- The conversation-owned fields have no `verification.` prefix: `sdk-settings-confirmation_mode`, `sdk-settings-security_analyzer`. Critic fields keep it (`sdk-settings-verification.critic_enabled`).
- Saving is tier-scoped: Basic or Advanced saves reset every field the tier hides to its schema default (`F15.view-scoped-save`). Make all edits on All when other non-default values must survive.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F16-application-settings.md b/.agents/skills/verify-openhands/references/feature-map/F16-application-settings.md
index 64916e382c..ce13d1d90f 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F16-application-settings.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F16-application-settings.md
@@ -30,7 +30,7 @@ Source: `src/routes/app-settings.tsx`, `src/components/features/settings/app-set
- `F16.phone`: the form fits a 390 px viewport without horizontal overflow and Save is reachable by scrolling.
- `F16.title-model-clear`: the Title generation model combobox has a clear (×) button, shown on hover; it resets the choice to Automatic (recommended).
- `F16.dropdown-filter`: typing into the Language (or any) combobox filters its options; Escape restores the stored value and leaves Save disabled.
-- `F16.backend-switch`: switching the active backend in the backend selector while this page is open shows the new backend's stored preferences (the footer already names it), also when switching back to a backend the page has already loaded. Known failure, see its bullet.
+- `F16.backend-switch`: switching the active backend in the backend selector while this page is open shows the new backend's stored preferences (the footer already names it), also when switching back to a backend the page has already loaded; an unsaved edit is dropped on switch.
- `F16.browser-scope`: backend-saved preferences (language and the rest of the Save group) follow the user to a fresh browser profile, while theme and checklist visibility return to their defaults there.
## How to get to it (user POV)
@@ -99,7 +99,7 @@ Preconditions:
- **Clear the title model (`F16.title-model-clear`).** Run `control-openhands browser click 'testid=title-llm-profile-input'`, `control-openhands browser click 'role=option >> has-text=deepseek-flash'` (Save enables), `control-openhands browser hover 'testid=title-llm-profile-input'`, then `control-openhands browser click 'testid=title-llm-profile-input >> xpath=../..//button[not(@aria-label)]'`. `control-openhands browser value 'testid=title-llm-profile-input'` reads `Automatic (recommended)` and `control-openhands browser enabled 'testid=submit-button'` is `false`. Without the hover the click times out (`<div ...> intercepts pointer events`).
- **Filter a dropdown by typing (`F16.dropdown-filter`).** Run `control-openhands browser fill 'testid=language-input' Deu` and `control-openhands browser snapshot 'role=listbox'`: the only option is `Deutsch`. Save is briefly `true` while the text is typed; `control-openhands browser press Escape` restores `control-openhands browser value 'testid=language-input'` to `English` and `control-openhands browser enabled 'testid=submit-button'` to `false`.
- **Browser vs backend scope (`F16.browser-scope`).** Choose `Deutsch` in `testid=language-input` and save, choose `Solarized Light` in `testid=color-theme-input`, and hide the checklist with `control-openhands browser click 'testid=show-getting-started-checklist-switch >> xpath=ancestor::label'` (the label text is German now). Then `control-openhands browser reset`, `control-openhands onboard --skip`, `control-openhands browser goto /settings/app`, `control-openhands browser wait 'testid=submit-button'`: `control-openhands browser text 'testid=submit-button'` is `Änderungen speichern` and `control-openhands browser value 'testid=language-input'` is `Deutsch` (backend), while `control-openhands browser value 'testid=color-theme-input'` is `OpenHands-Neutral`, `control-openhands browser eval "localStorage.getItem('openhands-color-theme')"` is `null` and `control-openhands browser count 'testid=sidebar-onboarding-checklist'` is `1` (browser-only, back to defaults). Restore English (the option keeps its English name in the German UI): `control-openhands browser click 'testid=language-input'`, `control-openhands browser click 'role=option[name="English"]'`, `control-openhands browser click 'testid=submit-button'`, `control-openhands browser reload`, `control-openhands browser wait 'testid=submit-button'`; `control-openhands browser text 'testid=submit-button'` reads `Save Changes` and `control-openhands api GET /api/settings --pick misc_settings.app_preferences.language` is `en`.
-- **Backend switch in place (`F16.backend-switch`).** Needs a second backend with a different Git username: the second stack from F25's Preconditions added as `QA_Second` (F25 Add bullet), arranged with `control-openhands api PATCH /api/settings --run "$OH_VERIFY_RUN_2" --write --data '{"misc_settings_diff":{"app_preferences":{"git_user_name":"QA Second Git"}}}'`. With Local active, run `control-openhands browser goto /settings/app` and `control-openhands browser reload`; `browser value 'testid=git-user-name-input'` is `openhands`. Switch without leaving the page: `control-openhands browser click 'testid=backend-selector'`, `control-openhands browser click 'testid=backend-selector >> role=option[name="Connected QA_Second"]'`, `control-openhands browser wait 'testid=environment-switch-overlay' --state detached --timeout 10000`, wait a few seconds: the value is `QA Second Git` and `browser text 'testid=backend-synced-settings-badge'` names QA_Second (a first visit to a backend loads fresh). Switch back the same way to `Connected Local`. Expected: `openhands` beside `synced from Local backend`. Known failure (reproduced 2026-10-13 on main 2bdc0e9d6 and on 553d45191, #18388): the footer names Local but the form still shows `QA Second Git`, and it stays so on further switches until the page remounts (another settings page and back, or a reload). The Git fields are uncontrolled inputs (`defaultValue` in `src/routes/app-settings.tsx`) that do not remount when the switched-to backend's settings come from the cache. A Save after any edit then sends the stale Git identity to the active backend (Local's `git_user_name` became `QA Second Git`, reproduced 2026-10-13). Switch back to Local and remove `QA_Second` afterwards (F25).
+- **Backend switch in place (`F16.backend-switch`).** Needs a second backend with a different Git username: the second stack from F25's Preconditions added as `QA_Second` (F25 Add bullet), arranged with `control-openhands api PATCH /api/settings --run "$OH_VERIFY_RUN_2" --write --data '{"misc_settings_diff":{"app_preferences":{"git_user_name":"QA Second Git"}}}'`. With Local active, run `control-openhands browser goto /settings/app` and `control-openhands browser reload`; `browser value 'testid=git-user-name-input'` is `openhands`. Switch without leaving the page: `control-openhands browser click 'testid=backend-selector'`, `control-openhands browser click 'testid=backend-selector >> role=option[name="Connected QA_Second"]'`, `control-openhands browser wait 'testid=environment-switch-overlay' --state detached --timeout 10000`, wait a few seconds: the value is `QA Second Git` and `browser text 'testid=backend-synced-settings-badge'` names QA_Second (a first visit to a backend loads fresh). Switch back the same way to `Connected Local`: the value is `openhands` beside `synced from Local backend`, and it keeps following the active backend on further switches. The form remounts per backend, so an unsaved edit is dropped: on Local run `control-openhands browser fill 'testid=git-user-name-input' 'QA Local Edit'` (`browser enabled 'testid=submit-button'` is `true`), switch to `QA_Second` (`QA Second Git`, Save `false`) and back (`openhands`, Save `false`); `control-openhands api GET /api/settings --pick misc_settings.app_preferences.git_user_name` is unchanged. At `browser viewport phone` the selector is in the drawer (`testid=sidebar-mobile-menu-toggle`, then `'testid=sidebar-mobile-drawer >> testid=backend-selector'` and `role=option[name="Connected QA_Second"]`; close the drawer with `browser mouse-click 370 400`) and the field follows the same way. Switch back to Local and remove `QA_Second` afterwards (F25).
- After the family, `control-openhands browser errors --app-only` reports `appErrors: 0`. Requests to `registry.npmjs.org` (update check) and Google Fonts fail through the sandbox proxy; they are not app errors.
## Gotchas
diff --git a/.agents/skills/verify-openhands/references/feature-map/F18-skills.md b/.agents/skills/verify-openhands/references/feature-map/F18-skills.md
index c1f593e66f..3cfdb79c66 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F18-skills.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F18-skills.md
@@ -91,7 +91,7 @@ Preconditions:
- `sidebar-skills-link` is labelled **Customize** and lands on `/mcp` on desktop (`/customize` redirects); Skills is one more click. Test ids embed the route: quote `'testid=sidebar-extensions-/skills'`.
- A `browser click` on an **enabled** card toggle is lost when the pointer was not already over it: `pointerenter` swaps the checkmark icon for the X icon between mousedown and mouseup, so Chrome fires no click and no `PATCH /api/settings` goes out (#17941). `browser hover` the toggle first (or click twice). Turning a skill on is not affected. Recorded as `fail` on `F18.toggle`.
-- The card toggle and the facet rows are `role=switch`/`role=checkbox` buttons: read `aria-checked`. The modal switch is a hidden `<input>` (#17900): click the row text and read `.checked` with `browser eval`.
+- The card toggle and the facet rows are `role=switch`/`role=checkbox` buttons: read `aria-checked`. The modal switch is an `sr-only` checkbox (keyboard reachable): click the row text and read `.checked` with `browser eval`.
- Typing writes `?q=` with `replace` after a 300 ms debounce, so read `browser url` a second later; facet clicks push a history entry immediately.
- Facet counts are disjunctive: a row counts skills matching the other groups' selections, so Disabled reads `0` (and is disabled) while Recommended is checked.
- Groups with fewer than two values are hidden (Source and Type on a fresh run) unless the URL selects one of their values.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F21-automations-dashboard.md b/.agents/skills/verify-openhands/references/feature-map/F21-automations-dashboard.md
index 61ef95be01..6e9ddb0e9e 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F21-automations-dashboard.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F21-automations-dashboard.md
@@ -28,7 +28,7 @@ Source: `src/routes/automations-list.tsx`, `src/components/features/automations/
- `F21.add-menu`: the header **Add Automation** menu offers **Create Automation** and **Import automation**; Create Automation opens the guided setup page at `/automations/setup` (F22).
- `F21.empty-state`: with zero automations and zero saved drafts the page shows **No automations configured**, the "Create an automation" instructions, a docs link, **Create Automation** (opens `/automations/setup`, F22, without creating a conversation) and the Recommended automations rail; the view toggle is disabled.
- `F21.empty-rail`: a Recommended automations card on the empty dashboard opens its template setup at `/automations/new/<template-id>` (F22 owns the setup flow).
-- `F21.overview-tiles`: four tiles summarize the loaded org list (expected: the automations the groups list; Known failure (reproduced 2026-10-13, #18381): they also count hidden DRAFT test automations, see Gotchas): Automations (`N active`), Needs attention (`Latest run failed` or `No latest-run failures`; an enabled automation counts when its latest run is Failed or its task outcome is `failed` or `blocked`, while Partial and Needs review do not count), Total runs and Average duration.
+- `F21.overview-tiles`: four tiles summarize the automations the groups list from the loaded org list (hidden DRAFT test automations from a draft Test are left out): Automations (`N active`), Needs attention (`Latest run failed` or `No latest-run failures`; an enabled automation counts when its latest run is Failed or its task outcome is `failed` or `blocked`, while Partial and Needs review do not count), Total runs and Average duration.
- `F21.search`: the search box narrows the visible automations client-side by name, prompt, repository or model, and the saved drafts by name, prompt, kind or trigger; the tiles do not change.
- `F21.filters`: the **Filters** popover filters by status (All / Active / Needs attention / Disabled) and trigger (All / Scheduled / Event-driven), counts non-default choices in a badge and offers **Reset all**; it stays open while the user changes several choices.
- `F21.sort`: Sort (Latest run / Most runs / Name) orders automations inside each group and counts as a non-default choice.
@@ -49,7 +49,7 @@ Source: `src/routes/automations-list.tsx`, `src/components/features/automations/
- `F21.edit`: kebab **Edit** opens the setup page in edit mode (`/automations/setup?automationId=<id>`) prefilled from the automation, with **Test**, **Save** and **Close** in its toolbar; Save sends one PATCH and toasts **Automation updated**, Close returns to `/automations` (field-level editing is F23). For a script (tarball) automation it should summarize the saved bundle (**Existing code bundle**). Known failure (reproduced 2026-10-13, #18341): from the kebab it shows generated starter Python and a default setup script instead; a cold load of the same URL shows the bundle summary.
- `F21.export`: kebab **Export** downloads `<name-slug>.automation.json` with `version`, `kind: "automation"` and the spec.
- `F21.delete`: kebab **Delete** asks `Delete "<name>"? This cannot be undone.`; Cancel and the X keep it, Delete removes it for good, and Escape dismisses the dialog.
-- `F21.saved-drafts`: saved setup drafts (#17242) appear in a **Drafts** group above Active/Inactive, one row each with the name (Resume), a pencil (**Edit**), Play (**Test draft**, tooltip **Test Run**; disabled for event-triggered or non-dispatchable drafts) and a trash (**Delete draft**). Resume and Edit open `/automations/setup?draftId=<id>`; Test dispatches a test run (toast **Test run started.**) on a materialized DRAFT-state automation that the groups do not list; Delete asks **Delete draft?** (`Delete “<name>”? This removes the saved setup draft, but does not delete any final automations.`) with Cancel / **Delete draft** and toasts **Draft deleted.**, also removing the draft's materialized test automation. Expected: the confirmation closes on Escape like the automation delete dialog. Known failure (reproduced 2026-10-13, #18382): Escape leaves it open and it has no dialog role.
+- `F21.saved-drafts`: saved setup drafts (#17242) appear in a **Drafts** group above Active/Inactive, one row each with the name (Resume), a pencil (**Edit**), Play (**Test draft**, tooltip **Test Run**; disabled for event-triggered or non-dispatchable drafts) and a trash (**Delete draft**). Resume and Edit open `/automations/setup?draftId=<id>`; Test dispatches a test run (toast **Test run started.**) on a materialized DRAFT-state automation that the groups do not list; Delete asks **Delete draft?** (`Delete “<name>”? This removes the saved setup draft, but does not delete any final automations.`) with Cancel / **Delete draft** and toasts **Draft deleted.**, also removing the draft's materialized test automation. The confirmation is the named modal dialog **Delete draft?** (the automation delete dialog with draft copy): it opens with focus on Cancel, keeps Tab inside, closes on Escape, X, Cancel or a backdrop click and returns focus to the trash button.
- `F21.phone`: at 390 px the dashboard has no horizontal overflow, tiles wrap 2×2 and the kebab menu stays inside the viewport.
## How to get to it (user POV)
@@ -89,7 +89,7 @@ Preconditions:
- **Empty state (`F21.empty-state`).** With no automations run `control-openhands browser goto /automations`, `control-openhands browser text 'testid=automations-empty'`, `control-openhands browser enabled 'testid=automations-view-toggle'` and `control-openhands browser screenshot --feature F21.empty-state --name empty`. The text starts `No automations configured`, contains `Create an automation`, `Learn more in the documentation` and `Create Automation`, and lists the Recommended automations; `enabled` is `false`; the tiles read `Automations 0 / 0 active` and `Needs attention 0 / No latest-run failures`. `control-openhands browser click 'testid=automations-empty >> testid=automations-create-automation' --expect-url '/automations/setup'` opens the setup page (form plus docked composer, F22) and `control-openhands conversation list` is unchanged (no conversation until the user sends a prompt). The empty-state copy still says "Start a new conversation and tell OpenHands to Create an automation".
- **Add Automation menu (`F21.add-menu`).** Run `control-openhands browser click 'testid=automations-add-automation'` and `control-openhands browser snapshot 'testid=automations-add-automation-menu'`: buttons `Create Automation` and `Import automation`. `control-openhands browser press Escape` (or a click on `role=heading[name="Dashboard"]`) makes `browser count 'testid=automations-add-automation-menu'` `0`. Reopen it and `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`: the agent-assisted setup page opens. Go back to `/automations`, reopen the menu, and `control-openhands browser click 'testid=automations-import-automation'`: the import modal is covered by F22.
- **Recommended rail on the empty dashboard (`F21.empty-rail`).** With no automations, `control-openhands browser click 'testid=automations-empty >> testid=recommended-automation-rail-card-custom-automation' --expect-url 'automations/new'` lands on `/automations/new/custom-automation` with the **Before you start** prerequisites dialog (F22). `control-openhands browser back` returns to `/automations`; there `browser attr 'testid=recommended-automations-rail-fade-left' data-visible` is `false` and `...-fade-right` is `true` (scrolling is F03.recommended-automations-rail).
-- **Saved drafts (`F21.saved-drafts`, `F21.search`).** Arrange one draft through the setup page (F22's **Open and save a draft**: name `QA_F22 Delta Draft`, prompt `Reply pong. Do not run any tools.`, custom cron `0 0 1 1 *`, Save draft), then `control-openhands browser goto /automations`. `control-openhands browser text 'role=main'` lists `Drafts 1` / `QA_F22 Delta Draft` above the groups, and `control-openhands browser testids 'role=main' --filter draft` lists `automation-setup-draft-<draft-id>`, `-resume-`, `-open-`, `-edit-` (`Edit`), `-test-` (`Test draft`) and `-delete-` (`Delete draft`). `control-openhands browser tooltip 'testid=automation-setup-draft-test-<draft-id>'` is `Test Run`; clicking it toasts `Test run started.` and the draft (`api GET /api/automation/v1/drafts/<draft-id>`) gains `materialized_automation_id` and `last_test_run_id`; that automation has `state` `DRAFT`, is not listed in any group, and its detail page shows the Draft badge (F23.header). `testid=automation-setup-draft-resume-<draft-id>` and `testid=automation-setup-draft-edit-<draft-id>` each open `/automations/setup?draftId=<draft-id>` (`--expect-url '/automations/setup'`) with the saved values. Search narrows drafts too: filling `Delta` leaves 0 rows and 1 draft; `zzqa-nomatch` shows the filtered empty state with no draft. Delete: click `testid=automation-setup-draft-delete-<draft-id>`; `control-openhands browser text 'testid=automation-setup-draft-delete-confirm >> xpath=ancestor::div[2]'` reads `Delete draft?` / `Delete “QA_F22 Delta Draft”? This removes the saved setup draft, but does not delete any final automations.` / `Cancel` / `Delete draft`. Its Cancel (`'testid=automation-setup-draft-delete-confirm >> xpath=ancestor::div[2] >> role=button[name="Cancel"]'`) and a backdrop click (`control-openhands browser mouse-click 1400 950`) close it. Expected: Escape closes it too. Known failure (reproduced 2026-10-13, #18382): `browser press Escape` leaves `browser count 'testid=automation-setup-draft-delete-confirm'` at `1`. `control-openhands browser click 'testid=automation-setup-draft-delete-confirm'` toasts `Draft deleted.`; drafts `total` is `0` and the materialized test automation answers `404`.
+- **Saved drafts (`F21.saved-drafts`, `F21.search`).** Arrange one draft through the setup page (F22's **Open and save a draft**: name `QA_F22 Delta Draft`, prompt `Reply pong. Do not run any tools.`, custom cron `0 0 1 1 *`, Save draft), then `control-openhands browser goto /automations`. `control-openhands browser text 'role=main'` lists `Drafts 1` / `QA_F22 Delta Draft` above the groups, and `control-openhands browser testids 'role=main' --filter draft` lists `automation-setup-draft-<draft-id>`, `-resume-`, `-open-`, `-edit-` (`Edit`), `-test-` (`Test draft`) and `-delete-` (`Delete draft`). `control-openhands browser tooltip 'testid=automation-setup-draft-test-<draft-id>'` is `Test Run`; clicking it toasts `Test run started.` and the draft (`api GET /api/automation/v1/drafts/<draft-id>`) gains `materialized_automation_id` and `last_test_run_id`; that automation has `state` `DRAFT`, is not listed in any group, and its detail page shows the Draft badge (F23.header). Once its run is `COMPLETED` (poll `api GET '/api/automation/v1/<materialized-id>/runs?limit=10'`), `control-openhands browser reload` leaves the tiles at `Automations 0 / 0 active`, `Total runs 0` and `Average duration —` (`browser text 'testid=overview-tile-total-runs'`): the test automation is not counted. `testid=automation-setup-draft-resume-<draft-id>` and `testid=automation-setup-draft-edit-<draft-id>` each open `/automations/setup?draftId=<draft-id>` (`--expect-url '/automations/setup'`) with the saved values. Search narrows drafts too: filling `Delta` leaves 0 rows and 1 draft; `zzqa-nomatch` shows the filtered empty state with no draft. Delete: click `testid=automation-setup-draft-delete-<draft-id>`; `control-openhands browser snapshot 'role=dialog[name="Delete draft?"]'` shows `button "Close"`, heading `Delete draft?`, `Delete “QA_F22 Delta Draft”? This removes the saved setup draft, but does not delete any final automations.`, `Cancel` and `Delete draft`, and `control-openhands browser eval "document.activeElement.textContent"` is `Cancel`; `browser press Tab` (repeated) cycles Delete draft, Close and Cancel without leaving the dialog. `control-openhands browser press Escape` makes `browser count 'role=dialog[name="Delete draft?"]'` `0` with the draft kept and focus back on `automation-setup-draft-delete-<draft-id>`; reopening and closing with `'role=dialog[name="Delete draft?"] >> role=button[name="Close"]'`, `... role=button[name="Cancel"]` and a backdrop click (`control-openhands browser mouse-click 1400 950`) each give `0` too. At phone width the confirm button is `insideViewport` (`browser bbox 'testid=automation-setup-draft-delete-confirm'`). `control-openhands browser click 'testid=automation-setup-draft-delete-confirm'` toasts `Draft deleted.`; drafts `total` is `0` and the materialized test automation answers `404`.
