Skip to content

🙌 OpenHands: AI-Driven Development

docs(verify-openhands): map maintenance 2026-10-09 (#18237)

Engel Nyst95dcdcdf12bd
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 46e1e40e31..f677b77bd4 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
@@ -36,6 +36,7 @@ Source: `src/components/features/sidebar/`, `src/components/features/command-men
- `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).
+- `F02.super-admin-setup-tour`: for the two setup steps done in Canvas (Choose an automation template, Add an integration), the guide walks the Super Admin through the page with a guided tour (`.oh-setup-tour-popover` with Back / Next / Done): the template step points at the recommended-automations section, then the setup dialog; the integration step at **Add custom server** on `/mcp`, then the server form. Finishing the guide's next step in Canvas opens the step after it the way **Start** does (a Canvas step's tour or page; an enterprise step reopens the guide to point at Start, or navigates on a locked-Cloud canvas). An enterprise page that links back with `?setup_tour=<step id>` starts that step's tour once and drops the parameter. Local backends ignore `?setup_tour=` (no tour, no guide, the parameter stays) (#18138). The positive path is blocked here, with the same Enterprise prerequisite as the setup guide above.
- `F02.update-tile`: when npm `latest` is newer than the running version, a **New version** tile (`agent-canvas-version-tile`) opens an update modal with the npm and Docker commands. The tile is never shown in locked-Cloud mode.
- `F02.alert-banner`: when the server config reports maintenance, faulty models or an error message, a dismissible banner (`alert-banner`) appears at the top of the content area.
- `F02.settings-404-modal`: if the backend's settings return 404, the LLM settings modal opens by itself. Other settings fetch errors show a toast.
@@ -84,6 +85,7 @@ Preconditions:
- **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.
- **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.
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 f6401db625..c1adb440f0 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F03-home.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F03-home.md
@@ -49,6 +49,7 @@ Automations on home:
- `F03.home-automations-list`: **Automations** lists enabled automations with schedule, last run and status, a sparkline, **+ Add**, and `No automations yet. Add one to automate repetitive work.` when empty.
- `F03.home-automation-links`: a row opens `/automations/<id>`, a sparkline bar opens `/automations/<id>?run=<runId>`, **+ Add** opens `/automations`.
- `F03.home-run-tooltip`: hovering a row shows its trigger, status, last run and task.
+- `F03.home-run-outcome-health`: an automation whose latest Completed run reported a task outcome shows it on Home (#17682): `blocked` and `partial_success` read **Blocked** and **Partial**, `unknown` or an unrecognised status reads **Needs review**, each with an amber (`bg-warning`) health dot, in the row text, the row link's aria-label, the tooltip's Status line and the pinned card's strip. A finish response without a string status counts as success.
- `F03.home-row-menu`: the row kebab offers Run now, View, Edit, Turn off and Pin to dashboard; Run now toasts `Automation dispatched` and is disabled while a run is in flight; Edit saves with `Automation updated`.
- `F03.cancel-in-flight-run`: **Cancel run** appears while a run is pending or running and cancels it (`Automation run cancelled`).
- `F03.turn-off-confirmation`: Turn off asks `Turn off "<name>"? It will stop running on its schedule until you turn it back on.`; Cancel keeps it on, Turn off disables it and drops it from home.
@@ -115,6 +116,7 @@ Automations on home (fixtures imported and turned on):
- **Activity list (`F03.home-automations-list`, `F03.home-automation-links`).** `control-openhands browser wait '[data-testid^="running-automation-row-"]'`, `control-openhands browser text 'testid=running-automations-list'`: both fixtures with `0 0 1 1 *` and `No activity yet` (`browser screenshot --feature F03.home-automations-list --name list`); no row has a sparkline link yet (the run-bar link is checked after the first run, in the next bullet). `control-openhands browser click '[data-testid^="running-automation-row-"] >> has-text=QA_F03 Beta >> role=link' --expect-url '/automations/[^/?]+'` opens the detail page; after `control-openhands browser goto /`, `control-openhands browser click 'testid=home-automations-manage' --expect-url '/automations$'` opens the dashboard.