- **Arrange two automations (import, F22).** Run `control-openhands browser click 'testid=automations-add-automation'`, `control-openhands browser click 'testid=automations-import-automation'`, `control-openhands browser upload 'testid=automations-import-file' "$OH_VERIFY_RUN/evidence/_fixtures/qa-f21-alpha.automation.json"`, `control-openhands browser click 'testid=import-automation-confirm'`, `control-openhands browser wait 'testid=import-automation-modal' --state detached`. Repeat with `qa-f21-beta.automation.json`. Both arrive disabled, under **Inactive 2**.
- **Grid cards (`F21.card-grid`).** Run `control-openhands browser reload`, `control-openhands browser snapshot 'role=main'` and `control-openhands browser screenshot --feature F21.card-grid --name grid`. Each card shows its name, prompt, pills (`pull_request.opened` + GitHub logo for Beta, `cron` for Alpha), **No activity yet**, and `Runs 0 / Recent success — / Avg. duration —`; both Play buttons are `[disabled]`.
- **Kebab menu (`F21.kebab-menu`).** Run `control-openhands browser click '[data-testid^="automation-card-"] >> has-text=QA_F21 Alpha >> role=button[name="Automation actions"]'` and `control-openhands browser snapshot 'role=list >> has-text=Export'`: buttons `Run now [disabled]`, `View`, `Export`, `Edit`, `Turn on`, `Delete`. `control-openhands browser press Escape`, then `control-openhands browser count 'role=list >> has-text=Export'` is `0`. Reopen it and click `role=heading[name="Dashboard"]`; the count is `0` again.
@@ -109,7 +109,7 @@ Preconditions:
- **Git Sync button (`F21.git-sync-button`).** `control-openhands browser click 'testid=automations-git-sync'`, then `control-openhands browser url`: `/automations/git-sync` (F24).
- **Needs attention, kebab Run now (`F21.overview-tiles`, `F21.run-now`, `F21.filters`).** Run `control-openhands browser goto /automations` (the Git Sync button left `/automations/git-sync`), import `qa-f21-fail.automation.json` as above, turn it on from its kebab, then `control-openhands browser click '[data-testid^="automation-card-"] >> has-text=QA_F21 Fail >> role=button[name="Automation actions"]'`, `control-openhands browser enabled 'role=list >> has-text=Export >> role=button[name="Run now"]'` (`true`), click it, `control-openhands browser wait-text 'Automation dispatched'`. The 1 s timeout run ends `FAILED` after about 3-4 minutes (check with `control-openhands api GET /api/automation/v1/<id>/runs`; the id is in `api GET /api/automation/v1`). Then `control-openhands browser reload`, `control-openhands browser text 'testid=overview-tile-needs-attention'` (`Needs attention 1 / Latest run failed`), and `control-openhands browser screenshot --feature F21.overview-tiles --name needs-attention`. The Fail card shows **Failed** with `Timed out: command timed out or was killed`, a red sparkline bar and `Recent success 0%`; Total runs is `2`. Open Filters and choose status `automations-filter-status-failing` (`testid=automations-filter-status >> testid=dropdown-trigger`, then `testid=automations-filter-status-failing`): only QA_F21 Fail is left. In the still-open popover click `testid=automations-filters-reset` (all three automations return), then close it with `control-openhands browser click 'role=heading[name="Dashboard"]'`: the next bullets need QA_F21 Fail and QA_F21 Beta visible.
- **Run now error (`F21.run-now-error`).** Arrange a stale card: with the dashboard open, run `control-openhands api DELETE /api/automation/v1/<QA_F21 Fail id> --write` and `control-openhands browser toasts --history --clear` (empties the toast history), then click `'[data-testid^="automation-card-"] >> has-text=QA_F21 Fail >> role=button[name="Run now"]'`, then `control-openhands browser toasts --history` and `control-openhands browser screenshot --feature F21.run-now-error --name error-toast`. The history holds exactly one toast, `Automation not found`, and no `Request failed with status code 404`. `browser errors --app-only` then lists the induced 404 on `/dispatch`. Expected: the card then drops out without a reload: `control-openhands browser wait '[data-testid^="automation-card-"] >> has-text=QA_F21 Fail' --state detached --timeout 10000` succeeds. Known failure: the wait times out and the card stays until `control-openhands browser reload` (#18062, see Gotchas).
-- **Delete with confirmation (`F21.delete`).** Open Beta's kebab and click `role=list >> has-text=Export >> role=button[name="Delete"]`. `control-openhands browser snapshot 'role=dialog[name="Delete automation"]'` shows dialog `Delete automation` with `Close`, the heading, `Delete "QA_F21 Beta"? This cannot be undone.`, `Cancel` and `Delete`; focus starts on Cancel (`control-openhands browser eval "document.activeElement.textContent"` is `Cancel`). Take `control-openhands browser screenshot --feature F21.delete --name dialog`. `control-openhands browser click 'role=dialog[name="Delete automation"] >> role=button[name="Cancel"]'` closes it (`control-openhands browser count 'role=dialog[name="Delete automation"]'` is `0`) with the card kept. Reopen and close it with the X (`control-openhands browser click 'role=dialog[name="Delete automation"] >> role=button[name="Close"]'`), then reopen and `control-openhands browser press Escape`: the dialog count is `0` and the card count `1` each time. Reopen, then `control-openhands browser click 'role=dialog[name="Delete automation"] >> role=button[name="Delete"]'`, `control-openhands browser reload` and `control-openhands browser count '[data-testid^="automation-card-"] >> has-text=QA_F21 Beta'`. The count is `0`. Delete the remaining fixtures the same way (rows work too) until `testid=automations-empty` returns.
+- **Delete with confirmation (`F21.delete`).** Open Beta's kebab and click `role=list >> has-text=Export >> role=button[name="Delete"]`. `control-openhands browser snapshot 'role=dialog[name="Delete automation"]'` shows dialog `Delete automation` with `Close`, the heading, `Delete "QA_F21 Beta"? This cannot be undone.`, `Cancel` and `Delete`; focus starts on Cancel (`control-openhands browser eval "document.activeElement.textContent"` is `Cancel`) and `browser press Tab` / `Shift+Tab` cycle Cancel, Delete and Close inside the dialog. Take `control-openhands browser screenshot --feature F21.delete --name dialog`. `control-openhands browser click 'role=dialog[name="Delete automation"] >> role=button[name="Cancel"]'` closes it (`control-openhands browser count 'role=dialog[name="Delete automation"]'` is `0`) with the card kept. Reopen and close it with the X (`control-openhands browser click 'role=dialog[name="Delete automation"] >> role=button[name="Close"]'`), then reopen and `control-openhands browser press Escape`: the dialog count is `0` and the card count `1` each time. Reopen, then `control-openhands browser click 'role=dialog[name="Delete automation"] >> role=button[name="Delete"]'`, `control-openhands browser reload` and `control-openhands browser count '[data-testid^="automation-card-"] >> has-text=QA_F21 Beta'`. The count is `0`. Delete the remaining fixtures the same way (rows work too) until `testid=automations-empty` returns.
- **Load more (`F21.load-more`).** Arrange 51 inert automations (event trigger that never fires): `for i in $(seq -w 1 51); do control-openhands api POST /api/automation/v1/preset/prompt --write --data "{\"name\":\"QA_F21 Page $i\",\"prompt\":\"QA paging fixture; never runs.\",\"trigger\":{\"type\":\"event\",\"source\":\"qa-f21\",\"on\":\"qa.f21.never\"}}"; done`. Then `control-openhands browser goto /automations`, `control-openhands browser wait 'role=button[name="Load more"]'`, row/card count `50`, `control-openhands browser click 'role=button[name="Load more"]'`, `control-openhands browser wait 'role=button[name="Load more"]' --state detached`, count `51`. The preset fixtures arrive enabled (`51 active`) but never fire. Keep them for Phone (it needs at least one automation) and clean up after it.
- **Phone (`F21.phone`, `F21.subpage-nav`).** With at least one automation, run `control-openhands browser viewport phone`, `control-openhands browser bbox 'role=main'` (`insideViewport` `true`, `pageHorizontalOverflow` `false`), `control-openhands browser screenshot --feature F21.phone --name dashboard`, `control-openhands browser click 'testid=automations-navbar-mobile >> testid=automations-navigation-templates'` (URL `/automations/templates`) and back through `automations-navigation-list`. Open the first card's kebab and run `control-openhands browser bbox 'role=list >> has-text=Export'`: the menu is `insideViewport` `true` and sits above the kebab (flipped). Take `control-openhands browser screenshot --feature F21.phone --name kebab`, then `control-openhands browser press Escape` and `control-openhands browser viewport desktop`. Now clean up the paging fixtures with `control-openhands api DELETE /api/automation/v1/<id> --write` for each `QA_F21 Page` id from `control-openhands api GET '/api/automation/v1?limit=100'` (UI deletion is covered above) until `total` is `0`.
- **Errors.** After each group, `control-openhands browser errors --app-only` shows no page errors; the only app errors are the induced 404s for the deleted QA_F21 Fail: `/dispatch` and the stale card's `/runs?limit=20&offset=0`. The backend-down recipes add 502s on `/api/automation/health`, `/sdk-version` and the list while the service is stopped; clear them with `browser errors --clear` after `restart`.
@@ -118,7 +118,6 @@ Preconditions:
- Both sub-page navs are always in the DOM (one hidden by CSS). Unscoped `testid=automations-navigation-templates` is a strict-mode violation; scope it with `testid=automations-navbar-desktop >> ` or `testid=automations-navbar-mobile >> `.
- The pin toggle is invisible until the Automate item is hovered on a fine pointer: `browser hover 'testid=sidebar-automations-link'` before clicking it. The pin is stored per backend and org in `localStorage` (`oh:pinned-home-route:<backend>:<org>`); unpin before leaving the run.
-- Overview tiles count DRAFT-state test automations that the groups hide: after one draft Test the tiles read `Automations 3`, `Total runs 2`, `Average duration 2m` while the page lists two automations with no runs. Known failure (reproduced 2026-10-13, #18381). Run `F21.overview-tiles` with no saved drafts, or delete the draft first (that removes its test automation).
- The Filters popover's status and trigger choices do not apply to the Drafts group: with status Active and trigger Event-driven, a cron draft stays listed and the filtered empty state does not show (search does filter drafts). Intent undocumented in #17242.
- The setup page's success toast (`Automation updated`) covers its toolbar's Test, Save and Close and does not time out while the pointer rests on it; a click there fails with "subtree intercepts pointer events". Move the pointer away (`browser hover` on a form field), wait about 5 s, then click.
- `testid=dropdown-trigger` is reused by the backend selector, the Filters button and every inner dropdown. Always scope it (`testid=automations-filters >> testid=dropdown-trigger >> nth=0` for the Filters button, `testid=automations-filter-status >> testid=dropdown-trigger` for a field).
diff --git a/.agents/skills/verify-openhands/references/feature-map/F22-automation-creation.md b/.agents/skills/verify-openhands/references/feature-map/F22-automation-creation.md
index 692265e723..008d261b9d 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F22-automation-creation.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F22-automation-creation.md
@@ -94,7 +94,7 @@ Preconditions:
- **Open and save a draft (`F22.agent-assisted-setup`, desktop and phone).** This scoped recipe needs a doctored baseline stack with automation draft support (1.19.3 here); its non-dispatch steps need no LLM key or profile. From `/automations` run `control-openhands browser click 'testid=automations-add-automation'`, then `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`. The form and docked composer are visible; `control-openhands conversation list` is unchanged. Run `control-openhands browser fill 'testid=automation-setup-name' 'QA_F22 Delta Draft'`, `control-openhands browser fill 'testid=automation-setup-prompt' 'Reply pong. Do not run any tools.'`, `control-openhands browser click 'testid=automation-setup-frequency-custom'`, `control-openhands browser fill 'testid=automation-setup-custom-schedule' '0 0 1 1 *'`, then `control-openhands browser click 'testid=automation-setup-save-draft'`. `control-openhands browser wait 'testid=automation-setup-draft-details' --timeout 10000` and `browser snapshot 'testid=automation-setup-draft-details'` show **Draft saved** and **Ready to test**. `control-openhands api GET /api/automation/v1/drafts` supplies `<draft-id>` and confirms the name, prompt and yearly cron; the URL carries `draftId`. After `browser reload`, read each field with `browser value 'testid=automation-setup-name'`, `browser value 'testid=automation-setup-prompt'` and `browser value 'testid=automation-setup-custom-schedule'`: the values match. Screenshot with `--feature F22.agent-assisted-setup --name saved-draft`. Saving a prompt draft does not prove model execution; use the configured flash profile and the **Genuine prompt/plugin draft Test** checks below for that path.
- **Credential-free script test (`F22.agent-assisted-setup`).** Run `control-openhands browser click 'testid=automation-setup-kind-custom'`, then inspect `browser value 'testid=automation-setup-custom-code'`: the generated Python prints its prompt and calls the real automation callback, without an agent/LLM. Keep its code, setup and entrypoint unchanged; click Save draft. Run `browser record start --feature F22.agent-assisted-setup --name custom-script-test --max-seconds 180`, then `browser click 'testid=automation-setup-draft-test'` and `browser wait 'testid=automation-setup-draft-runs-page' --timeout 10000`. Its snapshot initially shows Pending. Immediately run `api GET /api/automation/v1/drafts/<draft-id>` for `materialized_automation_id` and `last_test_run_id`, then `api GET '/api/automation/v1/<materialized-id>/runs?limit=10'`. Repeat the quoted runs GET after short `sleep 5` intervals until that run has terminal backend status (bounded to 90 seconds; no single wait over 45 seconds). Once it is COMPLETED, `browser wait 'testid=automation-setup-draft-run >> text=Successful' --timeout 45000` tests automatic UI completion. Expected: Successful without a reload. Record the actual row and a second backend receipt before any recovery. As a separate readback, run `browser reload`, click `testid=automation-setup-view-test-runs`, screenshot Successful, then `browser click 'testid=automation-setup-draft-runs-back'`: the draft details read **Latest test passed**. End with `browser record stop`. A saved draft, Pending row or successful reload does not prove automatic completion updates.
- **Resume at phone width (`F22.agent-assisted-setup`).** Run `browser viewport phone`, `browser goto /automations`, then click `testid=automation-setup-draft-resume-<draft-id>` with `--expect-url '/automations/setup'`. Read Name, rename to `QA_F22 Phone Draft`, Save draft and reload: `browser value 'testid=automation-setup-name'` and `api GET /api/automation/v1/drafts/<draft-id>` both have the new name. `browser bbox 'testid=automation-setup-form'` has `pageHorizontalOverflow: false`; this long form scrolls vertically and need not fit wholly inside the viewport. Screenshot with `--feature F22.agent-assisted-setup --name resumed-phone`. Repeat initial open/save/reload on a fresh phone draft when checking that entry point; desktop creation is not phone creation. Restore desktop afterward.
-- **Agent fills a prompt form (`F22.agent-assisted-setup`, desktop).** Start a fresh form with `control-openhands browser goto /automations`, `control-openhands browser click 'testid=automations-add-automation'` and `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`. Run `control-openhands conversation start --stay --prompt 'Call automation_form_update exactly once with fields {"name":"QA_F22 DeepSeek Desktop","kind":"prompt","prompt":"Reply pong. Do not run any tools.","frequency":"custom","customSchedule":"0 0 1 1 *","timezone":"UTC","model":"deepseek-flash"}. Do not call any other tools, run terminal commands, inspect files, or read environment variables. Briefly confirm the filled fields and stop.'`. `--stay` sends through the docked composer; omitting it would navigate away to New Chat. The returned `<setup-conversation-id>` opens `/conversations/<id>` with the form beside it. Run `control-openhands conversation wait <setup-conversation-id> --until finished,idle --timeout 60`, then `control-openhands conversation events <setup-conversation-id> --kinds ActionEvent --full`: a genuine `automation_form_update` supplied the exact fields. Read `control-openhands browser value 'testid=automation-setup-name'`, `control-openhands browser value 'testid=automation-setup-prompt'`, `control-openhands browser value 'testid=automation-setup-custom-schedule'` and `control-openhands browser value 'testid=automation-setup-model'`; they match the requested values. `control-openhands browser snapshot 'testid=automation-setup-form'` also shows **Filled by OpenHands** markers. Capture `control-openhands browser screenshot --feature F22.agent-assisted-setup --name desktop-genuine-tool-filled`; the UI field readbacks and real action establish the update, not the agent's confirmation alone.
+- **Agent fills a prompt form (`F22.agent-assisted-setup`, desktop).** Start a fresh form with `control-openhands browser goto /automations`, `control-openhands browser click 'testid=automations-add-automation'` and `control-openhands browser click 'testid=automations-add-automation-create' --expect-url '/automations/setup'`. Run `control-openhands conversation start --stay --prompt 'Call automation_form_update exactly once with fields {"name":"QA_F22 DeepSeek Desktop","kind":"prompt","prompt":"Reply pong. Do not run any tools.","frequency":"custom","customSchedule":"0 0 1 1 *","timezone":"UTC","model":"deepseek-flash"}. Do not call any other tools, run terminal commands, inspect files, or read environment variables. Briefly confirm the filled fields and stop.'`. `--stay` sends through the docked composer; omitting it would navigate away to New Chat. The returned `<setup-conversation-id>` opens `/conversations/<id>` with the form beside it. Run `control-openhands conversation wait <setup-conversation-id> --until finished,idle --timeout 60`, then `control-openhands conversation events <setup-conversation-id> --kinds ActionEvent --full`: a genuine `automation_form_update` supplied the exact fields. Read `control-openhands browser value 'testid=automation-setup-name'`, `control-openhands browser value 'testid=automation-setup-prompt'`, `control-openhands browser value 'testid=automation-setup-custom-schedule'` and `control-openhands browser text 'testid=automation-setup-model'`; they match the requested values. `control-openhands browser snapshot 'testid=automation-setup-form'` also shows **Filled by OpenHands** markers. Capture `control-openhands browser screenshot --feature F22.agent-assisted-setup --name desktop-genuine-tool-filled`; the UI field readbacks and real action establish the update, not the agent's confirmation alone. The sent prompt is not kept as a draft: `control-openhands browser eval "sessionStorage.getItem('oh:home-prompt-draft')"` is `null` (with no earlier Home draft), and the New Chat composer (`browser eval "document.querySelector('[data-testid=chat-input]').innerText"` after clicking `text=New Chat`) and a later `/automations/setup` docked composer are empty; return to `/conversations/<setup-conversation-id>` for the next bullet.
- **Desktop Hide/Show and phone views (`F22.agent-assisted-setup`).** On that associated conversation, `control-openhands browser attr 'testid=automation-setup-agent-toggle' aria-label` reads `Hide Agent`. Click the toggle: the label becomes `Show Agent`, and `control-openhands browser bbox 'testid=conversation-chat-panel'` has width `0`; form values remain. Click again: `Hide Agent` and a visible conversation panel return. Save with `control-openhands browser click 'testid=automation-setup-save-draft'`, `control-openhands browser wait 'testid=automation-setup-draft-details' --timeout 10000` and `control-openhands api GET /api/automation/v1/drafts` to obtain `<assisted-draft-id>` and the stored form. Run `control-openhands browser viewport phone`, `control-openhands browser click 'testid=automation-setup-mobile-view-form'`, read Name, then `control-openhands browser click 'testid=automation-setup-mobile-view-conversation'`; the selected view fits without horizontal overflow. From `control-openhands browser goto /automations`, click `testid=automation-setup-draft-resume-<assisted-draft-id>` with `--expect-url '/automations/setup'`: dashboard Resume opens `/automations/setup?draftId=<id>`, not the original conversation URL. Reload and read Name/prompt/schedule again. Separately `control-openhands browser goto /conversations/<setup-conversation-id>` and `control-openhands browser click 'testid=automation-setup-mobile-view-form'` restore the same draft inside its associated conversation. Capture each entry separately. `control-openhands browser click 'testid=automation-setup-create' --expect-url '/automations/[0-9a-f-]{8,}'` creates the final prompt automation; `control-openhands api GET /api/automation/v1/<id>` shows the requested fields and `enabled: false`.
- **Fresh phone prompt and genuine Run now (`F22.agent-assisted-setup`).** Stay at phone width, open a new form through Dashboard Add → Create, and send the exact one-tool prompt above with name `QA_F22 DeepSeek Phone`. Read the genuine action and fields, click `testid=automation-setup-mobile-view-form`, Save draft, reload and read back the saved values before clicking Create. On the resulting detail page run `control-openhands browser click 'role=switch[name="Turn on"]'`, `control-openhands browser click 'role=button[name="Run now"]'`, then `control-openhands browser wait 'testid=automation-activity-log >> testid=run-status-icon-completed' --timeout 60000` without reloading. `control-openhands api GET /api/automation/v1/<id>/runs` independently reports COMPLETED with a conversation id; `control-openhands conversation events <run-conversation-id> --kinds ActionEvent --full` contains a real `finish`, and MessageEvent output is `pong`. The ordinary detail row becomes Successful automatically. Desktop assisted-prompt creation and fresh phone prompt execution are separate scopes. Return to desktop before the next bullet.