- **Row menu, Run now and Cancel (`F03.home-row-menu`, `F03.cancel-in-flight-run`, `F03.home-automation-links`).** Run `control-openhands browser goto /` (the Manage link left the dashboard), then `control-openhands browser click '[data-testid^="running-automation-row-"] >> has-text=QA_F03 Pong >> [data-testid^="running-automation-menu-"]'` and `control-openhands browser snapshot '[data-testid^="running-automation-menu-panel-"]'`: Run now, View, Edit, Turn off, separator, Pin to dashboard. `control-openhands browser click '[data-testid^="running-automation-run-"]' --observe 'role=status'` observes `Automation dispatched`. Reopen the menu: `Run now [disabled]` and **Cancel run**. `control-openhands browser click '[data-testid^="running-automation-cancel-"]'` and `control-openhands browser toasts` reads `Automation run cancelled`; `control-openhands api GET /api/automation/v1/<automation-id>/runs` (id from `browser testids --filter running-automation-row`) shows `CANCELLED`, and after `browser reload` the row reads `Just now · Cancelled`. Pong now has a run: `control-openhands browser click '[data-testid^="running-automation-activity-"] >> role=link' --expect-url '/automations/[^/?]+\?run='` opens it; return with `control-openhands browser goto /`.
- **Tooltip (`F03.home-run-tooltip`).** `control-openhands browser tooltip '[data-testid^="running-automation-row-"] >> has-text=QA_F03 Pong'` returns `QA_F03 Pong`, `Trigger 0 0 1 1 *`, `Status Cancelled`, `Last run Just now`, `Task Cancelled by user`.
+- **Run outcomes (`F03.home-run-outcome-health`).** Arrange as `F23.run-task-outcome` does (API `preset/prompt`, yearly cron `0 0 1 1 *`, enabled) `QA_F03 Blocked`, which finishes with status `blocked`, and `QA_F03 Review`, which finishes with status `unknown`, each prompt ending `Do not run any tools.` On `/`, for each row open `'[data-testid^="running-automation-row-"] >> has-text=QA_F03 Blocked >> [data-testid^="running-automation-menu-"]'` and click `[data-testid^="running-automation-run-"]` (`Automation dispatched`); poll `control-openhands api GET /api/automation/v1/<id>/runs` until the run is `COMPLETED` (about 35 s), then `control-openhands browser reload`. `control-openhands browser text 'testid=running-automations-list'` shows `… · Blocked` and `… · Needs review`; `control-openhands browser eval "[...document.querySelectorAll('[data-testid^=running-automation-row-] a[aria-label]')].filter(a=>a.querySelector('.rounded-full')).map(a=>a.getAttribute('aria-label')+' | '+a.querySelector('.rounded-full').className.match(/bg-[a-z-]+/g)).join('; ')"` lists `QA_F03 Blocked Blocked | bg-warning` and `QA_F03 Review Needs review | bg-warning` (in list order, which can vary). `control-openhands browser tooltip` on each row reads `Status Blocked` / `Task <its summary>` and `Status Needs review` / `Task <its summary>` (before each tooltip read, `browser hover 'testid=chat-input'` and `browser wait 'role=tooltip' --state detached --timeout 3000`, or the previous row's tooltip is returned). Pin both from their row menus (`[data-testid^="running-automation-pin-"]`): each card's strip shows the Blocked (!) (`run-status-icon-warning`) or Needs review (?) (`run-status-icon-needs-review`) icon, then the run's summary and its conversation title. Expected: the summary is visible (`browser eval "[...document.querySelectorAll('[data-testid^=pinned-automation-card-] [data-testid=run-status-badge-icon]')].map(i=>Math.round(i.nextElementSibling.getBoundingClientRect().width))"` is above `0`). Known failure (reproduced 2026-10-09, #18234): when the run has a conversation, the summary is 0 px wide at desktop and phone, so only the conversation title shows (the summary is still in `browser text`). Repeat at `browser viewport phone`. The status-less success case has no live producer (automation 1.19.0 with SDK 1.53.0 always sends a status), as `F23.run-task-outcome` records: `blocked`. Delete both automations afterwards.
- **View and Edit (`F03.home-row-menu`).** From the Beta row menu, `control-openhands browser click '[data-testid^="running-automation-view-"]' --expect-url '/automations/[^/?]+$'` opens the detail page. Back on `/` (`control-openhands browser goto /`), reopen the menu and click `[data-testid^="running-automation-edit-"]`: the **Edit automation** modal shows `browser value 'label=Name'` `QA_F03 Beta`. `control-openhands browser fill 'label=Prompt' 'Reply with the single word: beta2. Do not run any tools.'`, `control-openhands browser click 'testid=edit-automation-save' --observe 'role=status'` (`Automation updated`); `api GET /api/automation/v1/<automation-id>` shows the new prompt. A timeout above 1800 is refused client-side with `Timeout cannot exceed 1800 seconds`.