- **Agent-created plugin (`F22.agent-assisted-setup`, desktop and fresh phone).** Arrange a public plugin fixture with `mkdir -p "$OH_VERIFY_RUN/workspace"` and `cp -R node_modules/@openhands/extensions/plugins/magic-test "$OH_VERIFY_RUN/workspace/qa-f22-real-plugin"`; `<plugin-path>` below is that absolute run-owned path. Open another fresh form and send the one-tool docked prompt with fields `{"name":"QA_F22 DeepSeek Real Plugin","kind":"plugin","pluginSource":"<plugin-path>","prompt":"Reply pong. Do not run any tools.","frequency":"custom","customSchedule":"0 0 1 1 *","timezone":"UTC","model":"deepseek-flash"}` (replace the path before sending). Read the real `automation_form_update`, the form and its saved values. Create, Turn on and Run now as above; the detail row becomes Successful without reload and its actual conversation finishes `pong`. `control-openhands conversation events <run-conversation-id> --grep magic-word --from-start` confirms that the public plugin's skill registered. This checks plugin loading/execution; `pong` does not exercise that skill's trigger, hooks, MCP or scripts. Repeat from a fresh phone form with name `QA_F22 DeepSeek Plugin Phone Draft`, using the Form view and the next draft-Test bullet before final Create/Turn on/Run now; desktop creation alone does not prove that phone path.
@@ -113,9 +113,9 @@ Preconditions:
- Expected: a completed draft test updates the open Test runs row to Successful automatically. Known failure (reproduced again 2026-10-13 on a desktop prompt draft: run COMPLETED at 08:13:39, row still Pending after a further 45 s wait; verified 2026-10-10, fresh Local stacks, Agent Server 1.54.0/automation 1.19.3): callback-only script tests and genuine DeepSeek prompt/plugin tests report COMPLETED while the open row stays Pending. The desktop prompt stayed Pending through a 45-second wait; the phone plugin capture was more than 201 seconds after completion. Reload and View test runs recover Successful ([#18263](https://github.com/OpenHands/OpenHands/issues/18263)). Ordinary final-automation Run now polling passes; reload recovery does not waive the draft-Test failure.
- The Save toast can cover the setup toolbar Test button; use the lower `automation-setup-draft-test` and verify the actual run outcome. The toast does not time out while the pointer rests on it, so move the pointer away (`browser hover` on a form field) before clicking the toolbar again.
-- A prompt sent from the setup page's docked composer is kept as the home prompt draft (`sessionStorage` `oh:home-prompt-draft`): afterwards the Home composer and every later `/automations/setup` composer (new or edit) open pre-filled with that already-sent text. Expected: a delivered prompt clears the draft. Reproduced twice 2026-10-13 (CLI `conversation start --stay` and a pure UI fill + `submit-button`); Known failure #18379. Clear the composer (`browser fill 'testid=automation-setup-docked-composer >> testid=chat-input' ''`) before reading it in later bullets.
+- The setup page's docked composer keeps no draft: it opens empty even when the Home composer has an unsent draft (`sessionStorage` `oh:home-prompt-draft`), a prompt sent from it never becomes the Home draft, and leaving it keeps an existing Home draft. Unsent text typed there is lost when the user leaves the page.
- In the desktop split view (form beside the conversation) the Frequency strip scrolls horizontally inside the narrower form and the selected **Custom** button sits off-screen; scroll it into view before reading it.
-- Dashboard draft Delete uses an inline confirmation, not a `role=dialog`. Deleting a draft also removes its materialized test automation.
+- Dashboard draft Delete asks in the named dialog `role=dialog[name="Delete draft?"]` (F21.saved-drafts). Deleting a draft also removes its materialized test automation.
- A card click and a deep link are different paths: from the grid, GitHub/Slack-only responders (`qa-changes`, `github-agents-md-maintainer`, `slack-channel-monitor`, …) first show the responder choice and then the GitHub/Slack install dialog, so their setup form is reachable locally only by URL (`/automations/new/<id>`).
- A card's `N MCPs to connect before launch` pill counts optional integrations too: Custom automation shows `1 MCPs to connect before launch` for GitHub, yet opens its setup at once because GitHub is optional there (#17959).
- The featured-section copy says picking a card opens "a pre-filled conversation"; most featured cards open a setup form instead.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F23-automation-detail.md b/.agents/skills/verify-openhands/references/feature-map/F23-automation-detail.md
index f3ca82048d..5d272fff44 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F23-automation-detail.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F23-automation-detail.md
@@ -49,15 +49,15 @@ Source: `src/routes/automation-detail.tsx`, `src/components/features/automations
- `F23.delete`: kebab **Delete** asks `Delete "<name>"? This cannot be undone.`; Cancel keeps it, Delete removes it and returns to `/automations`.
- `F23.edit-open`: kebab **Edit** opens the setup page `/automations/setup?automationId=<id>` (#17688; the **Edit automation** modal is gone) prefilled with Name, Prompt, agent profile and LLM profile pickers, the trigger (Schedule with Frequency, or Event), timezone and Timeout seconds, with **Test**, **Save** and **Close** in the toolbar and a docked composer; no conversation exists until a prompt is sent.
- `F23.edit-close`: **Close** leaves without saving and lands on `/automations` (the dashboard, not the detail page); Escape does nothing on the page.
-- `F23.edit-validation`: an empty prompt is refused with the toast **Add a prompt before continuing.** (no request); a bad cron, a cron that never runs or a timeout above 1800 should be refused with a readable message. Known failure (reproduced 2026-10-13, #18380; regression from #17688): Save sends the PATCH and shows only the toast `Request failed with status code 422`; the API detail (`Value error, Invalid cron expression: …`) is dropped. BASE's modal refused the cron inline (`Enter a valid cron expression`).
+- `F23.edit-validation`: an empty prompt is refused with the toast **Add a prompt before continuing.** (no request); a bad cron, a cron that never runs or a timeout above 1800 is sent to the API, refused with `422`, and its message is the single error toast (`Value error, Invalid cron expression: not a cron`, `Value error, Cron expression cannot produce any future fire times: 0 0 31 2 *`, `Value error, timeout must not exceed 1800 seconds (30 minutes)`); nothing is saved.
- `F23.edit-save`: Save sends one PATCH (also when nothing changed), shows **Saving…** then **Saved just now**, the toast **Automation updated**, and the detail page and API reflect it; an Active profile model stays `null`.
- `F23.edit-test`: on the edit page **Test** dispatches the automation as last saved (`POST /api/automation/v1/<id>/dispatch`, no PATCH first) and shows `Test run started.` on the form; unsaved edits are not part of that run. Whether it should test the current form is a product decision (Gotchas).
- `F23.edit-dirty-state`: the edit page shows no save state for unsaved edits (no **Unsaved changes**, which create mode shows, F28); the label appears only while saving (**Saving…**) and after (**Saved just now** or **Save failed**). **Close** discards the edits.
- `F23.edit-schedule`: Frequency offers Once, Hourly, Daily, Weekdays (Mon–Fri), Weekly and Custom; Weekly adds **Day of week** and **At**; they are saved as a cron expression and read back after a reload.
-- `F23.edit-time-cleared`: clearing **At** on a preset schedule and saving should be refused with a message instead of being dropped or replaced. Known failure (reproduced 2026-10-13, #17948 item, new symptom since #17688): Save silently stores midnight (`0 0 * * 3`) and toasts **Automation updated**; Save draft in the create flow stores `0 0 * * *` the same way. Open PR #18260 changes only `edit-automation-modal.tsx`, which nothing imports on TARGET, so it does not fix this.
+- `F23.edit-time-cleared`: clearing **At** on a preset schedule and saving should be refused with a message instead of being dropped or replaced. Known failure (reproduced 2026-10-13, #17948 item, new symptom since #17688): Save silently stores midnight (`0 0 * * 3`) and toasts **Automation updated**; Save draft in the create flow stores `0 0 * * *` the same way. Open PR #18260 (head `0f812ac68902d2c71896568311bffb8aa6f216a6`, checked 2026-10-15) targets the live setup panel: it rejects an invalid or cleared time before Save and Save draft, with an edit-mode regression test. It is not merged; the known failure above remains the observed behavior on this map’s baseline.
- `F23.edit-cron`: a custom cron schedule shows Frequency **Custom** with an editable **Custom** cron field and a timezone select.
- `F23.edit-event-readonly`: for an event automation the edit page shows Event selected with Event source, event key and filter controls (editable now, no longer read-only text) and no schedule fields.
-- `F23.edit-save-error`: a failed save shows exactly one error toast with the API message and keeps the edits on the page. Known failure (reproduced 2026-10-13, regression from #17688, #18380): the toast reads `Request failed with status code 404` instead of `Automation not found`.
+- `F23.edit-save-error`: a failed save shows exactly one error toast with the API message (`Automation not found` for an automation deleted meanwhile), the save state reads **Save failed** and the edits stay on the page; the toolbar **Test** on that page also toasts `Automation not found`.
- `F23.phone`: at 390×844 the detail page has no horizontal overflow, and the Edit page fits with its title, Test, Save and Close reachable.
## How to get to it (user POV)
@@ -95,17 +95,17 @@ Preconditions:
- **Export and Tarball (`F23.export`, `F23.tarball`).** `control-openhands browser click 'role=button[name="Automation actions"]'`, `control-openhands browser click 'role=list >> has-text=Tarball >> role=button[name="Export"]'`, `control-openhands browser downloads --last 1 --inspect`: `qa-f23-pong.automation.json` with `"version": 1`, `"kind": "automation"` and `spec` (name, trigger, `enabled`, prompt, `model` `null`, `timeout` `600`, `timezone`). Reopen the kebab and click `role=list >> has-text=Tarball >> role=button[name="Tarball"]`; `control-openhands browser downloads --last 1` lists `QA_F23_Pong.tar` (about 9 KB) and no toast appears. The file is gzip data despite the name: `tar -tzf <path>` lists `main.py`, `finish_tool_hook.py`, `prompt.txt`, `setup.sh`.
- **Turn off and the paused banner (`F23.disabled-reason`, `F23.toggle`).** While Pong is active `control-openhands browser count 'testid=automation-disabled-reason-banner'` is `0`. Open the kebab, `control-openhands browser click 'role=list >> has-text=Tarball >> role=button[name="Turn off"]'`, `control-openhands browser count 'testid=confirmation-modal'` (`0`), `control-openhands browser reload`, `control-openhands browser text 'testid=automation-disabled-reason-banner'`: `Why this is paused` / `Turned off manually` / `Paused Just now` (`automation-disabled-reason-text`, `automation-disabled-reason-timestamp`). The badge reads `Inactive` and Run now is disabled. Screenshot with `--feature F23.disabled-reason --name manual`.
- **Edit fields and closing (`F23.edit-open`, `F23.edit-close`).** On Pong run `control-openhands browser click 'role=button[name="Automation actions"]'` and `control-openhands browser click 'role=list >> has-text=Tarball >> role=button[name="Edit"]' --expect-url '/automations/setup'`: the URL is `/automations/setup?automationId=<id>`, `browser count 'testid=edit-automation-save'` is `0` (no modal), and `browser value` reads `automation-setup-name` (`QA_F23 Pong`), `automation-setup-prompt`, the stored schedule (a custom cron shows Frequency Custom with `automation-setup-custom-schedule`) and `automation-setup-timeout` (`600`); `browser text 'testid=automation-setup-agent-profile'` is `default` and `browser text 'testid=automation-setup-model'` names the active profile (the stored `model` stays `null`). `browser testids 'testid=automation-setup-toolbar'` lists `automation-setup-test`, `automation-setup-save` and `automation-setup-close`; `control-openhands conversation list` is unchanged. `control-openhands browser press Escape` leaves the URL unchanged; `control-openhands browser click 'testid=automation-setup-close' --expect-url '/automations$'` lands on the dashboard without a PATCH (the dashboard's own GETs follow). A cold `control-openhands browser goto '/automations/setup?automationId=<id>'` opens the same form.
-- **Validation (`F23.edit-validation`).** On the edit page run `control-openhands browser fill 'testid=automation-setup-prompt' ''`, `control-openhands browser network --clear` and `control-openhands browser click 'testid=automation-setup-save'`: `control-openhands browser toasts --history` includes the toast `Add a prompt before continuing.` and `browser network --filter 'automation/v1'` lists no request. Refill the prompt. On a Custom-cron automation (QA_F23 Fail) fill `testid=automation-setup-custom-schedule` with `not a cron` (then `0 0 31 2 *`, then a valid cron with `testid=automation-setup-timeout` `1801`), click Save and read `control-openhands browser toasts --history` after `browser toasts --history --clear`. Expected: each is refused with a readable message (the API answers `Value error, Invalid cron expression: not a cron`). Known failure (reproduced 2026-10-13, #18380; regression from #17688): each sends a `PATCH` that returns `422` and the only feedback is the toast `Request failed with status code 422`; nothing is saved. Before each new click move the pointer off the toast (`browser hover 'testid=automation-setup-name'`, about 5 s; see Gotchas).
+- **Validation (`F23.edit-validation`).** On the edit page run `control-openhands browser fill 'testid=automation-setup-prompt' ''`, `control-openhands browser network --clear` and `control-openhands browser click 'testid=automation-setup-save'`: `control-openhands browser toasts --history` includes the toast `Add a prompt before continuing.` and `browser network --filter 'automation/v1'` lists no request. Refill the prompt. On a Custom-cron automation (QA_F23 Fail) fill `testid=automation-setup-custom-schedule` with `not a cron` (then `0 0 31 2 *`, then a valid cron with `testid=automation-setup-timeout` `1801`), click Save and read `control-openhands browser toasts --history` after `browser toasts --history --clear`. Each sends a `PATCH` that returns `422` (`browser network --filter 'automation/v1'`), and the history holds exactly one toast with the API message: `Value error, Invalid cron expression: not a cron`, `Value error, Cron expression cannot produce any future fire times: 0 0 31 2 *` and `Value error, timeout must not exceed 1800 seconds (30 minutes)`; `api GET /api/automation/v1/<Fail id>` keeps `0 0 1 1 *` and timeout `600`. Before each new click move the pointer off the toast (`browser hover 'testid=automation-setup-name'`, about 5 s; see Gotchas).
- **No-change Save (`F23.edit-save`).** Open Edit, `control-openhands browser network --clear`, click `testid=automation-setup-save`, then `control-openhands browser network --filter 'automation/v1'`: one `PATCH` with the unchanged values and the toast `Automation updated` (the old modal sent nothing).
- **Weekly schedule, model and timeout (`F23.edit-schedule`, `F23.edit-save`).** On the edit page run `control-openhands browser click 'testid=automation-setup-frequency-weekly'`, `control-openhands browser select 'testid=automation-setup-weekday' 3` (Wednesday; options `0:Sunday` … `6:Saturday`), `control-openhands browser fill 'testid=automation-setup-time' 10:30`, `control-openhands browser click 'testid=automation-setup-model'`, `control-openhands browser click 'testid=automation-setup-model-option-deepseek-pro'`, `control-openhands browser fill 'testid=automation-setup-timeout' 300` and `control-openhands browser screenshot --feature F23.edit-schedule --name weekly-form`. Then `control-openhands browser network --clear`, `control-openhands browser click 'testid=automation-setup-save' --observe 'testid=automation-setup-save-state' --observe-ms 2000` (`Saving…` → `Saved just now`) and `control-openhands browser toasts --history` (`Automation updated`). `browser network --bodies --filter 'automation/v1'` shows one `PATCH /api/automation/v1/<id>` with `schedule` `30 10 * * 3`, `model` `deepseek-pro` and `timeout` `300`. After `control-openhands browser reload` the weekday is `3`, At `10:30`, the model `deepseek-pro` and Timeout `300`; the detail page Configuration reads `Schedule` `30 10 * * 3` and `LLM profile` `deepseek-pro`.
-- **Cleared time (`F23.edit-time-cleared`).** On the weekly edit page `control-openhands browser fill 'testid=automation-setup-time' ''`, `control-openhands browser network --clear`, click Save. Expected: an inline message and nothing saved. Known failure (reproduced 2026-10-13, #17948 item, new symptom since #17688): `browser network --bodies --filter 'automation/v1'` shows a `PATCH` with `schedule` `0 0 * * 3`, the toast reads `Automation updated` and the automation now runs at midnight. Save draft in the create flow (`/automations/setup`) stores `0 0 * * *` the same way; open PR #18260 touches only the unused `edit-automation-modal.tsx`.
+- **Cleared time (`F23.edit-time-cleared`).** On the weekly edit page `control-openhands browser fill 'testid=automation-setup-time' ''`, `control-openhands browser network --clear`, click Save. Expected: an inline message and nothing saved. Known failure (reproduced 2026-10-13, #17948 item, new symptom since #17688): `browser network --bodies --filter 'automation/v1'` shows a `PATCH` with `schedule` `0 0 * * 3`, the toast reads `Automation updated` and the automation now runs at midnight. Save draft in the create flow (`/automations/setup`) stores `0 0 * * *` the same way; open PR #18260 targets both paths in `automation-setup-panel.tsx` at head `0f812ac68902d2c71896568311bffb8aa6f216a6` (checked 2026-10-15), but is not merged.
- **Custom cron (`F23.edit-cron`).** Open QA_F23 Fail and Edit: Frequency **Custom** is selected and `control-openhands browser value 'testid=automation-setup-custom-schedule'` is `0 0 1 1 *` beside the timezone select (`UTC`). Fill `0 0 2 1 *` and `testid=automation-setup-timeout` with `1` (this makes the next run fail), Save and `browser toasts --history` (`Automation updated`); after a reload of the detail page the Configuration shows `0 0 2 1 *`. Invalid crons: see Validation.
- **Failed run, its row and Debug with OpenHands (`F23.run-status-polling`, `F23.run-no-conversation`, `F23.run-logs-modal`, `F23.debug-with-openhands`).** On QA_F23 Fail (timeout `1` from the previous bullet) click `role=button[name="Run now"]` and `control-openhands browser wait-text 'Automation dispatched'`. The run stays Running for about 3-4 minutes and ends `FAILED` (poll `control-openhands api GET /api/automation/v1/<id>/runs`; drive the event automation bullets meanwhile). Then `control-openhands browser goto /automations`, open QA_F23 Fail, `control-openhands browser wait 'testid=automation-activity-log >> testid=run-status-icon-failed'` and `control-openhands browser text 'testid=automation-activity-log'`: the newest row reads `<time>`, `Timed out: command timed out or was killed`, `Failed`. Expected beside the time: `No Conversation` (this is a prompt automation); today it reads `No conversation — this run executed a script. Use View logs for its output.` Run `control-openhands browser click 'testid=automation-activity-log >> title=View logs >> nth=0'` and `control-openhands browser snapshot 'role=dialog'`: `Run` `FAILED`, `Task` `No task outcome reported.`, `System` `Error: Timed out: command timed out or was killed` and `Status detail: ...`, the Output tab with the `[setup] Fetching SDK version ...` lines, and button `Debug with OpenHands`. Screenshot with `--feature F23.debug-with-openhands --name failed-logs`. `control-openhands browser click 'testid=debug-automation-button' --expect-url '/conversations/'` opens a new conversation; `control-openhands conversation events <conversation id from browser url> --kinds MessageEvent` starts `The scheduled automation "QA_F23 Fail" failed during a run. Please investigate the error and fix the root cause.`, and `control-openhands browser eval "document.body.innerText.includes('Timed out: command timed out or was killed')"` is `true` (the page also shows the run id). The agent then works on that prompt with the active profile.
- **Task outcome badge (`F23.run-task-outcome`).** Arrange (API, arrange only) `control-openhands api POST /api/automation/v1/preset/prompt --write --data '{"name":"QA_F23 Blocked","prompt":"Do not run any tools. Call the finish tool right away with status \"blocked\" and outcome_summary \"QA blocked on purpose\".","trigger":{"type":"cron","schedule":"0 0 1 1 *","timezone":"UTC"},"enabled":true}'` (a good thing to do while the Fail run is still running). Open QA_F23 Blocked, click `role=button[name="Run now"]`, and poll `control-openhands browser text 'testid=automation-activity-log'` until it leaves Running (about 35 s): `<time>`, `QA blocked on purpose`, a cost such as `$0.0016`, `Blocked`; `control-openhands browser testids 'testid=automation-activity-log'` lists `run-status-icon-warning` (not `-completed`) although `api GET /api/automation/v1/<id>/runs` has `status` `COMPLETED` and `run_metadata.finish_tool_response.status` `blocked`. `control-openhands browser click 'testid=automation-activity-log >> title=View logs'` and `browser snapshot 'role=dialog'`: `Run` `COMPLETED`, `Task` `Blocked` / `QA blocked on purpose` / `Task metadata` `{ "message": ... }`. Screenshot with `--feature F23.run-task-outcome --name blocked`, Escape. `partial_success` (**Partial**, `run-status-icon-warning`) and `unknown` (**Needs review**, `run-status-icon-needs-review`) follow the same path with the same prompt shape. The normal automation 1.19.0 prompt/plugin presets configure SDK 1.53.0 FinishTool with the required `TaskOutcome.status`, limited to `success`, `partial_success`, `blocked`, `failed` and `unknown`; SDK custom response schemas are a separate producer boundary. Its normal prompt path therefore does not establish the new fallback for a finish response without a string status. That positive boundary needs a genuine completed run from a legacy or custom-schema producer whose real finish arguments have that shape; a run with no finish metadata checks only the ordinary Completed fallback. Arbitrary unrecognised strings and summary retention on a failed run need the corresponding genuine producer too. Until available, record those checks as blocked and do not treat injected run metadata or mocked LLM output as live proof (#17682).
- **Event automation (`F23.prompt-collapse`, `F23.configuration-event`, `F23.plugins-repo`, `F23.export`).** Open QA_F23 Event. `control-openhands browser bbox 'testid=automation-prompt-content'` is 240 px high, `testid=automation-prompt-fade` is present and `control-openhands browser text 'testid=automation-prompt-toggle'` is `View More`; screenshot with `--feature F23.prompt-collapse --name collapsed`. Click the toggle: height 360, no fade, text `View Less`; click again to collapse. `control-openhands browser text 'role=heading[name="Configuration"] >> xpath=ancestor::div[2]'` reads `Repositories qa-example/qa-repo main`, `Trigger Event`, `Event Source github`, `Event Type pull_request.opened`, `Filter pull_request.draft == `false` && pull_request.base.ref == re…`, `Show more`, `LLM profile Active profile`. `control-openhands browser click 'role=button[name="Show more"]'` shows the full filter and `control-openhands browser attr 'role=button[name="Show less"]' aria-expanded` is `true`. `control-openhands browser text 'testid=automation-repository'` is `qa-example/qa-repo` / `main` (the fixture's `https://github.com/qa-example/qa-repo` as `owner/repo`, with its branch badge) and `control-openhands browser text 'role=heading[name="Plugins"] >> xpath=ancestor::div[2]'` is `Plugins` / `github:qa-example/qa-plugin`; screenshot with `--feature F23.plugins-repo --name event-repo-plugins`. Export keeps them: `control-openhands browser click 'role=button[name="Automation actions"]'`, `control-openhands browser click 'role=list >> has-text=Tarball >> role=button[name="Export"]'` and `control-openhands browser downloads --last 1 --contains '"branch": "main"'` (`qa-f23-event.automation.json`, `contains` `true`); `--contains '"repository": "https://github.com/qa-example/qa-repo"'` and `--contains 'github:qa-example/qa-plugin'` are `true` too (the `--inspect` head ends inside the long prompt).
- **Event trigger in Edit (`F23.edit-event-readonly`).** On QA_F23 Event open Edit: Event is selected, `browser testids 'testid=automation-setup-form' --filter event` lists `automation-setup-event-source` (GitHub), `automation-setup-event-source-toggle`, `automation-setup-event-public-url-notice`, `automation-setup-event-key`, `automation-setup-event-filter` and `automation-setup-event-filter-remove` (the fixture has a filter), and there are no Frequency controls for a schedule. The trigger is editable here (it was read-only text in the old modal). Screenshot with `--feature F23.edit-event-readonly --name event`, then Close.
- **Phone (`F23.phone`).** Run `control-openhands browser viewport phone`, open QA_F23 Pong, `control-openhands browser bbox 'role=heading[name="QA_F23 Pong"] >> xpath=ancestor::div[3]'` (`insideViewport` `true`, `pageHorizontalOverflow` `false`) and `control-openhands browser screenshot --feature F23.phone --name detail` (the name and Run now wrap). Open Edit (kebab, `--expect-url '/automations/setup'`), then `control-openhands browser bbox 'testid=automation-setup-save'` and `... 'testid=automation-setup-close'` (both `insideViewport` `true`, Save at `y` about 65) and `control-openhands browser bbox 'testid=automation-setup-form'` (`pageHorizontalOverflow` `false`); screenshot with `--feature F23.phone --name edit-setup-page`. Close and `control-openhands browser viewport desktop`.
-- **Run now and save errors (`F23.run-now-error`, `F23.edit-save-error`).** Arrange a stale page: open QA_F23 Fail, then `control-openhands api DELETE /api/automation/v1/<Fail id> --write`. Run `control-openhands browser toasts --history --clear` (empties the toast history), click `role=button[name="Run now"]`, then `control-openhands browser toasts --history` and `control-openhands browser screenshot --feature F23.run-now-error --name toasts`: exactly one toast, `Automation not found`, and no `Request failed with status code 404`. Expected: the page then shows **Automation not found**: `control-openhands browser wait 'role=heading[name="QA_F23 Fail"]' --state detached --timeout 15000` succeeds (the not-found state takes about 7 s, as in States) and `control-openhands browser snapshot 'role=paragraph >> text=Automation not found >> xpath=..'` lists link `Back to Automations` (do not use `wait-text 'Automation not found'` here: the toast has the same text). Once the heading is gone, run `control-openhands browser toasts --history` again: it still holds only `Automation not found` (a late second toast, `Request failed with status code 404`, would come from the page's own refetch of the deleted automation). Known failure: the wait times out and the page keeps heading `QA_F23 Fail`, `Active` and an enabled **Run now** (`control-openhands browser enabled 'role=button[name="Run now"]'` is `true`) until a reload (#18062). The save error does not reuse that page (it is gone once #18062 is fixed): arrange a second fixture, `control-openhands api POST /api/automation/v1/preset/prompt --write --data '{"name":"QA_F23 Stale","prompt":"Reply with the single word: pong.","trigger":{"type":"cron","schedule":"0 0 1 1 *","timezone":"UTC"},"enabled":false}'` (`body.id` is `<stale-id>`), open QA_F23 Stale, its kebab and `role=list >> has-text=Tarball >> role=button[name="Edit"]` (`--expect-url '/automations/setup'`). With the edit page open, `control-openhands api DELETE /api/automation/v1/<stale-id> --write` and `control-openhands browser toasts --history --clear`. Fill `testid=automation-setup-name` with `QA_F23 Stale renamed` and click `testid=automation-setup-save`: the page keeps the edit (`browser value 'testid=automation-setup-name'` still `QA_F23 Stale renamed`) and `browser toasts --history` holds exactly one toast. Expected: `Automation not found` (BASE's modal showed it). Known failure (reproduced 2026-10-13, regression from #17688, #18380): the toast reads `Request failed with status code 404`. Screenshot with `--feature F23.edit-save-error --name toast`, move the pointer off the toast and Close. `browser errors --app-only` lists the induced `POST /api/automation/v1/<Fail id>/dispatch` and `PATCH /api/automation/v1/<stale-id>` 404s; both fixtures are already deleted.
+- **Run now and save errors (`F23.run-now-error`, `F23.edit-save-error`).** Arrange a stale page: open QA_F23 Fail, then `control-openhands api DELETE /api/automation/v1/<Fail id> --write`. Run `control-openhands browser toasts --history --clear` (empties the toast history), click `role=button[name="Run now"]`, then `control-openhands browser toasts --history` and `control-openhands browser screenshot --feature F23.run-now-error --name toasts`: exactly one toast, `Automation not found`, and no `Request failed with status code 404`. Expected: the page then shows **Automation not found**: `control-openhands browser wait 'role=heading[name="QA_F23 Fail"]' --state detached --timeout 15000` succeeds (the not-found state takes about 7 s, as in States) and `control-openhands browser snapshot 'role=paragraph >> text=Automation not found >> xpath=..'` lists link `Back to Automations` (do not use `wait-text 'Automation not found'` here: the toast has the same text). Once the heading is gone, run `control-openhands browser toasts --history` again: it still holds only `Automation not found` (a late second toast, `Request failed with status code 404`, would come from the page's own refetch of the deleted automation). Known failure: the wait times out and the page keeps heading `QA_F23 Fail`, `Active` and an enabled **Run now** (`control-openhands browser enabled 'role=button[name="Run now"]'` is `true`) until a reload (#18062). The save error does not reuse that page (it is gone once #18062 is fixed): arrange a second fixture, `control-openhands api POST /api/automation/v1/preset/prompt --write --data '{"name":"QA_F23 Stale","prompt":"Reply with the single word: pong.","trigger":{"type":"cron","schedule":"0 0 1 1 *","timezone":"UTC"},"enabled":false}'` (`body.id` is `<stale-id>`), open QA_F23 Stale, its kebab and `role=list >> has-text=Tarball >> role=button[name="Edit"]` (`--expect-url '/automations/setup'`). With the edit page open, `control-openhands api DELETE /api/automation/v1/<stale-id> --write` and `control-openhands browser toasts --history --clear`. Fill `testid=automation-setup-name` with `QA_F23 Stale renamed` and click `testid=automation-setup-save`: the page keeps the edit (`browser value 'testid=automation-setup-name'` still `QA_F23 Stale renamed`) and `browser toasts --history` holds exactly one toast, `Automation not found`. Screenshot with `--feature F23.edit-save-error --name toast`. Move the pointer off the toast (`browser hover 'testid=automation-setup-name'`, about 5 s), `browser toasts --history --clear` and click `testid=automation-setup-test`: again exactly one toast, `Automation not found`. Close. `browser errors --app-only` lists the induced `POST /api/automation/v1/<Fail id>/dispatch`, `PATCH /api/automation/v1/<stale-id>` and `POST /api/automation/v1/<stale-id>/dispatch` 404s; both fixtures are already deleted.
- **Load more runs (`F23.activity-log-more`).** Not run: it needs more than 20 runs of one automation (about 35 s and one model call each). With them, `control-openhands browser click 'testid=automation-activity-log >> role=button[name="Load more"]'` grows the list from 20 to 40 rows.
- **Script automation (`F23.script-section`).** Arrange a script automation through F22's bundle entry (UI arrange; F22 owns the creation proof): `control-openhands browser goto /automations/templates`, `control-openhands browser click 'testid=recommended-automation-card-news-digest' --expect-url '/automations/new/news-digest'`, `control-openhands browser fill 'testid=setup-field-schedule' '0 0 1 1 *'`, `control-openhands browser click 'testid=setup-continue-button'`, `control-openhands browser wait 'testid=setup-review' --timeout 10000`, `control-openhands browser click 'testid=setup-continue-button' --expect-url '/automations/[0-9a-f-]{8,}' --timeout 60000`. The detail page of **Daily news digest** shows `role=heading[name="Script"]` and no Prompt (`browser count 'role=heading[name="Prompt"]'` is `0`); `control-openhands browser text 'testid=automation-script-entrypoint'` is `python3 main.py`, and `browser eval "Array.from(document.querySelectorAll('[data-testid=automation-script-file]')).map(e=>e.innerText.split('\n').slice(0,2).join(' | '))"` lists `main.py | Entrypoint` first, then `config.json` (both with line-numbered source). Screenshot with `--feature F23.script-section --name news-digest`. `automation-script-unavailable` (bundle unreadable) was not reached. Do not Run now: the script fetches public feeds and starts a model-backed conversation. Delete it in the Delete bullet.
- **Edit-mode Test and unsaved edits (`F23.edit-test`, `F23.edit-dirty-state`).** Open QA_F23 Event and its kebab **Edit** (`--expect-url 'automationId='`). Fill `testid=automation-setup-name` with `QA_F23 Unsaved`: `control-openhands browser count 'testid=automation-setup-save-state'` is `0`. `control-openhands browser network --clear`, click `testid=automation-setup-test`; `control-openhands browser network --filter 'automation/v1'` shows `POST …/<id>/dispatch 201` and no `PATCH`, and the form shows `Test run started.` Move the pointer off the toast (`control-openhands browser hover 'testid=automation-setup-prompt'`) and click `testid=automation-setup-close` (`--expect-url '/automations$'`): the card still reads `QA_F23 Event`. The dispatched run is not checked here; Delete below removes it with the automation.
@@ -118,7 +118,7 @@ Preconditions:
- The Logs dialog's panel keeps long content reachable (`F23.run-logs-modal`, #18175 / #18173). Re-verified 2026-10-13 (revision `2bdc0e9d`, SDK 1.54.0, automation 1.19.3) with a 30-line `outcome_summary` at 1440×1000, 390×844 and 390×500: the panel is `testid=run-logs-body` (`max-h-[80vh]`, `overflow-y: auto`) with a sticky header (Logs + X stay visible); the Output box no longer scrolls on its own, the whole panel does, and its end (`=== RESULT === ALL_OK`) is reachable at the bottom. Verified 2026-10-10 on an independent fresh Linux Local stack at product revision `5a8c3832154b14d2dad651801c0ab3f74c96f02e`, SDK 1.54.0 and automation 1.19.3: a genuine DeepSeek finish produced a 4,596-character `outcome_summary` and 1,079-character metadata `message`; summary/metadata endings, Output/Error tabs and the true stdout end were reachable at 1440×1000, 390×844, 1440×500 and 390×500. The header/X remained visible, closing did not navigate, and phone had no horizontal overflow. Preserve those long-content and short-height checks when retesting. The verified scope is pinned Linux Local; macOS, Cloud and legacy/custom finish schemas need their own checks. `browser bbox 'role=dialog'` measures the full-screen wrapper, not the panel: measure the panel with `browser eval`.
- The detail page has no `main` landmark: `browser snapshot 'role=main'` times out. Snapshot the page, or scope by section headings (`role=heading[name="Configuration"] >> xpath=ancestor::div[2]`).
- The kebab menu has no test ids; `role=list >> has-text=Tarball` scopes the detail kebab (the dashboard kebab has no Tarball, so F21 uses `has-text=Export`).
-- Since #17688 kebab **Edit** (detail and dashboard) opens `/automations/setup?automationId=<id>`; the old **Edit automation** modal (`edit-automation-*` test ids) is no longer rendered anywhere, so its Escape (#17563), phone clipping (#17563) and silently-ignored cleared time (#17948) symptoms cannot be reached. The edit page's Save reports API failures with the raw `Request failed with status code <N>` (see `F23.edit-validation`, `F23.edit-save-error`), and its `Automation updated` toast covers Test/Save/Close until the pointer moves away.
+- Kebab **Edit** (detail and dashboard) opens `/automations/setup?automationId=<id>`. The edit page's Save and Test report API failures with the API's message (see `F23.edit-validation`, `F23.edit-save-error`), and its `Automation updated` toast covers Test/Save/Close until the pointer moves away.
- Known defect: clearing **At** on a preset schedule and saving stores midnight with no message (#17948 item; with the old modal it was silently ignored).
- On a local backend the Logs dialog reads every run's output from the server-level `/api/bash/bash_events/search`, with or without a conversation. It caches what it fetched: reload before checking its requests with `browser network`.
- Known defect: prompt runs also have a `bash_command_id`, so every row shows the logs icon, and a prompt run that ended without a conversation is labelled as a script run (`F23.run-no-conversation`) (#17948).
diff --git a/.agents/skills/verify-openhands/references/feature-map/F25-backends-and-cloud.md b/.agents/skills/verify-openhands/references/feature-map/F25-backends-and-cloud.md
index eef03f5d54..c52c4c24ec 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F25-backends-and-cloud.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F25-backends-and-cloud.md
@@ -10,7 +10,8 @@ backends in recovery mode. Cloud backends add organization rows, a Cloud
settings link and public sharing of conversations (read-only at
`/shared/conversations/:id`); `/oauth/device/verify` is the device-approval page.
When the page is served with the DigitalOcean Managed Agents bridge (the
-desktop app, or a web server started with `MARS_WEB=1`), Add Backend gains a
+desktop app's preload, or the web server's own bridge, which the launcher's
+ingress mounts by default behind the session key), Add Backend gains a
**Providers** tab whose DigitalOcean entry signs in with a personal access
token and turns a Managed Agents session into a backend.
Everything except the Cloud and DigitalOcean accounts themselves can be driven
@@ -21,7 +22,7 @@ Source: `src/components/features/backends/`, `src/contexts/active-backend-contex
## Sub-features
- `F25.selector-dropdown`: clicking (or hovering) the selector opens a listbox of backends, each option named `<status> <name>` with a status dot, and a footer with **Add Backend** and **Manage Backends**.
-- `F25.add-backend-modal`: **Add Backend** opens "Choose how you want to connect" with a description, a **Learn more** link to the deployment docs and two tabs, **OpenHands Cloud** (selected) and **Agent-server**. Escape does not close it; the X does.
+- `F25.add-backend-modal`: **Add Backend** opens "Choose how you want to connect" with a description, a **Learn more** link to the deployment docs and the tabs **OpenHands Cloud** (selected) and **Agent-server** (plus **Providers** with the bridge, see the Providers row). It takes keyboard focus on open (the close button), Tab and Shift+Tab wrap inside it, and at desktop width closing returns focus to the backend selector (at phone width the drawer closes with the modal and focus falls back to the page body). Escape does not close it; the X does.
- `F25.cloud-advanced-host`: on the Cloud tab, **Advanced** reveals a **Host** field (placeholder `https://app.all-hands.dev`) for a self-hosted Cloud.
- `F25.cloud-device-flow`: **Connect to OpenHands** opens a popup and shows `Starting authentication...`, then `Waiting for authorization...` with the verification link and **Cancel**; Cancel returns to the idle button and closes the popup. An unreachable host shows `Failed to start device flow: ...` with **Try again**. Approving the code (adds a backend named OpenHands Cloud) needs a Cloud account.
- `F25.agent-server-guidance`: the Agent-server tab has a **Local / Remote** toggle. Local shows a collapsible **Before you connect** with `agent-canvas --backend-only --port 8001` and a local-setup docs link; Remote shows **Recommended setup**, **Connection details** and a remote-setup docs link.
@@ -29,7 +30,7 @@ Source: `src/components/features/backends/`, `src/contexts/active-backend-contex
- `F25.switch-backend`: choosing another option shows a full-screen `Switching to <name>` overlay for about a second, then the sidebar and pages show that backend's data; on `/conversations/<id>` the app moves to `/conversations` (BM-002).
- `F25.backend-pinned-url`: in-app links carry `?backend=<id>`; opening such a URL pins the tab to that backend; an unknown id is ignored.
- `F25.per-tab-backend`: the active backend is per browser tab: switching it in one tab leaves the other open tabs where they were, a new tab opened from a plain link starts from the last choice made in any tab, and a sidebar conversation row opened in a new tab (Control-click) lands on the backend that owns the conversation because its link carries `?backend=<owner>`.
-- `F25.manage-backends`: Manage backends lists every backend with status dot, name, version (`v1.50.1`), host, status text and a LOCAL/CLOUD pill, plus Edit and Remove; Escape, a backdrop click, X and **Done** close it.
+- `F25.manage-backends`: Manage backends lists every backend with status dot, name, version (`v1.50.1`), host, status text and a LOCAL/CLOUD pill, plus Edit and Remove; it takes keyboard focus on open and keeps Tab inside; Escape, a backdrop click, X and **Done** close it and return focus to the selector.
- `F25.manage-add`: **Add Backend** in Manage backends opens the same Add form stacked over Manage; its X returns to Manage, and a successful add keeps Manage open with the new row while the new backend becomes active.
- `F25.manage-select`: clicking a healthy row makes that backend active and closes the modal; on a detail page it leaves it for that section's list, like the selector does (BM-002).
- `F25.edit-backend`: the pencil opens "Edit backend" pre-filled (key masked) with a `Connected · Local v<version>` badge; empty name and invalid host show `Name is required` and `Enter a valid URL (e.g. http://localhost:8080)` and disable Save; Save re-tests the connection, keeps the modal open with `Could not connect to <host>` on failure, and persists on success.
@@ -48,7 +49,7 @@ Source: `src/components/features/backends/`, `src/contexts/active-backend-contex
- `F25.cloud-settings-link`: with a Cloud backend active the gear next to the selector opens `<cloud host>/settings?org=<org>` in a new tab.
- `F25.cloud-sandbox-states`: Cloud conversations show waiting, archived and error sandbox states in chat.
- `F25.cloud-org-suspended`: when a Cloud call made for the selected organization (`X-Org-Id`) fails with 403 and `detail` `Organization is suspended` or `User membership is suspended`, the app is replaced by `<org> is suspended. Contact your administrator to restore access.` (or `Your access to <org> is suspended. Contact your administrator to restore it.`), then `Switch to another workspace` and one button per other organization (`Personal Workspace` for the user's own). A button selects that organization and the app returns; with no other organization only the message shows. A reload clears the in-memory record. Blocked here: needs an OpenHands Enterprise instance with `ENABLE_SUPER_ADMIN` where a Super Admin suspended the organization or this member's membership, signed in as a non-Super-Admin member with that organization selected (#17988).
-- `F25.providers-tab`: only when the page has the Managed Agents bridge (`window.marsBridge`), Add Backend shows a third tab **Providers** ("Connect through DigitalOcean and more.") with a **DigitalOcean** card ("Your team's OpenHands agents") and `More providers coming soon.`; without the bridge (the default web build) there are only the Cloud and Agent-server tabs and no Managed Agents UI anywhere.
+- `F25.providers-tab`: only when the page has the Managed Agents bridge (`window.marsBridge`), Add Backend shows a third tab **Providers** ("Connect through DigitalOcean and more.") with a **DigitalOcean** card ("Your team's OpenHands agents") and `More providers coming soon.`. Every launcher-served web page has it (the ingress answers `/mars/health`); without it (`MARS_WEB=0`, or a non-loopback bind with no session key at all) there are only the Cloud and Agent-server tabs and no Managed Agents UI anywhere. A non-loopback bind that has a key (F26 `--host 0.0.0.0` with `LOCAL_BACKEND_API_KEY`) mounts the bridge behind that key and offers the tab too, although the page itself carries no key (see the sign-in bullet's Known failure).