- **Turn off (`F03.turn-off-confirmation`).** From the Beta row menu click `[data-testid^="running-automation-turn-off-"]`; `control-openhands browser snapshot 'role=dialog'` reads `Turn off "QA_F03 Beta"? It will stop running on its schedule until you turn it back on.` with Cancel and Turn off. `control-openhands browser click 'role=dialog >> role=button[name="Cancel"]'` keeps it enabled; repeat and `control-openhands browser click 'role=dialog >> role=button[name="Turn off"]'`, `browser reload`: `control-openhands browser count '[data-testid^="running-automation-row-"] >> has-text=QA_F03 Beta'` is `0` and the API shows `enabled: false`. Turn it back on from `/automations` (Preconditions) for the pin checks.
- **Pin (`F03.pinned-automations-grid`).** Run `control-openhands browser goto /` (turning Beta back on left `/automations`); from Pong's row menu click `[data-testid^="running-automation-pin-"]`; `control-openhands browser text 'testid=pinned-automations-dashboard'` shows `Pinned automations`, `QA_F03 Pong`, the prompt, `cron` and the last status; `browser storage --values` holds the id under `oh:home-pinned-automations:<backend>:-`. Pin Beta too; `control-openhands browser eval "[...document.querySelectorAll('[data-testid^=pinned-automation-card-]')].map(e=>e.innerText.split('\n')[0]).join(',')"` is `QA_F03 Pong,QA_F03 Beta`. `control-openhands browser click '[data-testid^="pinned-automation-card-"] >> role=link[name="QA_F03 Pong"]' --expect-url '/automations/[^/?]+$'` opens the detail page.
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 3f28cb0c06..a70e2325ff 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F05-composer.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F05-composer.md
@@ -24,7 +24,7 @@ Source: `src/components/features/chat/`, `src/components/features/controls/tools
- `F05.dictation`: the mic button dictates into the composer; without a microphone it toasts a failure and resets.
- `F05.llm-profile-picker`: the model pill switches the running conversation's LLM profile; the next reply uses the new model.
- `F05.profile-identity`: the pill names the profile the conversation was started with, also after a reload and when another profile uses the same model; changing the default profile in Settings changes the home pill, not an existing conversation's.
-- `F05.overflow-menu`: at very narrow widths the model pill collapses into a "More input actions" menu whose Model submenu fits the viewport and switches the profile.
+- `F05.overflow-menu`: at very narrow widths the model pill collapses into a "More input actions" menu whose Model submenu fits the viewport and switches the profile; Escape closes the menu and returns focus to its trigger.
- `F05.context-window-meter`: the ring shows context use; its popover opens the Usage tab and compacts the context.
- `F05.stop-resume`: Stop pauses a running agent (status `Stopped`, Play button); Play resumes it.
- `F05.queued-message`: a follow-up typed while the agent runs is accepted and answered.
@@ -71,7 +71,7 @@ Preconditions:
- **Dictation failure (`F05.dictation`).** Run `control-openhands browser click 'testid=chat-dictation-button' --observe '[role=status]' --observe-ms 3000`: the observed toasts include `Dictation failed. Check microphone access and your voice input settings.` and `control-openhands browser attr 'testid=chat-dictation-button' aria-label` is back to `Start dictation`. Real dictation is blocked without a microphone.
- **Switch LLM profile (`F05.llm-profile-picker`).** Run `control-openhands browser click 'testid=chat-input-llm-profile'` and `control-openhands browser text 'testid=chat-input-llm-profile-popover'`: `AVAILABLE PROFILES`, `deepseek-flash` / `deepseek/deepseek-flash`, `deepseek-pro` / `deepseek/deepseek-v4-pro`, `LLM Profiles`. Run `control-openhands browser click 'testid=chat-input-llm-profile-option-deepseek-pro'`, `control-openhands browser wait-text 'Switched to profile' --timeout 15000`, `control-openhands browser text 'testid=chat-input-llm-profile'` (`deepseek-pro`) and `control-openhands conversation status <id>` (`model` is `deepseek/deepseek-v4-pro`). Send `Reply with only: pro-ok` as in Type and send; the reply is `pro-ok`. After `control-openhands browser reload` the pill still reads `deepseek-pro`. Switch back with `control-openhands browser click 'testid=chat-input-llm-profile'` and `control-openhands browser click 'testid=chat-input-llm-profile-option-deepseek-flash'`.