- `F25.digitalocean-sign-in`: the DigitalOcean card opens "Connect your DigitalOcean team" with a Back button, a masked **Personal access token** field, a **Create a token** link to `https://cloud.digitalocean.com/account/api/tokens` and **Connect with token**, disabled until a token is typed (the **Sign in with DigitalOcean** OAuth button exists only in the desktop app). A token of the wrong shape reads `That does not look like a DigitalOcean personal access token.`; a well-formed token DigitalOcean rejects shows its HTTP error (`401 Unauthorized`). Back returns to the Providers tab.
- `F25.digitalocean-agents`: signed in, the panel lists the team's OpenHands agents and their sessions, **New Agent** (name and sandbox size), refresh and sign-out; connecting a session (or **Resume and connect** a paused one) adds it as the active backend, the sidebar's conversation list becomes that agent's session panel (`mars-session-panel`), and Manage Backends' Managed Agents view offers Pause, Stop, Disconnect and Open per session (blocked: needs a DigitalOcean account with Managed Agents).
- `F25.digitalocean-backend-removal`: removing a DigitalOcean backend in Manage Backends explains that it only disconnects and offers **Disconnect only** or **Stop session** (blocked: needs a connected DigitalOcean session).
@@ -74,21 +75,21 @@ Preconditions:
- A second stack to add as a backend, started without a browser and never driven directly: `export OH_VERIFY_RUN_2=$(OH_VERIFY_RUN= control-openhands launch --new --build never --no-browser --print-run)`, then `control-openhands status --run "$OH_VERIFY_RUN_2"`; its `baseUrl` is `<second-url>` below and its key is `$OH_VERIFY_RUN_2/private/session-key`. It has no conversations.
- No backend named `QA_Second` or `QA_Renamed` exists (the selector lists only `Local`).
- Bullets run in order: Add creates `QA_Second`, Edit renames it to `QA_Renamed`, Remove deletes it.
-- The Managed Agents rows need the bridge, which `control-openhands launch` cannot turn on (harness gap: no `--mars-web` flag, and `MARS_WEB` is not passed to the launcher). Until it exists, on a run you launched yourself, add `"MARS_WEB": "1"` to `<run>/private/launcher-env.json` (keep a copy), `control-openhands restart`, `browser reload`, and check `control-openhands browser eval "!!window.marsBridge"` is `true` (`curl -s <baseUrl>/mars/health` answers `{"ok":true,"authRequired":true}`). Restore the file and `restart` afterwards. `F25.digitalocean-agents` and `F25.digitalocean-backend-removal` stay blocked without a DigitalOcean account.
-- Blocked here (no OpenHands Cloud account): approving the device flow, `F25.cloud-org-rows`, `F25.cloud-log-back-in`, `F25.cloud-settings-link`, `F25.cloud-sandbox-states`, `F25.cloud-org-suspended` (also needs a suspended OpenHands Enterprise organization or membership), `F25.share-publicly` (toggle), and `F25.locked-cloud`, which also needs a launcher flag (see Gotchas).
+- The Managed Agents rows need the bridge, which every `control-openhands launch` run has: `control-openhands browser eval "!!window.marsBridge"` is `true` and `curl -s <baseUrl>/mars/health` answers `{"ok":true,"authRequired":true}`. `F25.digitalocean-agents` and `F25.digitalocean-backend-removal` stay blocked without a DigitalOcean account.
+- Cloud account: where this machine's proxy injects OpenHands Cloud credentials (`curl -s -o /dev/null -w '%{http_code}' https://app.all-hands.dev/api/organizations` is `200` without a token), the **Cloud backend** bullet below can approve the device flow in its popup; otherwise it is blocked. Blocked here even with such an account: `F25.cloud-log-back-in` (needs a logged-out Cloud row), org switching in `F25.cloud-org-rows` (needs a second organization), the archived and error states of `F25.cloud-sandbox-states`, `F25.cloud-org-suspended` (also needs a suspended OpenHands Enterprise organization or membership), `F25.share-publicly` (toggle), and `F25.locked-cloud`, which also needs a launcher flag (see Gotchas).
- **Open the selector (`F25.selector-dropdown`).** From `/conversations` run `control-openhands browser click 'testid=backend-selector'` and `control-openhands browser snapshot 'testid=backend-selector'`. The snapshot shows `listbox` with `option "Connected Local"`, a separator, `button "Add Backend"` and `button "Manage Backends"`. Screenshot with `control-openhands browser screenshot --feature F25.selector-dropdown --name open`. Hover opens it too: `browser press Escape`, move the pointer away with `control-openhands browser hover 'testid=command-menu-trigger'` (hovering a selector the pointer is already on does nothing), `control-openhands browser hover 'testid=backend-selector'`, then the same snapshot shows `combobox "Local" [expanded]` and the two footer buttons.
-- **Add Backend modal (`F25.add-backend-modal`).** Run `control-openhands browser click 'testid=add-backend-menu-item'` and `control-openhands browser snapshot 'testid=add-backend-modal'`: heading `Choose how you want to connect`, link `Compare Cloud, self-hosted, and Enterprise deployment options` (`/url: https://docs.openhands.dev/overview/introduction`, text `Learn more`) and tab `OpenHands Cloud ...` `[selected]` next to tab `Agent-server ...`. `control-openhands browser press Escape` leaves `browser count 'testid=add-backend-modal'` at `1`; `control-openhands browser click 'testid=add-backend-close'` makes it `0`. Keyboard focus in a light theme: arrange `control-openhands browser eval "localStorage.setItem('openhands-color-theme','light-plus')"` and `browser reload`, reopen the modal, run `control-openhands browser focus 'testid=add-backend-option-cloud'` and `browser press ArrowRight`, then `control-openhands browser eval "getComputedStyle(document.activeElement).outlineColor"`. Expected: the theme's focus colour (`--oh-focus`, `#007ACC` in Light+). Known failure (reproduced 2026-10-13, #18324): a fixed pale `outline-blue-300` (`oklab(0.79 -0.03 -0.10)`), about 1.8:1 on the light tab (`browser screenshot 'testid=add-backend-modal' --feature F25.add-backend-modal --name tab-focus-light-plus`). Restore with `browser eval "localStorage.removeItem('openhands-color-theme')"`.
+- **Add Backend modal (`F25.add-backend-modal`).** Run `control-openhands browser click 'testid=add-backend-menu-item'` and `control-openhands browser snapshot 'testid=add-backend-modal'`: heading `Choose how you want to connect`, link `Compare Cloud, self-hosted, and Enterprise deployment options` (`/url: https://docs.openhands.dev/overview/introduction`, text `Learn more`) and tab `OpenHands Cloud ...` `[selected]` next to tab `Agent-server ...`. Keyboard focus: `control-openhands browser eval "(()=>{const a=document.activeElement;const d=document.querySelector('[data-testid=add-backend-modal]');return {tid:a.getAttribute('data-testid'),inDialog:!!(d&&d.contains(a))}})()"` is `{tid: "add-backend-close", inDialog: true}` right after opening; after eight `control-openhands browser press Tab` and three `browser press Shift+Tab` (reading the same eval after each) `inDialog` stays `true` (Tab past **Advanced** wraps to the close button), and `browser press q` then `control-openhands browser eval "document.querySelector('[data-testid=chat-input]').textContent"` is `""` (nothing reaches the Home composer behind). `browser screenshot --feature F25.add-backend-modal --name focus-trap-open`. `control-openhands browser press Escape` leaves `browser count 'testid=add-backend-modal'` at `1`; `control-openhands browser click 'testid=add-backend-close'` makes it `0`, and `control-openhands browser eval "!!document.activeElement.closest('[data-testid=backend-selector]')"` is `true` (focus is back on the selector). Keyboard focus in a light theme: arrange `control-openhands browser eval "localStorage.setItem('openhands-color-theme','light-plus')"` and `browser reload`, reopen the modal, run `control-openhands browser focus 'testid=add-backend-option-cloud'` and `browser press ArrowRight`, then `control-openhands browser eval "getComputedStyle(document.activeElement).outlineColor"`. Expected: the theme's focus colour (`--oh-focus`, `#007ACC` in Light+). Known failure (reproduced 2026-10-13, #18324): a fixed pale `outline-blue-300` (`oklab(0.79 -0.03 -0.10)`), about 1.8:1 on the light tab (`browser screenshot 'testid=add-backend-modal' --feature F25.add-backend-modal --name tab-focus-light-plus`). Restore with `browser eval "localStorage.removeItem('openhands-color-theme')"`.
- **Advanced host (`F25.cloud-advanced-host`).** Reopen with `browser click 'testid=backend-selector'` and `browser click 'testid=add-backend-menu-item'`, run `control-openhands browser click 'testid=add-backend-advanced-toggle'`, then `control-openhands browser attr 'testid=add-backend-advanced-toggle' aria-expanded` (`true`) and `control-openhands browser snapshot 'testid=add-backend-cloud-panel'` (textbox `Host` with placeholder `https://app.all-hands.dev`).
- **Device flow error (`F25.cloud-device-flow`).** Run `control-openhands browser fill 'testid=add-backend-cloud-host' http://127.0.0.1:9`, `control-openhands browser click 'testid=add-backend-login-button' --observe 'testid=add-backend-device-flow'` and `control-openhands browser wait 'testid=add-backend-auth-error' --timeout 15000`. The observation goes `Starting authentication...` → `Failed to start device flow: Failed to fetch` / `Try again`, and `control-openhands browser network --last 5` shows `POST /oauth/device/authorize` to `http://127.0.0.1:9`. `control-openhands browser click 'testid=add-backend-auth-retry'` starts again and fails the same way. Each attempt leaves an `about:blank` popup: `control-openhands browser tabs`, then `control-openhands browser close-tab <i>` for each extra page and `control-openhands browser tab 0`. Close the modal with `testid=add-backend-close`.
-- **Device flow awaiting and Cancel (`F25.cloud-device-flow`).** Reopen Add Backend (Cloud tab, default host), run `control-openhands browser click 'testid=add-backend-login-button' --observe 'testid=add-backend-device-flow' --observe-ms 6000`, then `control-openhands browser snapshot 'testid=add-backend-device-flow'`. If `https://app.all-hands.dev` is reachable (`control-openhands browser network --external --last 5` lists it), the snapshot shows `Waiting for authorization...`, `Browser opened. Complete sign-in to continue. If browser didn't open, visit:`, a link `https://app.all-hands.dev/oauth/device/verify?user_code=<code>` and button `Cancel`; `browser tabs` lists a second page on that URL. Screenshot with `--feature F25.cloud-device-flow --name awaiting`, then `control-openhands browser click 'testid=add-backend-auth-cancel'`: the panel is back to `Connect to OpenHands` / `Advanced` and `browser tabs` lists one page. Never approve the code. If Cloud is unreachable you get the error state of the previous bullet instead.
+- **Device flow awaiting and Cancel (`F25.cloud-device-flow`).** Reopen Add Backend (Cloud tab, default host), run `control-openhands browser click 'testid=add-backend-login-button' --observe 'testid=add-backend-device-flow' --observe-ms 6000`, then `control-openhands browser snapshot 'testid=add-backend-device-flow'`. If `https://app.all-hands.dev` is reachable (`control-openhands browser network --external --last 5` lists it), the snapshot shows `Waiting for authorization...`, `Browser opened. Complete sign-in to continue. If browser didn't open, visit:`, a link `https://app.all-hands.dev/oauth/device/verify?user_code=<code>` and button `Cancel`; `browser tabs` lists a second page on that URL. Screenshot with `--feature F25.cloud-device-flow --name awaiting`, then `control-openhands browser click 'testid=add-backend-auth-cancel'`: the panel is back to `Connect to OpenHands` / `Advanced` and `browser tabs` lists one page. Do not approve the code here (the **Cloud backend** bullet does, when it can). If Cloud is unreachable you get the error state of the previous bullet instead.
- **Agent-server guidance (`F25.agent-server-guidance`).** Run `control-openhands browser click 'testid=add-backend-option-agent-server'` and `control-openhands browser snapshot 'testid=add-backend-agent-server-panel'`: radiogroup `Agent-server location` with `Local` checked, heading `Before you connect`, textboxes `Host Name`, `Host`, `API Key` and `button "Connect" [disabled]`. `control-openhands browser click 'testid=add-backend-local-guidance-toggle'` and `browser snapshot 'testid=add-backend-local-guidance'` show the code `agent-canvas --backend-only --port 8001` and link `Read the local backend setup guide`. `control-openhands browser click 'testid=add-backend-location-option-remote'` and `control-openhands browser click 'testid=add-backend-remote-guidance-toggle'`; `browser snapshot 'testid=add-backend-remote-guidance'` shows `Recommended setup`, `Connection details` and `Read the remote backend setup guide`.
- **Connect rules and errors (`F25.add-agent-server`).** Still on Remote: `control-openhands browser fill 'testid=add-backend-name' QA_Second`, `control-openhands browser fill 'testid=add-backend-host' 127.0.0.1:9`; `control-openhands browser enabled 'testid=add-backend-submit'` is `false` (Remote needs a key). `control-openhands browser fill 'testid=add-backend-api-key' wrong-key` makes it `true`. `control-openhands browser click 'testid=add-backend-location-option-local'` keeps the fields. `control-openhands browser click 'testid=add-backend-submit' --observe 'testid=add-backend-submit'` shows `Checking…`, then `control-openhands browser text 'testid=add-backend-error'` starts `Could not connect to http://127.0.0.1:9` / `Disconnected (check URL or network). ...`. Fill `add-backend-host` with `<second-url>` and submit again: the error reads `Could not connect to <second-url>` / `Invalid API key`. On Local the key is optional for Connect but not for a key-protected host: `control-openhands browser fill 'testid=add-backend-api-key' ''` leaves `control-openhands browser enabled 'testid=add-backend-submit'` at `true` (`browser value 'testid=add-backend-api-key'` reports `length` `0`), and `control-openhands browser click 'testid=add-backend-submit'` then `control-openhands browser wait 'testid=add-backend-error' --timeout 15000` shows the same `Could not connect to <second-url>` / `Invalid API key` (`browser screenshot --feature F25.add-agent-server --name local-no-key-error`). Close with `testid=add-backend-close`.
- **Add the second backend (`F25.add-agent-server`).** Run `control-openhands browser goto /conversations/<id>`, `browser click 'testid=backend-selector'`, `browser click 'testid=add-backend-menu-item'`, `browser click 'testid=add-backend-option-agent-server'`, fill `add-backend-name` with `QA_Second` and `add-backend-host` with `<second-url>`, then `control-openhands browser fill 'testid=add-backend-api-key' --value-file "$OH_VERIFY_RUN_2/private/session-key"` and `control-openhands browser click 'testid=add-backend-submit' --expect-url '/conversations(\?|$)'`. The URL is `/conversations`, `browser count 'testid=add-backend-modal'` is `0`, `browser snapshot 'testid=backend-selector'` shows `combobox "QA_Second"`, and `browser count 'testid=conversation-card'` is `0`. Before anything reloads the page, `control-openhands browser toasts --history` lists only `Loading...` and the new backend's `Your LLM isn't set up yet, so conversations won't run. ...` (the second stack has no LLM profile; no error toast), and `control-openhands browser network --external --filter 'ids='` lists no request: the new backend is never asked for the first stack's conversation id. Known failure (reproduced 2026-10-13, #18390): on the run's first activation of `QA_Second` the list sometimes holds `GET <second-url>/api/conversations?ids=<id>` (`200`); record the bullet as `fail` then (see Gotchas). (`--external` leaves out this stack's own `GET /api/conversations?ids=<id>` from the `goto /conversations/<id>` above, which the request log keeps across page loads; `<second-url>` is another origin, so a request to it is listed.) The new backend asks its own telemetry consent ("This preference is saved for the local backend “QA_Second”"): run `control-openhands onboard --skip` (it leaves the browser on `/`). After `control-openhands browser reload` the selector still reads `QA_Second`.
- **Switch with the overlay (`F25.switch-backend`).** From `/` run `browser click 'testid=backend-selector'` and `control-openhands browser click 'testid=backend-selector >> role=option[name="Connected Local"]' --observe 'testid=environment-switch-overlay' --observe-ms 2500`. The observation shows `Switching to Local`, then `<absent>` about one second later; the selector reads `Local` and `browser count 'testid=conversation-card'` is `1`. Then `control-openhands browser click 'testid=conversation-card' --expect-url '/conversations/[0-9a-f-]+'`, open the selector and `control-openhands browser click 'testid=backend-selector >> role=option[name="Connected QA_Second"]' --expect-url '/conversations(\?|$)'`: the URL is `/conversations`, the selector reads `QA_Second` and the card count is `0`.
- **Pinned URL (`F25.backend-pinned-url`).** With `QA_Second` active run `control-openhands browser goto '/conversations/<id>?backend=default-local'` and `control-openhands browser wait 'testid=chat-interface'`: the selector reads `Local` and the conversation opens. `control-openhands browser goto '/conversations?backend=qa-unknown-id'` keeps `Local` (card count `1`). The sidebar card links themselves end in `?backend=default-local` (`browser snapshot`).
- **Per-tab backend (`F25.per-tab-backend`).** With `Local` active and `QA_Second` registered, run `control-openhands browser goto /conversations` and `control-openhands browser eval "JSON.parse(sessionStorage.getItem('openhands-active-backend')).backendId"` (`default-local`); note QA_Second's id with `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-backends')).find(b=>b.name==='QA_Second').id"` (`<second-id>`, a generated UUID). Open a plain second tab, as a user opening the app again: `control-openhands browser tab new /conversations` prints `index` `1` and makes it the daemon's active tab (`control-openhands browser tabs` lists two pages at `/conversations` with `active` `1`); `control-openhands browser wait-tab '/conversations$' --new --timeout 5000` then prints `index` `0`, the tab you came from, because `--new` leaves out the tab the daemon is on and that is now the new one (run from `browser tab 0` it prints `index` `1`, the second tab). `control-openhands browser tab 1`, `control-openhands browser wait 'testid=backend-selector'`: `browser snapshot 'testid=backend-selector'` reads `combobox "Local"` and `control-openhands browser eval "sessionStorage.getItem('openhands-active-backend')"` is `null` (a new tab inherits nothing and starts from the localStorage fallback). Switch this tab: `browser click 'testid=backend-selector'`, `control-openhands browser click 'testid=backend-selector >> role=option[name="Connected QA_Second"]'`, `control-openhands browser wait 'testid=environment-switch-overlay' --state detached --timeout 10000`; the selector reads `QA_Second`, `browser count 'testid=conversation-card'` is `0` and `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-active-backend')).backendId"` is `<second-id>`. Back in `control-openhands browser tab 0` nothing moved: the selector still reads `Local`, the card count is `1`, the sessionStorage eval is still `default-local` while the localStorage eval is `<second-id>` (`browser screenshot --feature F25.per-tab-backend --name first-tab-unchanged`). A plain new tab starts from the last choice made in any tab: from tab 0 run `control-openhands browser tab new /conversations` (`index` `2`, now active), `browser wait 'testid=backend-selector'`: the selector reads `QA_Second`, the card count is `0`, its sessionStorage item is `null` and the localStorage eval is `<second-id>` (`--name plain-new-tab-last-choice`). A conversation row opens on its owner instead: `control-openhands browser tab 0`, `control-openhands browser click 'testid=conversation-card' --modifiers Control` (add `>> nth=0` when the run holds more conversations than the precondition's one: the click is strict), `control-openhands browser wait-tab '/conversations/<id>' --timeout 20000` prints `index` `3` and the URL `/conversations/<id>?backend=default-local`; `control-openhands browser tab 3`, `control-openhands browser wait 'testid=chat-interface' --timeout 30000`, `browser url` still ends in `/conversations/<id>?backend=default-local` (no bounce to `/conversations`), the selector reads `Local` and `browser toasts --history` lists only `Loading...` (no `This conversation does not exist…`); `browser screenshot --feature F25.per-tab-backend --name new-tab-on-owner`. Close the extra tabs highest first: `control-openhands browser close-tab 3`, `control-openhands browser close-tab 2`, `control-openhands browser close-tab 1`, `control-openhands browser tab 0`; `browser tabs` lists one page and the selector reads `Local`. Opening the pinned link also recorded `default-local` as the last choice (the localStorage eval in tab 3 reads `default-local`), so run the plain-new-tab check before the row check.
-- **Manage list and closing (`F25.manage-backends`).** Run `browser click 'testid=backend-selector'`, `control-openhands browser click 'testid=manage-backends-menu-item'` and `control-openhands browser snapshot 'testid=manage-backends-modal'`: heading `Manage backends`, two rows `Connected Local v1.50.1 <this run's URL> Connected Local` and `Connected QA_Second v1.50.1 <second-url> Connected Local`, each with `Edit` and `Remove`, then `Add Backend` and `Done`. `browser text 'testid=manage-backends-status-QA_Second'` is `Connected` and `browser text 'testid=manage-backends-version-QA_Second'` is `v1.50.1` (whatever `doctor` reports). Screenshot with `--feature F25.manage-backends --name list`. Close it four ways, reopening between: `browser press Escape`, `control-openhands browser mouse-click 20 500` (backdrop), `browser click 'testid=close-manage-backends-modal'`, `browser click 'testid=manage-backends-done'`; each time `browser count 'testid=manage-backends-modal'` is `0`.
+- **Manage list and closing (`F25.manage-backends`).** Run `browser click 'testid=backend-selector'`, `control-openhands browser click 'testid=manage-backends-menu-item'` and `control-openhands browser snapshot 'testid=manage-backends-modal'`: heading `Manage backends`, two rows `Connected Local v1.50.1 <this run's URL> Connected Local` and `Connected QA_Second v1.50.1 <second-url> Connected Local`, each with `Edit` and `Remove`, then `Add Backend` and `Done`. `browser text 'testid=manage-backends-status-QA_Second'` is `Connected` and `browser text 'testid=manage-backends-version-QA_Second'` is `v1.50.1` (whatever `doctor` reports). Screenshot with `--feature F25.manage-backends --name list`. Focus is inside: the Add Backend bullet's focus eval with `manage-backends-modal` in place of `add-backend-modal` reads `close-manage-backends-modal` on open and stays `inDialog` `true` over nine Tabs and two Shift+Tabs (with two backends listed, Tab 8 reaches **Done** and Tab 9 wraps to the close button); after Escape, focus is back on the selector. Close it four ways, reopening between: `browser press Escape`, `control-openhands browser mouse-click 20 500` (backdrop), `browser click 'testid=close-manage-backends-modal'`, `browser click 'testid=manage-backends-done'`; each time `browser count 'testid=manage-backends-modal'` is `0`.
- **Select from Manage (`F25.manage-select`).** With `Local` active, `control-openhands browser click 'testid=conversation-card' --expect-url '/conversations/[0-9a-f-]+'`, open Manage Backends, then `control-openhands browser click 'testid=manage-backends-row-QA_Second >> role=button >> nth=0' --expect-url '/conversations(\?|$)'`. Like the selector, the app leaves the detail page for `/conversations` (BM-002): `browser count 'testid=manage-backends-modal'` is `0`, the selector reads `QA_Second`, `browser count 'testid=chat-interface'` and `browser count 'testid=conversation-card'` are `0`, and `browser screenshot --feature F25.manage-select --name after-select` shows QA_Second's Home with its empty conversation list.
- **Edit validation and errors (`F25.edit-backend`).** Open Manage Backends, `control-openhands browser click 'testid=manage-backends-edit-QA_Second'`, then `browser value 'testid=edit-backend-host'` (`<second-url>`), `browser attr 'testid=edit-backend-api-key' type` (`password`) and `browser text 'testid=edit-backend-status'` (`Connected`, `·`, `Local`, `v1.50.1` on separate lines). Fill `edit-backend-name` with an empty string and `edit-backend-host` with `not a host`, then `browser focus 'testid=edit-backend-api-key'` and `browser snapshot 'testid=edit-backend-form'`: alerts `Name is required` and `Enter a valid URL (e.g. http://localhost:8080)`; `browser enabled 'testid=edit-backend-submit'` is `false`.
- **Escape in Edit (`F25.edit-escape`).** In the same Edit modal press `control-openhands browser press Escape`. Expected: the Edit modal stays (it disables Escape) or only it closes. Today `browser count 'testid=edit-backend-modal'` and `browser count 'testid=manage-backends-modal'` are both `0`: the whole stack closes and the draft is lost. Nothing was saved (`browser snapshot 'testid=backend-selector'` still reads `QA_Second`).
@@ -99,9 +100,10 @@ Preconditions:
- **Add from Manage, Edit Cancel and X (`F25.manage-add`, `F25.edit-cancel`).** The second stack is running again after the previous bullet. Run `browser goto /conversations/<id>`, `browser wait 'testid=chat-interface'`, open the selector, click `manage-backends-menu-item`, then `control-openhands browser click 'testid=manage-backends-add'` and `control-openhands browser wait 'testid=add-backend-modal' --timeout 5000` (`visible`); `browser count 'testid=manage-backends-modal'` is `1` underneath. `browser click 'testid=add-backend-close'`: the Add count is `0` and the Manage count stays `1`. Click `manage-backends-add` again, `add-backend-option-agent-server`, fill `add-backend-name` with `QA_Second`, `add-backend-host` with `<second-url>` and `add-backend-api-key` with `--value-file "$OH_VERIFY_RUN_2/private/session-key"`, then `browser click 'testid=add-backend-submit' --expect-url '/conversations(\?|$)'`: the URL is `/conversations`, `count 'testid=add-backend-modal'` is `0`, `count 'testid=manage-backends-modal'` is still `1`, `count 'testid=manage-backends-row-QA_Second'` is `1` and the selector reads `QA_Second` (screenshot `--feature F25.manage-add --name after-add`). No consent prompt this time: it is stored on the second backend. Then `control-openhands browser click 'testid=manage-backends-edit-QA_Second'`, `browser fill 'testid=edit-backend-name' QA_Draft` and `control-openhands browser click 'testid=edit-backend-cancel'`: `count 'testid=edit-backend-modal'` `0`, `count 'testid=manage-backends-modal'` `1`, row `QA_Second` `1`, row `QA_Draft` `0`. Repeat with `control-openhands browser click 'testid=edit-backend-close'`: same counts, and reopening the pencil shows `browser value 'testid=edit-backend-name'` `QA_Second`; leave with `edit-backend-cancel`. Remove it again: `browser click 'testid=manage-backends-remove-QA_Second'`, `browser click 'testid=confirmation-modal >> testid=confirm-button'`, `browser click 'testid=manage-backends-done'` and `browser reload`: the selector reads `Local`.
- **Collapsed rail (`F25.collapsed-entry`).** Run `control-openhands browser click 'testid=sidebar-collapse-toggle'` (`attr 'aside[data-collapsed]' data-collapsed` is `true`), `control-openhands browser hover 'testid=collapsed-backend-selector-link'`, `control-openhands browser click 'testid=add-backend-menu-item'` and `control-openhands browser wait 'testid=add-backend-modal' --timeout 5000` (`visible`). Close it, hover again, click `manage-backends-menu-item` and `browser wait 'testid=manage-backends-modal' --timeout 5000`. Close with Done and expand with `browser click 'testid=sidebar-collapse-toggle'`.
- **Phone (`F25.phone`).** Run `control-openhands browser viewport phone`, `browser goto /conversations`, `control-openhands browser click 'testid=sidebar-mobile-menu-toggle'`, `control-openhands browser click 'testid=sidebar-mobile-drawer >> testid=backend-selector'`, `browser click 'testid=add-backend-menu-item'`, `browser click 'testid=add-backend-option-agent-server'` and `control-openhands browser bbox 'testid=add-backend-modal'`: `insideViewport` `true`, `pageHorizontalOverflow` `false`. Screenshot `--feature F25.phone --name add-modal`. Close it (the drawer stays open), open Manage Backends from the drawer selector the same way and check `bbox 'testid=manage-backends-modal'` the same. Click `testid=manage-backends-edit-Local` and check `bbox 'testid=edit-backend-modal'`: `insideViewport` `true`; repeat at `browser viewport 320x568`. At `browser viewport 320x400`, clear `edit-backend-name`, fill `edit-backend-host` with `not a host` and focus `edit-backend-api-key` to show both validation messages. Run `browser scroll 'testid=edit-backend-form' --by 600`: Cancel and Save must come into view while the title and `edit-backend-close` stay visible. Check the modal and all three controls with `bbox`, then close with Cancel. Return with `browser viewport desktop`; reopen Edit and confirm its width is still 520px.
-- **Providers tab (`F25.providers-tab`).** Without the bridge (the default run), `browser snapshot 'testid=add-backend-modal'` lists only the `OpenHands Cloud` and `Agent-server` tabs and `browser eval "!!window.marsBridge"` is `false`. With the bridge on (Preconditions), reopen Add Backend: the tablist adds `tab "Providers Connect through DigitalOcean and more."` (`testid=add-backend-option-providers`); `control-openhands browser click 'testid=add-backend-option-providers'` and `browser snapshot 'testid=add-backend-providers-panel'` show `button "DigitalOcean Your team's OpenHands agents"` and `More providers coming soon.` (`browser screenshot --feature F25.add-backend-modal --name with-providers-tab`). At phone width the modal stays inside the viewport with the three tab titles cut to `Ope…`, `Age…`, `Pro…` (`--feature F25.providers-tab --name phone-tabs`).
-- **DigitalOcean sign-in (`F25.digitalocean-sign-in`).** With the bridge on, run `control-openhands browser click 'testid=add-backend-provider-digitalocean'` and `browser wait 'testid=managed-agents-sign-in'`: heading `Connect your DigitalOcean team`, `textbox "Personal access token"` (`browser attr 'testid=managed-agents-token' type` is `password`), link `Create a token` (`/url: https://cloud.digitalocean.com/account/api/tokens`) and `button "Connect with token" [disabled]`; no OAuth button in a browser (`browser count 'testid=managed-agents-oauth'` is `0`). Fill `testid=managed-agents-token` with `dop_v1_qa_invalid_token_0000` and click `testid=managed-agents-token-submit`: `browser text 'testid=managed-agents-auth-error'` reads `That does not look like a DigitalOcean personal access token.` (`POST /mars/rpc/savePat` `500`). Fill a well-formed dummy, `dop_v1_` plus 64 zeros, and submit again: after a few seconds the error reads `401 Unauthorized` (`POST /mars/rpc/savePat` `401`, DigitalOcean's answer; `browser screenshot --feature F25.digitalocean-sign-in --name rejected-token`). Never type a real token here. `control-openhands browser click 'testid=add-backend-providers-back'` returns to the Providers panel (`browser count 'testid=managed-agents-sign-in'` is `0`; the providers panel itself stays mounted behind the sign-in screen, so its count is `1` either way); close with `testid=add-backend-close`. At phone width the sign-in screen fits (`bbox 'testid=add-backend-modal'` `insideViewport` `true`).
+- **Providers tab (`F25.providers-tab`).** On the default run `browser eval "!!window.marsBridge"` is `true` (Preconditions); `curl -s "http://127.0.0.1:$(control-openhands status | jq -r .ports.frontend)/mars/health"` (the static child behind the ingress) returns the HTML index, not JSON: one bridge per stack. Open Add Backend: the tablist adds `tab "Providers Connect through DigitalOcean and more."` (`testid=add-backend-option-providers`); `control-openhands browser click 'testid=add-backend-option-providers'` and `browser snapshot 'testid=add-backend-providers-panel'` show `button "DigitalOcean Your team's OpenHands agents"` and `More providers coming soon.` (`browser screenshot --feature F25.add-backend-modal --name with-providers-tab`). At phone width the modal stays inside the viewport with the three tab titles cut to `Ope…`, `Age…`, `Pro…` (`--feature F25.providers-tab --name phone-tabs`).
+- **DigitalOcean sign-in (`F25.digitalocean-sign-in`).** On the default run, run `control-openhands browser click 'testid=add-backend-provider-digitalocean'` and `browser wait 'testid=managed-agents-sign-in'`: heading `Connect your DigitalOcean team`, `textbox "Personal access token"` (`browser attr 'testid=managed-agents-token' type` is `password`), link `Create a token` (`/url: https://cloud.digitalocean.com/account/api/tokens`) and `button "Connect with token" [disabled]`; no OAuth button in a browser (`browser count 'testid=managed-agents-oauth'` is `0`). Fill `testid=managed-agents-token` with `dop_v1_qa_invalid_token_0000` and click `testid=managed-agents-token-submit`: `browser text 'testid=managed-agents-auth-error'` reads `That does not look like a DigitalOcean personal access token.` (`POST /mars/rpc/savePat` `500`). Fill a well-formed dummy, `dop_v1_` plus 64 zeros, and submit again: after a few seconds the error reads `401 Unauthorized` (`POST /mars/rpc/savePat` `401`, DigitalOcean's answer; `browser screenshot --feature F25.digitalocean-sign-in --name rejected-token`). Never type a real token here. `control-openhands browser click 'testid=add-backend-providers-back'` returns to the Providers panel (`browser count 'testid=managed-agents-sign-in'` is `0`; the providers panel itself stays mounted behind the sign-in screen, so its count is `1` either way); close with `testid=add-backend-close`. At phone width the sign-in screen fits (`bbox 'testid=add-backend-modal'` `insideViewport` `true`). Both errors also raise an error toast with the same text. Expected on a page served without a baked key (F26 `--host 0.0.0.0`, Docker), once the server's key was typed on the API-key screen: the same two answers as above. Known failure (reproduced 2026-10-14, #18434): every token, even the malformed one, ends in `session key required` (`POST /mars/rpc/getAuthState` and `savePat` `401`), because the web bridge client sends only the baked page key, never the typed one; it predates #18349 (BASE with `MARS_WEB=1` fails the same way). A `--frontend-only` page connected to another stack's backend fails the same way for a different reason (the typed key belongs to the other stack); #18434 leaves that case to a maintainer decision.
- **Managed Agents with an account (`F25.digitalocean-agents`, `F25.digitalocean-backend-removal`).** Blocked: needs a DigitalOcean account with Managed Agents. With one, after a real token the panel lists agents (`digitalocean-agent-<id>`), **New Agent** opens `digitalocean-new-agent-form` (name, size), a session's Connect adds a backend and the sidebar shows `mars-session-panel`; removing that backend opens `mars-backend-removal-modal` with `disconnect-only-button` and `stop-session-button`.
+- **Cloud backend (`F25.cloud-device-flow`, `F25.cloud-org-rows`, `F25.cloud-settings-link`, `F25.share-publicly`, `F25.cloud-sandbox-states`).** Only with a Cloud account (Preconditions); it creates a device grant and one conversation on that account, and shows the account's own conversations in the sidebar (keep those screenshots private). Reopen Add Backend on the Cloud tab, `browser click 'testid=add-backend-login-button'`, then `control-openhands browser tab 1` (the verify popup: `Device Authorization Request`, the code, **Authorize Device**), `control-openhands browser click 'role=button[name="Authorize Device"]'` (`Success!`) and `control-openhands browser tab 0`. Within about 15 s (token polling answers `400` until approved, then `200`) `browser count 'testid=add-backend-modal'` is `0`, the popup is gone and `browser snapshot 'testid=backend-selector'` reads `combobox "OpenHands Cloud": OpenHands Cloud – Personal Workspace`; the registry holds `{name: "OpenHands Cloud", host: "https://app.all-hands.dev", kind: "cloud"}` (`browser eval "JSON.parse(localStorage.getItem('openhands-backends')).map(b=>({name:b.name,host:b.host,kind:b.kind}))"`). Open the selector: `option "Connected OpenHands Cloud – Personal Workspace"` (one option per organization) next to `Connected Local`; Manage Backends shows the row with `Personal Workspace` and the `Cloud` pill. `browser attr 'testid=backend-selector-settings-link' href` is `https://app.all-hands.dev/settings?org=<org id>` with `target` `_blank`; clicking it opens that tab (`browser wait-tab 'app.all-hands.dev/settings'`, then `browser close-tab 1`, `browser tab 0`). Start one conversation from Home: fill `testid=chat-input` with `Reply with exactly: QA_F25_CLOUD. Do not run any tools.` and `browser click 'testid=submit-button' --expect-url '/conversations/'`: the URL is `/conversations/task-<id>` while the sandbox starts (toasts `Creating conversation…`, `Sending...`), then moves to `/conversations/<id>?backend=<cloud id>&org=<org id>` and the reply `QA_F25_CLOUD` appears (about a minute). Its `conversation-name >> ellipsis-button` menu has `share-publicly-button` (**Public Share**, off); do not flip it unless the conversation may be public. Delete it with `delete-button` and `'role=dialog >> testid=confirm-button'`, then remove the Cloud backend in Manage Backends (`manage-backends-remove-OpenHands Cloud`, confirm): after `browser reload` the selector reads `Local`.
- **Device verify page (`F25.device-verify-page`).** Run `control-openhands browser goto /oauth/device/verify` and `browser snapshot`: heading `Device Authorization`, `Enter the code displayed on your device:`, textbox `Device Code:` and button `Continue`, inside the normal app shell. `control-openhands browser fill 'role=textbox[name="Device Code:"]' QA-0000` and `browser click 'role=button[name="Continue"]'` go straight to the result: heading `Error`, `Failed to authorize device. Please try again.`, button `Try Again`; `browser network --filter verify-authenticated` shows `POST /oauth/device/verify-authenticated` `404`. Then `control-openhands browser goto '/oauth/device/verify?user_code=QA-0000'`: heading `Device Authorization Request`, `DEVICE CODE`, `QA-0000`, `Security Notice`, buttons `Cancel` and `Authorize Device` (screenshot `--feature F25.device-verify-page --name with-code`). `browser click 'role=button[name="Cancel"]'` does nothing in a tab the page did not open (URL unchanged). `browser click 'role=button[name="Authorize Device"]'` and `browser wait 'role=heading[name="Error"]'` show the same Error card; **Try Again** reloads to the request.
- **Shared view (`F25.shared-conversation-view`).** Run `control-openhands browser goto /shared/conversations/<id>` and `control-openhands browser wait-text 'Conversation not found' --timeout 15000`. The page has no sidebar; `browser network --filter shared` shows `GET /api/shared-conversations` and `/api/shared-events/search` `404`. Screenshot `--feature F25.shared-conversation-view --name local-not-found`.
- **No Public Share locally (`F25.share-publicly`).** Run `browser goto /conversations/<id>`, `control-openhands browser click 'testid=conversation-name >> testid=ellipsis-button'` and `control-openhands browser count 'testid=share-publicly-button'`: `0` on a local backend (`browser testids --filter button` lists `rename-button`, `show-skills-button`, `show-agent-tools-button`, `export-transcript-button`, `download-trajectory-button`, `display-cost-button` and `delete-button`; `show-hooks-button` and `stop-button` appear only while the conversation is active (idle, running, waiting for confirmation or finished), so a paused or errored conversation has neither). The toggle, `copy-share-link-button` and `open-share-link-button` need a Cloud backend: blocked.
@@ -122,9 +124,9 @@ Preconditions:
- The Add Backend Agent-server form has no inline "required" messages; it only disables **Connect**. The messages exist in the Edit form.
- Footer clicks from the collapsed popover or the phone drawer open the modals a moment later: `browser wait` for them instead of `count`.
- The device-flow popup survives errors: every failed attempt leaves an `about:blank` tab (#17953). Cancel closes it; after an error close it with `browser close-tab`.
-- With the default host the device flow talks to the real `app.all-hands.dev` (`POST /oauth/device/authorize`, then token polling). Starting and cancelling is harmless; never approve a code.
+- With the default host the device flow talks to the real `app.all-hands.dev` (`POST /oauth/device/authorize`, then token polling). Starting and cancelling is harmless; approve a code only in the **Cloud backend** bullet, with an account you may use.
- While the second stack's agent server is stopped or restarting, its ingress answers without CORS headers, so the browser logs `Access to fetch at '<second-url>/api/settings' ... blocked by CORS policy` (also `/server_info`, `/api/conversations/search`, `/api/llm/models/verified`). That is the dead upstream, not a CORS bug; a healthy second stack sends `access-control-allow-origin` for this origin.
-- Expected: Escape never clears the backend selector. Known failure (reproduced 2026-10-09, 2026-10-12 and 2026-10-13, #18233): Escape on the focused backend selector while its menu is closed (a second Escape after the one that closes the menu, and sometimes the Escape that closes Manage Backends opened from it) clears the selection: the closed selector shows the active backend's name as an italic placeholder (`browser value 'testid=backend-selector >> role=combobox'` is `""`) and its status dot is gone, so `browser wait 'testid=backend-selector >> role=status[name="Disconnected"]'` times out although the backend is down; a reload restores it. Reload before a closed-selector status check, as the Edit bullet does. Repro: `browser click 'testid=backend-selector'`, `browser press Escape` twice (#18233). On 2026-10-12 `bug proof` verified fix PR #18238 at 77de0ad24 against main f07aebdf5 (two `xfail`, two `xpass`, desktop). `control-openhands bug run F25.selector-dropdown` replays it (desktop, after a reload with the mouse off the selector: a reload under a mouse left on it opens its menu and empties its field, which would look like the bug) and reports `xfail` while the bug lasts:
+- Expected: Escape never clears the backend selector. Known failure (reproduced 2026-10-09, 2026-10-12, 2026-10-13 and 2026-10-14, #18233): Escape on the focused backend selector while its menu is closed (a second Escape after the one that closes the menu, and sometimes the Escape that closes Manage Backends opened from it) clears the selection: the closed selector shows the active backend's name as an italic placeholder (`browser value 'testid=backend-selector >> role=combobox'` is `""`) and its status dot is gone, so `browser wait 'testid=backend-selector >> role=status[name="Disconnected"]'` times out although the backend is down; a reload restores it. Reload before a closed-selector status check, as the Edit bullet does. Repro: `browser click 'testid=backend-selector'`, `browser press Escape` twice (#18233). On 2026-10-12 `bug proof` verified fix PR #18238 at 77de0ad24 against main f07aebdf5 (two `xfail`, two `xpass`, desktop). `control-openhands bug run F25.selector-dropdown` replays it (desktop, after a reload with the mouse off the selector: a reload under a mouse left on it opens its menu and empties its field, which would look like the bug) and reports `xfail` while the bug lasts:
```sh known-bug F25.selector-dropdown escape-closed
control-openhands browser viewport desktop
@@ -144,7 +146,7 @@ Preconditions:
- `/shared/conversations/<id>` on a local backend also raises a raw toast `HTTP request failed (404 Not Found): {"detail":"Not Found"}` next to `Conversation not found` (#17953).
- Known failure (repro candidate, not an exemption): `F25.edit-escape` closes the Manage modal underneath and drops the Edit draft (Cancel and X close only Edit, `F25.edit-cancel`) (#17953).