- **Profile identity (`F05.profile-identity`).** Arrange two profiles on one model right before this bullet: `QA_DUMMY_KEY=qa control-openhands llm set --profile qa-alpha --model openai/gpt-4o --api-key-env QA_DUMMY_KEY --no-validate --no-activate` and `QA_DUMMY_KEY=qa control-openhands llm set --profile qa-zeta --model openai/gpt-4o --api-key-env QA_DUMMY_KEY --no-validate` (`qa-zeta` becomes the default: `control-openhands api GET /api/profiles --pick active_profile` is `qa-zeta`). Run `control-openhands conversation start --prompt "Reply with only: identity-ok"` (no `--wait`; with the dummy key the agent errors at once, which does not matter here) and note its `id` as `<zeta-id>`. `control-openhands browser wait 'testid=chat-input-llm-profile'`, then `control-openhands browser text 'testid=chat-input-llm-profile'` is `qa-zeta`, not the alphabetically first profile on that model. `control-openhands browser reload`, the wait and the text again: still `qa-zeta`. With a DeepSeek key, arrange both profiles on `deepseek/deepseek-flash` with `--api-key-env DEEPSEEK_API_KEY` and without `--no-validate`, and start with `--prompt "Reply with only: identity-ok. Do not run any tools." --wait --timeout 180`. Then the reply also survives the reload: `control-openhands browser text 'testid=agent-message >> nth=-1'` is `identity-ok` before and after it (driven 2026-10-08). The `default` agent profile still names `deepseek-flash` here, so the launch runs the pill's profile through agent settings (`F13.llm-pill-precedence`). The name is stamped on the conversation in this browser: `control-openhands browser eval "JSON.parse(localStorage.getItem('openhands-agent-server-conversation-metadata'))['<zeta-id>'].active_profile"` is `qa-zeta`. Change the default in Settings: `control-openhands browser goto /settings/llm`, `control-openhands browser wait 'testid=add-llm-profile'`, `control-openhands browser click '[data-testid=profile-row]:has([title="qa-alpha"]) >> testid=profile-menu-trigger'`, `control-openhands browser click 'testid=profile-actions-menu >> testid=profile-set-active'`, `control-openhands browser wait-text 'Switched to profile "qa-alpha"'` (`control-openhands api GET /api/profiles --pick active_profile` is `qa-alpha`). The home pill follows the default: `control-openhands browser goto /`, the wait, and `control-openhands browser text 'testid=chat-input-llm-profile'` is `qa-alpha`; the started conversation keeps its own: `control-openhands browser goto /conversations/<zeta-id>`, the wait, and the text is `qa-zeta` (`control-openhands browser screenshot 'testid=interactive-chat-box' --feature F05.profile-identity --name conversation-keeps-qa-zeta`). Restore: make `deepseek-flash` the default again the same way (its row's `profile-menu-trigger`, then `profile-set-active`), then `control-openhands api DELETE /api/profiles/qa-alpha --write` and `control-openhands api DELETE /api/profiles/qa-zeta --write` (arrange, not proof) so `/model` below still counts two profiles. In a model-free run make `qa-zeta` the default instead and delete only `qa-alpha`. Return with `control-openhands browser goto /conversations/<id>`.
-- **Overflow menu (`F05.overflow-menu`).** Run `control-openhands browser viewport narrow` and `control-openhands browser snapshot 'testid=interactive-chat-box'`: the model pill is gone and `button "More input actions"` appears (at `phone` and `350x800` the pill stays inline). `control-openhands browser click 'role=button[name="More input actions"]' --observe '[data-testid=chat-input-overflow-menu]' --observe-ms 1000`: `observed` goes from `<absent>` to `Model` (the menu appears in the observed sequence; its delay varies from run to run, so do not assert a time) and has no later `<absent>`. `control-openhands browser count 'testid=chat-input-overflow-menu'` is `1` and `control-openhands browser attr 'role=button[name="More input actions"]' aria-expanded` is `true`. Run `control-openhands browser click 'testid=overflow-model-button'`; `control-openhands browser text 'testid=chat-input-overflow-menu'` reads `Model`, `AVAILABLE PROFILES`, `deepseek-flash` / `deepseek/deepseek-flash`, `deepseek-pro` / `deepseek/deepseek-v4-pro`, `LLM Profiles`. Run `control-openhands browser bbox 'testid=chat-input-llm-profile-option-deepseek-flash'` and `control-openhands browser bbox 'role=link[name="LLM Profiles"]'`. Expected: both `insideViewport` `true`. Known failure (#18063): the submenu opens to the right of the menu, so the profile row starts at x about 270 with width 210 and the `LLM Profiles` link sits at y about 725, below the 700 px viewport (both `insideViewport` `false`); the profile clicks below still work because the first two rows keep a strip of about 50 px on screen (from x about 270 to the 320 px edge), while a click on the `LLM Profiles` link times out (`browser screenshot --feature F05.overflow-menu --name model-submenu-narrow`). `control-openhands browser click 'testid=chat-input-llm-profile-option-deepseek-pro'` and `control-openhands browser wait-text 'Switched to profile' --timeout 15000` close the menu (count `0`), and `control-openhands conversation status <id>` shows `deepseek/deepseek-v4-pro`. Switch back the same way: the trigger, `testid=overflow-model-button`, then `testid=chat-input-llm-profile-option-deepseek-flash` (`conversation status <id>` shows `deepseek/deepseek-flash`). Open the menu once more; a second click on the trigger closes it (`aria-expanded` `false`). Return with `control-openhands browser viewport desktop`.