- The add bullet's `network --external --filter 'ids='` check guards an earlier race: on a backend's first activation from `/conversations/<id>`, about 1 attempt in 5 asked the new backend for the old conversation id. It did not happen in 15 first activations at 4eff80237: each time `QA_Second` was removed, its consent unset with `control-openhands api PATCH /api/settings --run "$OH_VERIFY_RUN_2" --write --data '{"misc_settings_diff":{"app_preferences":{"user_consents_to_analytics":null}}}'` (arrange only) and the backend added again from `/conversations/<id>`. A backend that lacks an id answers `[null]`, which now reads as "not found" (only the `This conversation does not exist…` toast), not as incompatible data. No merged change was shown to fix the race itself (#18009 only changed how a `[null]` answer reads; #18024 only moved the redirect into a shared hook, still called after `addBackend`), so a hit (an `ids=` request to `<second-url>`) is a bug to file. At 95115e84f6dd it hit on 1 of 2 first activations from `/conversations/<id>` (one `GET <second-url>/api/conversations?ids=<id>` answered `200`, with no `This conversation does not exist…` toast after it). The hit followed a failed Connect in the same modal (an empty key, `Invalid API key`) before the successful one; the clean retry, which connected once, had none, so a repro should try both sequences. On 2026-10-08 at d2c89252d it did not reproduce in 17 first activations (8 clean, 9 failed-then-clean, some under CPU load and some human-paced): every `ids=` capture was empty, and no merged change touched the activation logic since the hit, so it stays a watch item rather than a filed defect. Record the Add bullet as `fail` when the request appears, even though the redirect and the selector look right. Known failure (reproduced 2026-10-13, #18390): two independent first activations hit it that day, each the run's first activation of `QA_Second` after a clean Connect. On 2026-10-13 at 2bdc0e9d6 it hit on 1 of 6 first activations from `/conversations/<id>`: the first one of the run (`GET <second-url>/api/conversations?ids=<id>` `200`, no toast), a clean Connect in a modal opened after earlier failed Connects in other Add Backend openings; the next 5 (2 failed-then-clean, 3 clean, plus one Manage → Add) had none.
-- Known failure (reproduced 2026-10-13, #18389): Add Backend and Manage Backends neither take nor contain keyboard focus. On open, focus stays on the backend selector input behind the modal; the first Tab goes to the page behind (the Settings link, the Code/Automate toggle, the composer) and never enters the modal, and typing lands in the Home composer behind the open modal. Only after focus has been put inside the modal (for example `browser focus 'testid=add-backend-option-cloud'`) does tabbing past its last control leave for the body, the logo link and the sidebar toggle. `src/components/shared/modals/modal-backdrop.tsx` sets no initial focus, trap or restore; #18222 changes only `confirmation-modal.tsx` and does not cover these modals (same class as #18196).
+- Add Backend and Manage Backends take and keep keyboard focus, and Edit backend takes it on open (`ModalBackdrop`; at phone width open the selector with `'testid=sidebar-mobile-drawer >> testid=backend-selector'`, since the bare test id matches two selectors). At desktop width focus returns to the selector input when they close (at phone width Add Backend's X closes the drawer too and focus falls to the page body), so an Escape pressed after closing lands on the selector and can trigger #18233 below.
- Toasts are not page or HTTP errors, so `browser errors --app-only` misses them: read `browser toasts --history` after every add or switch, before anything reloads the page (the history starts with each page load).
- `F25.locked-cloud` needs `scripts/static-server.mjs --lock-to-cloud <url>`, which neither `bin/agent-canvas.mjs` nor `control-openhands launch` can pass today; even with it, reconnecting needs a Cloud account.
- The Cloud tab title truncates to `OpenHan…` at 390 px when the Agent-server tab is selected; the layout still fits. With the Providers tab present all three titles are cut to three letters.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F26-runtime-variants.md b/.agents/skills/verify-openhands/references/feature-map/F26-runtime-variants.md
index ced2b93559..f101764527 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F26-runtime-variants.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F26-runtime-variants.md
@@ -35,6 +35,7 @@ Source: `bin/agent-canvas.mjs`, `scripts/dev-with-automation.mjs`, `scripts/dev-
- `F26.host-bind-lan-optin`: the opt-in the launcher's warning recommends (`--allow-lan-session-key`) should restore key injection on a LAN bind. Known failure (reproduced 2026-10-08): the npm launcher ignores the flag and the page still carries only `__AGENT_CANVAS_AUTH_REQUIRED__`; #17949.
- `F26.host-bind-env`: `OH_BIND_HOST=0.0.0.0` (the environment form of `--host`) prints the same two "not loopback" warnings and does not inject the session key, even with `LOCAL_BACKEND_API_KEY` set.
- `F26.automation-startup-failure`: when the requested automation backend exits before it is ready, the launcher should print no success banner, stop the services it started and exit nonzero with the automation's exit reason. Known failure (reproduced 2026-10-13, #18362): it prints the **Main UI** banner and keeps the stack up with `/api/automation/health` answering `502`.
+- `F26.mars-web-bridge`: every launcher-served web stack hosts the DigitalOcean Managed Agents bridge on its ingress by default (`MARS_WEB=0` opts out): `/mars/health` answers `{"ok":true,"authRequired":true}` on a full, `--backend-only`, `--frontend-only` and `--host 0.0.0.0` launcher, `/mars/rpc/*` answers `401` without the session key and works with it, the static child behind the ingress hosts none, and the Electron stack hosts none either (its preload provides `window.marsBridge`).
- `F26.runtime-services`: `/server_info.runtime_services` lists the agent-server, ingress, frontend and automation URLs as the agent sees them, and a new conversation's system prompt carries a matching `<RUNTIME_SERVICES>` block.
- `F26.runtime-services-agent-use`: with that block alone, the agent can reach the automation backend from its terminal: asked to, it reads the OpenAPI at the URL the block names, creates an enabled cron automation, and dispatches a run that completes; the automation and its run show up in the Automations UI.
- `F26.lib-build`: `npm run build:lib` produces `dist/` where every `package.json` `exports` target exists and the main entry exports `AgentServerUIProviders`, `AgentServerUIRoot`, `CloudOrganizationBoundary` and the telemetry helpers.
@@ -83,8 +84,9 @@ Preconditions:
- **Remote backend from a normal Canvas (`F26.remote-backend`).** Needs the backend-only instance. Run `control-openhands browser goto /`, `control-openhands browser click 'testid=backend-selector'`, `control-openhands browser click 'testid=add-backend-menu-item'`, `control-openhands browser click 'testid=add-backend-option-agent-server'` and `control-openhands browser click 'testid=add-backend-location-option-remote'`. `control-openhands browser text 'testid=add-backend-agent-server-panel'` includes `Run the remote backend with --public, set a strong LOCAL_BACKEND_API_KEY, and expose it through an SSH tunnel, ngrok, or a TLS reverse proxy.` Fill `testid=add-backend-name` with `QA_Remote`, `testid=add-backend-host` with `http://127.0.0.1:18960` and `testid=add-backend-api-key` with `qa-f26-backend-key`, click `testid=add-backend-submit`, then `control-openhands browser wait 'testid=add-backend-modal' --state detached --timeout 15000`, `control-openhands browser reload` and `control-openhands browser snapshot 'testid=backend-selector'`: `status "Connected"`, `combobox "QA_Remote"` (the new backend becomes active). If you run this bullet without Cross-connect first, the backend-only state has not answered telemetry consent yet: the consent dialog (`This preference is saved for the local backend “QA_Remote” at http://127.0.0.1:18960.`) opens over the page and blocks the selector; answer it as in Cross-connect (`control-openhands browser wait 'testid=telemetry-consent-form' --timeout 10000`, `control-openhands browser uncheck 'testid=telemetry-consent-form >> role=checkbox'`, `control-openhands browser click 'testid=confirm-telemetry-preferences'`). Switch back with `control-openhands browser click 'testid=backend-selector'` and `control-openhands browser click 'role=option[name="Local"]'`, then `control-openhands browser wait 'testid=backend-selector >> role=combobox[name="Local"]' --timeout 5000` (the switch applies after a short delay: a snapshot taken right after the click still reads `combobox "QA_Remote": Local`; after the wait it shows `combobox "Local"`). Clean up: `backend-selector` → `testid=manage-backends-menu-item`; `control-openhands browser text 'testid=manage-backends-row-QA_Remote'` reads `QA_Remote`, `v<version>`, `http://127.0.0.1:18960`, `Connected`, `LOCAL`, where `<version>` is the backend-only instance's own report, `curl -s http://127.0.0.1:18960/server_info | jq -r .version` (the `versions.agentServer` pin in `config/defaults.json` unless your shell exports an `OH_AGENT_SERVER_*` override, which a hand-started launcher inherits); it passes the UI's gate: `printf '%s\n' "$(jq -r .compatibility.minimumAgentServer config/defaults.json)" "$(curl -s http://127.0.0.1:18960/server_info | jq -r .version)" | sort -VC && echo meets-minimum || echo below-minimum` prints `meets-minimum`. Then click `testid=manage-backends-remove-QA_Remote`, `testid=confirmation-modal >> testid=confirm-button`, `testid=manage-backends-done`, `browser reload`.
- **Seeded Local backend (`F26.seeded-local-backend`).** On your run, `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-backends')).map(b=>({id:b.id,name:b.name,host:b.host,kind:b.kind,hasKey:!!b.apiKey}))"` returns exactly one entry `{id: "default-local", name: "Local", host: "http://127.0.0.1:<ingress>", kind: "local", hasKey: true}` (after the Remote cleanup above), and `control-openhands browser eval "({key: Boolean(window.__AGENT_CANVAS_SESSION_API_KEY__), authRequired: window.__AGENT_CANVAS_AUTH_REQUIRED__ === true})"` is `{key: true, authRequired: false}`.
- **Rotated session key (`F26.session-key-rotated`).** Still on your run's `/`, `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-backends'))[0].apiKey === window.__AGENT_CANVAS_SESSION_API_KEY__"` is `true` (compare keys inside the page; never print one). Run `md5sum $OH_VERIFY_RUN/private/session-key | cut -c1-12`, then `control-openhands restart --rotate-key` (`rotatedKey` `true`: the launcher comes back injecting a new key) and the `md5sum` line again; the hash changed. The open page still holds the old key, so its polls get `401` until it reloads: run `control-openhands browser errors --clear`, then `control-openhands browser reload` and `control-openhands browser wait 'testid=home-screen' --timeout 20000`. `control-openhands browser count` is `1` for `testid=root-layout` and `0` for `testid=api-key-entry-screen`, `testid=agent-server-onboarding-screen` and `testid=onboarding-modal`: Home, not the key prompt public mode shows in the same state (`F01.api-key-entry`). Second view: `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-backends')).map(b=>({id:b.id,host:b.host,keyMatchesPage:b.apiKey===window.__AGENT_CANVAS_SESSION_API_KEY__}))"` is `[{id: "default-local", host: "http://127.0.0.1:<ingress>", keyMatchesPage: true}]` against the new page key, `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-agent-server-config')).sessionApiKey === window.__AGENT_CANVAS_SESSION_API_KEY__"` is `true` (the legacy key is overwritten too), `control-openhands browser network --filter 'api/settings' --last 2` shows `status` `200` under `recent`, and `control-openhands browser errors --app-only` has `pageErrors` `0` and `appErrors` `0`. `control-openhands api GET /api/settings` works with the rotated key, and `control-openhands browser screenshot --feature F26.session-key-rotated --name home-after-rotation` shows Home.
+- **Managed Agents bridge (`F26.mars-web-bridge`).** With both partial stacks up: `curl -s "$(control-openhands status | jq -r .baseUrl)/mars/health"`, `curl -s http://127.0.0.1:18960/mars/health` and `curl -s http://127.0.0.1:18970/mars/health` each print `{"ok":true,"authRequired":true}`, while `curl -s "http://127.0.0.1:$(control-openhands status | jq -r .ports.frontend)/mars/health" | head -c 15` prints `<!DOCTYPE html>` (the static child behind your run's ingress has no second bridge). `curl -s -o /dev/null -w '%{http_code}\n' -X POST http://127.0.0.1:18960/mars/rpc/getAuthState` is `401` (no key). The frontend-only page has the bridge but no key to use it: after the cross-connect, `control-openhands browser eval "({marsBridge: !!window.marsBridge, key: Boolean(window.__AGENT_CANVAS_SESSION_API_KEY__)})"` on `http://127.0.0.1:18970/` is `{marsBridge: true, key: false}` (what that does to DigitalOcean sign-in is F25's `F25.digitalocean-sign-in` known failure).
- **Stop the partial stacks.** `kill -TERM $(cat $OH_VERIFY_RUN/private/f26-frontend/pid) $(cat $OH_VERIFY_RUN/private/f26-backend/pid)`; after about 5 s the ports 18960–18963 and 18970–18973 are free again (check as in Preconditions). Kill by the saved pid only, never by pattern.
-- **LAN bind (`F26.host-bind-lan`).** Start a full stack on all interfaces: `Q=$OH_VERIFY_RUN/private/f26-lan; mkdir -p $Q; HOME=$Q/home OH_CANVAS_SAFE_STATE_DIR=$Q/state OH_CANVAS_SAFE_BACKEND_PORT=18961 OH_CANVAS_SAFE_AUTOMATION_PORT=18962 OH_CANVAS_SAFE_VITE_PORT=18963 LOCAL_BACKEND_API_KEY=qa-f26-lan-key OH_SECRET_KEY=qa-f26-secret DO_NOT_TRACK=1 nohup node bin/agent-canvas.mjs --host 0.0.0.0 --port 18960 > $Q/launcher.log 2>&1 & echo $! > $Q/pid`, and wait until `curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:18960/` is `200`. `grep -a 'not loopback' $Q/launcher.log` shows `[auth] Bind host 0.0.0.0 is not loopback — session key will not be injected into HTML` and `[static] WARNING: bind host 0.0.0.0 is not loopback; refusing to inject the session API key into HTML.` The served page proves it: `curl -s http://127.0.0.1:18960/ | grep -o '__AGENT_CANVAS_\(SESSION_API_KEY\|AUTH_REQUIRED\)__' | sort -u` prints only `__AGENT_CANVAS_AUTH_REQUIRED__` and `curl -s http://127.0.0.1:18960/ | grep -c qa-f26-lan-key` prints `0`, while your loopback run's page, `curl -s "$(control-openhands status | jq -r .baseUrl)/" | grep -o '__AGENT_CANVAS_\(SESSION_API_KEY\|AUTH_REQUIRED\)__' | sort -u`, prints only `__AGENT_CANVAS_SESSION_API_KEY__`. The bind is real: with `LAN` from the loopback bullet, `curl -s -o /dev/null -w '%{http_code}\n' --connect-timeout 3 http://$LAN:18960/` is `200` here (refused on your run, `F26.loopback-bind-default`; with an empty `LAN` this check is `not-run`). The browser part below expects a profile that has never visited `127.0.0.1:18960` (true in this file's order; on a repeat pass run `control-openhands browser reset` first and restore your run afterwards, see Gotchas). Then `control-openhands browser goto http://127.0.0.1:18960/ --allow-external` and `control-openhands browser eval "({key: Boolean(window.__AGENT_CANVAS_SESSION_API_KEY__), authRequired: window.__AGENT_CANVAS_AUTH_REQUIRED__ === true})"` → `{key: false, authRequired: true}`. The onboarding shows `testid=onboarding-step-check-backend` with `control-openhands browser enabled 'testid=onboarding-backend-next'` `false` (no key typed). `control-openhands browser click 'testid=onboarding-skip'` and `control-openhands browser wait 'testid=api-key-entry-screen' --timeout 15000` succeed; after `sleep 12`, `control-openhands browser toasts --history` is `[]` (no error toast on either screen); `control-openhands browser screenshot --feature F26.host-bind-lan --name api-key-screen`. Fill `testid=api-key-entry-name` with `QA_LAN` and `testid=api-key-entry-api-key` with `qa-f26-lan-key`, click `testid=api-key-entry-submit`; `control-openhands browser wait 'testid=root-layout' --timeout 10000` succeeds. Stop it with `kill -TERM $(cat $Q/pid)`.
+- **LAN bind (`F26.host-bind-lan`).** Start a full stack on all interfaces: `Q=$OH_VERIFY_RUN/private/f26-lan; mkdir -p $Q; HOME=$Q/home OH_CANVAS_SAFE_STATE_DIR=$Q/state OH_CANVAS_SAFE_BACKEND_PORT=18961 OH_CANVAS_SAFE_AUTOMATION_PORT=18962 OH_CANVAS_SAFE_VITE_PORT=18963 LOCAL_BACKEND_API_KEY=qa-f26-lan-key OH_SECRET_KEY=qa-f26-secret DO_NOT_TRACK=1 nohup node bin/agent-canvas.mjs --host 0.0.0.0 --port 18960 > $Q/launcher.log 2>&1 & echo $! > $Q/pid`, and wait until `curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:18960/` is `200`. `grep -a 'not loopback' $Q/launcher.log` shows `[auth] Bind host 0.0.0.0 is not loopback — session key will not be injected into HTML` and `[static] WARNING: bind host 0.0.0.0 is not loopback; refusing to inject the session API key into HTML.` The served page proves it: `curl -s http://127.0.0.1:18960/ | grep -o '__AGENT_CANVAS_\(SESSION_API_KEY\|AUTH_REQUIRED\)__' | sort -u` prints only `__AGENT_CANVAS_AUTH_REQUIRED__` and `curl -s http://127.0.0.1:18960/ | grep -c qa-f26-lan-key` prints `0`, while your loopback run's page, `curl -s "$(control-openhands status | jq -r .baseUrl)/" | grep -o '__AGENT_CANVAS_\(SESSION_API_KEY\|AUTH_REQUIRED\)__' | sort -u`, prints only `__AGENT_CANVAS_SESSION_API_KEY__`. The bind is real: with `LAN` from the loopback bullet, `curl -s -o /dev/null -w '%{http_code}\n' --connect-timeout 3 http://$LAN:18960/` is `200` here (refused on your run, `F26.loopback-bind-default`; with an empty `LAN` this check is `not-run`). The browser part below expects a profile that has never visited `127.0.0.1:18960` (true in this file's order; on a repeat pass run `control-openhands browser reset` first and restore your run afterwards, see Gotchas). Then `control-openhands browser goto http://127.0.0.1:18960/ --allow-external` and `control-openhands browser eval "({key: Boolean(window.__AGENT_CANVAS_SESSION_API_KEY__), authRequired: window.__AGENT_CANVAS_AUTH_REQUIRED__ === true})"` → `{key: false, authRequired: true}`. The onboarding shows `testid=onboarding-step-check-backend` with `control-openhands browser enabled 'testid=onboarding-backend-next'` `false` (no key typed). `control-openhands browser click 'testid=onboarding-skip'` and `control-openhands browser wait 'testid=api-key-entry-screen' --timeout 15000` succeed; after `sleep 12`, `control-openhands browser toasts --history` is `[]` (no error toast on either screen); the bridge is mounted behind the key (`F26.mars-web-bridge`): `curl -s http://127.0.0.1:18960/mars/health` is `{"ok":true,"authRequired":true}`, and `POST /mars/rpc/getAuthState` with `-H 'Content-Type: application/json' -d '{"args":[]}'` is `401` without and `200` with `-H 'X-Session-API-Key: qa-f26-lan-key'`; `control-openhands browser screenshot --feature F26.host-bind-lan --name api-key-screen`. Fill `testid=api-key-entry-name` with `QA_LAN` and `testid=api-key-entry-api-key` with `qa-f26-lan-key`, click `testid=api-key-entry-submit`; `control-openhands browser wait 'testid=root-layout' --timeout 10000` succeeds. A few seconds later the telemetry consent dialog for the new `QA_LAN` backend opens and blocks clicks; answer it (`role=button[name="Confirm preferences"]`) before driving that page further. Stop it with `kill -TERM $(cat $Q/pid)`.
- **LAN opt-in (`F26.host-bind-lan-optin`).** Expected: the opt-in the warning names (`Pass --host 127.0.0.1 (default) for local mode, or --allow-lan-session-key only if you accept LAN exposure.`) makes the npm launcher inject the key again. Start the LAN stack as above with `--host 0.0.0.0 --allow-lan-session-key --port 18960` (log to `$Q/launcher2.log`), wait for `200`, then `curl -s http://127.0.0.1:18960/ | grep -o '__AGENT_CANVAS_\(SESSION_API_KEY\|AUTH_REQUIRED\)__' | sort -u`. Today it prints only `__AGENT_CANVAS_AUTH_REQUIRED__` and the log repeats both "not loopback" warnings: the launcher ignores the flag (fail; see Gotchas). Stop it with `kill -TERM $(cat $Q/pid)`.
- **Generated session key persists (`F26.session-key-persist`).** With 18960–18963 free, define a helper and start a loopback stack without a key: `Q=$OH_VERIFY_RUN/private/f26-keys; mkdir -p $Q; start(){ HOME=$Q/home OH_CANVAS_SAFE_STATE_DIR=$Q/state OH_CANVAS_SAFE_BACKEND_PORT=18961 OH_CANVAS_SAFE_AUTOMATION_PORT=18962 OH_CANVAS_SAFE_VITE_PORT=18963 OH_SECRET_KEY=qa-f26-secret DO_NOT_TRACK=1 "$@" > $Q/launcher.log 2>&1 & echo $! > $Q/pid; for i in $(seq 1 60); do c=$(curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:18960/); [ "$c" = 200 ] && break; sleep 3; done; echo "code=$c"; }; start env -u LOCAL_BACKEND_API_KEY nohup node bin/agent-canvas.mjs --port 18960` (`code=200`). `curl -s http://127.0.0.1:18960/ | grep -o '__AGENT_CANVAS_SESSION_API_KEY__[^;<]*' | md5sum` prints a hash (never print the key itself); `$Q/home/.openhands/agent-canvas/api-key.txt` exists (64 hex characters) and `curl -s -o /dev/null -w '%{http_code}' -H "X-Session-API-Key: $(cat $Q/home/.openhands/agent-canvas/api-key.txt)" http://127.0.0.1:18960/api/settings` is `200`. `kill -TERM $(cat $Q/pid); sleep 6`, run the same `start ...` line again in the same shell and repeat the `md5sum` line: the hash is identical.
- **Pinned session key (`F26.session-key-pinned`).** Same shell, after `kill -TERM $(cat $Q/pid); sleep 6`: `start env LOCAL_BACKEND_API_KEY=qa-f26-pin-key nohup node bin/agent-canvas.mjs --port 18960`. `curl -s http://127.0.0.1:18960/ | grep -o '__AGENT_CANVAS_SESSION_API_KEY__[^;<]*'` prints `__AGENT_CANVAS_SESSION_API_KEY__="qa-f26-pin-key"`; `/api/settings` is `200` with `-H 'X-Session-API-Key: qa-f26-pin-key'` and `401` with the saved generated key. In the browser, `control-openhands browser goto http://127.0.0.1:18960/ --allow-external` and `control-openhands browser eval "({key: window.__AGENT_CANVAS_SESSION_API_KEY__ === 'qa-f26-pin-key', authRequired: window.__AGENT_CANVAS_AUTH_REQUIRED__ === true})"` → `{key: true, authRequired: false}`.
@@ -97,7 +99,7 @@ Preconditions:
- **Library build (`F26.lib-build`).** Never build inside the shared checkout (it regenerates i18n and typegen files the other runs hash); build in a copy: `L=$OH_VERIFY_RUN/private/f26-lib; mkdir -p $L; tar --exclude=./node_modules --exclude=./build --exclude=./.git --exclude=./dist -cf - . | tar -xf - -C $L; ln -s "$PWD/node_modules" $L/node_modules; (cd $L && npm run build:lib); echo "exit=$?"`. It ends with `✓ built in` and `exit=0` (about 30 s). Then `(cd $L && node -e 'const p=require("./package.json"),fs=require("fs");let ok=0,miss=[];for(const[k,v]of Object.entries(p.exports))for(const f of(typeof v==="string"?[v]:Object.values(v)))fs.existsSync(f)?ok++:miss.push(k+" -> "+f);console.log(JSON.stringify({entries:Object.keys(p.exports),ok,missing:miss}))')` prints 9 entries (`.`, `./browser`, `./conversation`, `./files`, `./settings`, `./sidebar`, `./terminal`, `./i18n`, `./package.json`), `ok` 25 and `missing` `[]`. `grep -ohE 'AgentServerUIProviders|AgentServerUIRoot|CloudOrganizationBoundary|configureTelemetry|setTelemetryConsent' $L/dist/lib/index.d.ts | sort -u` lists all five. Remove the copy with `rm -rf $L`.
- **Style scope (`F26.lib-style-scope`).** On your run's home page (`control-openhands browser goto /`) run `control-openhands browser eval "(() => { let total=0, unscoped=[]; for (const sh of document.styleSheets) { let rules; try { rules = sh.cssRules } catch { continue } const walk = (rs) => { for (const r of rs) { if (r.selectorText) { total++; if (!r.selectorText.includes('data-agent-server-ui')) unscoped.push(r.selectorText) } else if (r.cssRules) walk(r.cssRules) } }; walk(rules) } return { scopeRoots: document.querySelectorAll('[data-agent-server-ui]').length, total, unscopedCount: unscoped.length, sample: unscoped.slice(0, 12) } })()"`. `total` is in the thousands and `unscopedCount` is `2`: only `.go<hash>` and `.go<hash> > *`, react-hot-toast's runtime classes. `control-openhands browser eval "getComputedStyle(document.querySelector('[data-agent-server-ui]')).getPropertyValue('--oh-color-base').trim()"` is `#181818`. This proves the bundle is scoped, not that a host page is unaffected (`F26.lib-host-app`).
- **Blocked (`F26.lib-host-app`, `F26.docker-image`, `F26.docker-conversation-runtime`, `F26.helm-chart`).** If `docker info` cannot reach the daemon but `dockerd` is installed, start it (`setsid dockerd`); without one, record both Docker rows `blocked` naming the missing daemon. With a daemon, run the image with `-p 127.0.0.1:8000:8000`, open `http://localhost:8000/canvas` and expect the **Add a backend** step (`testid=onboarding-step-check-backend`, Next disabled until a key is typed); `onboarding-skip` leads to `testid=api-key-entry-screen`. With `-e AGENT_CANVAS_ALLOW_LAN_SESSION_KEY=true` the key is injected and first run opens on Choose your agent, under the telemetry consent dialog. `npm run build:docker` behind a TLS-intercepting proxy needs the proxy CA inside the build stages (without it `uv pip install` / `pip install` in the final stage fail with `CERTIFICATE_VERIFY_FAILED: self-signed certificate in certificate chain`), and Docker Hub may answer 429. Build a scratch copy of the Dockerfile that only adds the CA, never an edited `docker/Dockerfile`: after `FROM node:24-slim AS frontend-build` add `COPY --from=ccrca ca-bundle.crt /tmp/qa-ca.crt` and `ENV NODE_EXTRA_CA_CERTS=/tmp/qa-ca.crt npm_config_cafile=/tmp/qa-ca.crt`, after the first `USER root` add the same `COPY` and `ENV SSL_CERT_FILE=/tmp/qa-ca.crt PIP_CERT=/tmp/qa-ca.crt REQUESTS_CA_BUNDLE=/tmp/qa-ca.crt UV_NATIVE_TLS=1`, then `docker build -f <copy> --build-context ccrca=<dir holding ca-bundle.crt> --network host --build-arg AGENT_SERVER_IMAGE=<images.agentServer>:<versions.agentServer>-python --build-arg AGENT_SERVER_VERSION=<versions.agentServer> --build-arg AUTOMATION_VERSION=<versions.automation> --build-arg VITE_BASE_PATH=/canvas -t qa-agent-canvas:target .` (the pins `scripts/docker-build.mjs` reads from `config/defaults.json`; about 5 min on a warm cache). Publish it on a free port (`-p 127.0.0.1:<free>:8000`; 8000 and the 188xx blocks may belong to other runs). Docker Hub may answer `429 Too Many Requests` for `node:24-slim`; in the scratch copy only, replace both `FROM node:24-slim` lines with `FROM mirror.gcr.io/library/node:24-slim` (arrange; say so in the note). Run the unmodified image with `docker run -d --name <qa-name> -p 127.0.0.1:<free>:8000 qa-agent-canvas:target`. On a kernel without IPv6, `docker logs <qa-name>` shows `IPv6 is not available; binding the frontend to 0.0.0.0.` and `Static-server + proxy listening on http://0.0.0.0:8000/canvas/`, `/canvas/` answers `200` within about 25 s and `docker inspect -f '{{.State.Status}}' <qa-name>` is still `running` minutes later (#18178 fixed by #18180, verified 2026-10-13); `docker exec <qa-name> ls /opt/agent-canvas` lists the `mars-*.mjs` bridge modules (#18343). The container's generated key is `docker exec <qa-name> cat /home/openhands/.openhands/agent-canvas/api-key.txt` (write it to `$OH_VERIFY_RUN/private/`, fill with `--value-file`). Settings → Application (`/canvas/settings/app`) shows the image's `Version` (`agent-canvas-update-card`), which matches `jq -r .version package.json`. Remove the containers and the image afterwards (`docker rm -f`, `docker rmi`). Record `F26.lib-host-app` `blocked` naming the missing host-app example. `command -v helm kubectl` prints nothing here: record `F26.helm-chart` `blocked` naming the missing `helm` binary and cluster; with them, `helm lint helm/agent-canvas` and `helm template qa helm/agent-canvas` render the StatefulSet, Service and Ingress, and `helm install` follows `helm/agent-canvas/README.md`.
-- **Desktop splash and window (`F26.desktop-boot-splash`, `F26.desktop-main-window`).** Start a private display and the app in one process group: `Q=$OH_VERIFY_RUN/private/f26-desktop; mkdir -p $Q/shots; setsid nohup sh -c "echo \$\$ > $Q/pgid; Xvfb :78 -screen 0 1440x1000x24 -nolisten tcp & sleep 1; DISPLAY=:78 HOME=$Q/home OH_CANVAS_SAFE_STATE_DIR=$Q/state LOCAL_BACKEND_API_KEY=qa-f26-desktop-key OH_SECRET_KEY=qa-f26-secret DO_NOT_TRACK=1 exec node_modules/.bin/electron --no-sandbox electron" > $Q/desktop.log 2>&1 &`, then capture frames: `for i in $(seq 1 40); do sleep 1; DISPLAY=:78 import -window root $Q/shots/s$(printf %02d $i).png; done`. Read the frames: the first seconds show the splash (logo, `OpenHands Agent Canvas`, `AI coding agent interface`, spinner, `Starting backend services…`, then lines such as `agent-server: {"asctime": ...`, the note `First launch downloads Python + the OpenHands agent server.` and a **Show details** button); a later frame shows the main window with a `File Edit View Window` menu and the first-run **Choose your agent** step under the consent dialog naming `the local backend "Local" at http://localhost:8000` (about 30 s on a warm cache, minutes on the first download). Copy the frames you cite into `$OH_VERIFY_RUN/evidence/F26.desktop-boot-splash/` and `.../F26.desktop-main-window/`. Stop in two steps: `kill -TERM $(cat $Q/pgid)` signals only the Electron launcher (it quits and stops its backend services); after about 10 s ports 8000, 18000, 18001 and 3001 are free again. Then `kill -TERM -- -$(cat $Q/pgid)` ends Xvfb; `ps -o pid,cmd -g $(cat $Q/pgid)` lists nothing.
+- **Desktop splash and window (`F26.desktop-boot-splash`, `F26.desktop-main-window`).** Start a private display and the app in one process group: `Q=$OH_VERIFY_RUN/private/f26-desktop; mkdir -p $Q/shots; setsid nohup sh -c "echo \$\$ > $Q/pgid; Xvfb :78 -screen 0 1440x1000x24 -nolisten tcp & sleep 1; DISPLAY=:78 HOME=$Q/home OH_CANVAS_SAFE_STATE_DIR=$Q/state LOCAL_BACKEND_API_KEY=qa-f26-desktop-key OH_SECRET_KEY=qa-f26-secret DO_NOT_TRACK=1 exec node_modules/.bin/electron --no-sandbox electron" > $Q/desktop.log 2>&1 &`, then capture frames: `for i in $(seq 1 40); do sleep 1; DISPLAY=:78 import -window root $Q/shots/s$(printf %02d $i).png; done`. Read the frames: the first seconds show the splash (logo, `OpenHands Agent Canvas`, `AI coding agent interface`, spinner, `Starting backend services…`, then lines such as `agent-server: {"asctime": ...`, the note `First launch downloads Python + the OpenHands agent server.` and a **Show details** button); a later frame shows the main window with a `File Edit View Window` menu and the first-run **Choose your agent** step under the consent dialog naming `the local backend "Local" at http://localhost:8000` (about 30 s on a warm cache, minutes on the first download). While it runs, `curl -s http://127.0.0.1:8000/mars/health | head -c 15` prints `<!DOCTYPE html>`: the desktop stack's ingress hosts no bridge (`marsWebBridge: false`, `F26.mars-web-bridge`). Copy the frames you cite into `$OH_VERIFY_RUN/evidence/F26.desktop-boot-splash/` and `.../F26.desktop-main-window/`. Stop in two steps: `kill -TERM $(cat $Q/pgid)` signals only the Electron launcher (it quits and stops its backend services); after about 10 s ports 8000, 18000, 18001 and 3001 are free again. Then `kill -TERM -- -$(cat $Q/pgid)` ends Xvfb; `ps -o pid,cmd -g $(cat $Q/pgid)` lists nothing.
- **External links (`F26.desktop-external-links`).** Not driven: record `not-run` (harness gap: no verb clicks inside the Electron window or observes the OS browser handoff).
## Gotchas
diff --git a/.agents/skills/verify-openhands/references/feature-map/F27-workspace-tools.md b/.agents/skills/verify-openhands/references/feature-map/F27-workspace-tools.md
index cf64050762..c001b10b4f 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F27-workspace-tools.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F27-workspace-tools.md
@@ -105,6 +105,6 @@ Preconditions:
- At phone width (and in a narrow drawer) the tab row shows only the tabs that fit; the rest are reachable only through the tabs menu (`conversation-tabs-menu-open-<tab>`), and which tabs fit depends on the active tab's label width.
- Plain `agent-canvas` instances on one machine share the tmux socket `/tmp/tmux-<uid>/openhands` (OpenHands/OpenHands#17946), so another instance stopping can reset this one's terminal. `control-openhands launch` gives each run its own `TMUX_TMPDIR`, so in verification runs a reset means this run's agent server restarted.
- Agent Tools & Metadata remembers the last tab (it reopens on Available Tools).
-- The overview drawer's Automations and MCP sections have no opener: the Overview menu opens only Pull Requests (`conversation-overview-menu-open-git-pull_requests`). Their in-range changes (Add automation now launches `/automations/setup` instead of a modal, #17688; the MCP enable toggle shows its own error toast, #18182) cannot be driven from the UI (checked 2026-10-13).
+- The Overview menu opens only Pull Requests (`conversation-overview-menu-open-git-pull_requests`); its testids are `conversation-overview-menu-open-*` and `-pin-*` for `workspace` and the Git rows only.
- Known open issues: terminal history after reload (#17566) still fails (reproduced 2026-10-06); info modals at phone width (#17562) pass.
- Agent tab requests (`open_tab`, `navigate_to_file`, chat file links) are saved per conversation like a user tab click (#18045): after the agent opens Terminal or a file and the page reloads, the drawer starts closed and the panel toggle reopens it on that tab; clicking that tab then closes the drawer. At phone width they navigate to `/conversations/<id>/panel`.
diff --git a/.agents/skills/verify-openhands/references/feature-map/README.md b/.agents/skills/verify-openhands/references/feature-map/README.md
index 03e0aa03db..0782a0f699 100644
--- a/.agents/skills/verify-openhands/references/feature-map/README.md
+++ b/.agents/skills/verify-openhands/references/feature-map/README.md
@@ -5,7 +5,7 @@ of Agent Canvas. Read this index before driving the app, then open the matching
feature file and run its recipe with `control-openhands`. The map is not a claim
that every row passes today; the run's evidence ledger is.
-Maintenance baseline: main@2bdc0e9d6c5940bc3d4667fb2127fbb753b2f7c8 (2026-10-13). The next maintenance pass starts from this commit; a pass proposes the next baseline in its PR, and merging that PR accepts it.
+Maintenance baseline: main@8b9823a89ce5844d0808398c2f740d9748da66df (2026-10-14). The next maintenance pass starts from this commit; a pass proposes the next baseline in its PR, and merging that PR accepts it.
## Baseline preconditions
@@ -230,7 +230,7 @@ How to write and prove new entries: [../mapping.md](../mapping.md).
| ID | Family | What it covers | Entry points | Needs | Sub-features |
|---|---|---|---|---|---|
| F01 | [First run, onboarding and sign-in](F01-first-run-and-sign-in.md) | telemetry consent, onboarding modal, public-mode backend step and API-key screen, route error page | any URL on a fresh browser profile; `launch --public` | fresh run or `browser reset`; LLM for say-hello | 22 |
-| F02 | [App shell, sidebar and command menu](F02-app-shell.md) | rail and phone drawer, collapse, home pinning, getting-started checklist, command menu, toasts | every page; `Control+k` / `Meta+k` | baseline; LLM for checklist progress | 27 |
+| F02 | [App shell, sidebar and command menu](F02-app-shell.md) | rail and phone drawer, collapse, home pinning, getting-started checklist, command menu, toasts | every page; `Control+k` / `Meta+k` | baseline; LLM for checklist progress | 28 |
| F03 | [Home and starting work](F03-home.md) | home composer, workspace picker, plugin picker, recommended automations rail, no-LLM banner | `/`, **New Chat**, command menu | runs before and after `llm preset`; fixture repo | 42 |
| F04 | [Conversation list and folders](F04-conversation-list.md) | sidebar list, View presets, workspace folders, tags, rename/pin/archive/delete, load more | sidebar on every page | several conversations; fixture repos | 41 |
| F05 | [Composer, slash commands and plan mode](F05-composer.md) | send, drafts, attachments, model pill, slash commands, `/goal`, plan mode, dictation | home and conversation composer | LLM | 25 |
@@ -242,7 +242,7 @@ How to write and prove new entries: [../mapping.md](../mapping.md).
| F11 | [Provider connections](F11-provider-connections.md) | add, edit, rotate key, delete connections; profiles from a connection | `/settings/llm` | fresh run; DeepSeek key | 21 |
| F12 | [Model router](F12-model-router.md) | meta-profiles, templates, run on first message, routing effect | `/settings/meta-llm` | LLM profiles | 25 |
| F13 | [Agent profiles](F13-agent-profiles.md) | OpenHands and ACP profiles, editor, name rules, active profile, MCP and secret scoping | `/settings/agents` | fresh run; LLM | 22 |
-| F14 | [Secrets](F14-secrets.md) | list, add, edit, rename, delete, agent access | `/settings/secrets` | LLM for agent access | 11 |
+| F14 | [Secrets](F14-secrets.md) | list, add, edit, rename, delete, agent access | `/settings/secrets` | LLM for agent access | 12 |
| F15 | [Condenser, agent context and verification](F15-agent-behavior-settings.md) | condenser fields, agent context, confirmation mode and critic, their effect on runs | `/settings/condenser`, `/agent-context`, `/verification` | LLM for the effects | 23 |
| F16 | [Application settings](F16-application-settings.md) | language, theme, analytics, sound, checklist, title model, voice input, git identity | `/settings/app` | LLM for title and git checks | 20 |
| F17 | [Customize hub and MCP servers](F17-mcp-servers.md) | catalog, install, test, edit, delete, custom servers, agent use | **Customize**, `/mcp` | `uvx`/`npx`; LLM | 31 |
@@ -254,11 +254,11 @@ How to write and prove new entries: [../mapping.md](../mapping.md).
| F23 | [Automation detail, runs and editing](F23-automation-detail.md) | detail page, runs, logs, edit dialog, triggers, debug | `/automations/<id>` | LLM; runs | 41 |
| F24 | [Automation Git Sync](F24-git-sync.md) | configure, sync cycles, encryption, status | `/automations/git-sync` | `fixture git-repo` | 26 |
| F25 | [Backends, Cloud and sharing](F25-backends-and-cloud.md) | add, edit, remove and switch backends, per-backend consent, Cloud login, shared pages | backend selector | a second stack; Cloud account (blocked) | 33 |
-| F26 | [Launcher modes, Docker, desktop and library](F26-runtime-variants.md) | launcher flags, partial stacks, LAN bind, Docker, Electron, embeddable library | a terminal | Docker/Electron where available | 33 |
+| F26 | [Launcher modes, Docker, desktop and library](F26-runtime-variants.md) | launcher flags, partial stacks, LAN bind, Docker, Electron, embeddable library | a terminal | Docker/Electron where available | 34 |
| F27 | [Workspace tools](F27-workspace-tools.md) | terminal, browser, planner, task list, usage, `canvas_ui_control` | drawer tabs | LLM | 30 |
| F28 | [Automation setup page](F28-automation-setup-page.md) | blank form, kind toggle, schedule presets, event fields, timeout, draft save and validation, draft Test, Create | **Add Automation → Create Automation**, `/automations/setup`, draft Resume, kebab Edit | automation service; no LLM | 13 |
-28 families, 775 sub-features. `control-openhands map ids` lists every ID with its file.
+28 families, 778 sub-features. `control-openhands map ids` lists every ID with its file.
### Neighbouring families
diff --git a/.agents/skills/verify-openhands/references/mapping.md b/.agents/skills/verify-openhands/references/mapping.md
index 57d5c3939d..0bc6fbd675 100644
--- a/.agents/skills/verify-openhands/references/mapping.md
+++ b/.agents/skills/verify-openhands/references/mapping.md
@@ -83,6 +83,14 @@ that ID instead of re-mapping it.
## 3. Write each entry
+Describe current behavior and how to verify it. Once a fix is proven, keep
+the corrected expectations and normal recipe steps, and remove the fix's
+PR or closed-issue references, verification dates or device lists, old UI
+wording, and descriptions of checkouts without the fix. Keep that history
+in the maintenance PR and evidence ledger. Retain links for known open
+bugs, authorizing PRs for retired IDs, and version or environment
+distinctions that still affect a supported run.
+
Each file starts with an H1 title, one paragraph describing the user-visible
behavior, and one `Source:` line with the main implementation paths (for drift
checks and `map affected`; keep other implementation detail out). Every path