+- **Overflow menu (`F05.overflow-menu`).** Run `control-openhands browser viewport narrow` and `control-openhands browser snapshot 'testid=interactive-chat-box'`: the model pill is gone and `button "More input actions"` appears (at `phone` and `350x800` the pill stays inline). `control-openhands browser click 'role=button[name="More input actions"]' --observe '[data-testid=chat-input-overflow-menu]' --observe-ms 1000`: `observed` goes from `<absent>` to `Model` (the menu appears in the observed sequence; its delay varies from run to run, so do not assert a time) and has no later `<absent>`. `control-openhands browser count 'testid=chat-input-overflow-menu'` is `1` and `control-openhands browser attr 'role=button[name="More input actions"]' aria-expanded` is `true`. Run `control-openhands browser click 'testid=overflow-model-button'`; `control-openhands browser text 'testid=chat-input-overflow-menu'` reads `Model`, `AVAILABLE PROFILES`, `deepseek-flash` / `deepseek/deepseek-flash`, `deepseek-pro` / `deepseek/deepseek-v4-pro`, `LLM Profiles`. Run `control-openhands browser bbox 'testid=chat-input-llm-profile-option-deepseek-flash'` and `control-openhands browser bbox 'role=link[name="LLM Profiles"]'`. Expected: both `insideViewport` `true`. Known failure (#18063): the submenu opens to the right of the menu, so the profile row starts at x about 270 with width 210 and the `LLM Profiles` link sits at y about 725, below the 700 px viewport (both `insideViewport` `false`); the profile clicks below still work because the first two rows keep a strip of about 50 px on screen (from x about 270 to the 320 px edge), while a click on the `LLM Profiles` link times out (`browser screenshot --feature F05.overflow-menu --name model-submenu-narrow`). `control-openhands browser click 'testid=chat-input-llm-profile-option-deepseek-pro'` and `control-openhands browser wait-text 'Switched to profile' --timeout 15000` close the menu (count `0`), and `control-openhands conversation status <id>` shows `deepseek/deepseek-v4-pro`. Switch back the same way: the trigger, `testid=overflow-model-button`, then `testid=chat-input-llm-profile-option-deepseek-flash` (`conversation status <id>` shows `deepseek/deepseek-flash`). Open the menu once more; a second click on the trigger closes it (`aria-expanded` `false`). Escape closes it too: record it with `control-openhands browser record start --feature F05.overflow-menu --name escape-narrow`, open the menu, `control-openhands browser press Escape` and `control-openhands browser wait 'testid=chat-input-overflow-menu' --state detached --timeout 3000`; `browser attr 'role=button[name="More input actions"]' aria-expanded` is `false` and `control-openhands browser eval "document.activeElement.getAttribute('aria-label')"` is `More input actions`. Repeat with focus on a menu item (`control-openhands browser focus 'testid=overflow-model-button'` before Escape) and with the Model submenu open (`browser focus 'testid=chat-input-llm-profile-option-deepseek-flash'`): the same result. `control-openhands browser record stop --gif`. Return with `control-openhands browser viewport desktop`.
- **Context window (`F05.context-window-meter`).** Run `control-openhands browser attr 'testid=context-window-meter' aria-label` (`Context window usage: 1% used (99% left)`), `control-openhands browser click 'testid=context-window-meter'` and `control-openhands browser text 'testid=context-window-meter-popover'` (`Context Window`, the percentage, `Compact context`, tokens such as `11.7k / 1.0M`, `Usage`). `control-openhands browser click 'testid=context-window-meter-bar-button'` (the progress bar) closes the popover and opens the Usage tab (`control-openhands browser testids --filter usage` lists `conversation-tab-usage`, `usage-panel`). The popover's `Usage` row does the same: from another tab (`control-openhands browser click 'testid=conversation-tab-files'`), reopen the ring, `control-openhands browser click 'testid=context-window-plan-usage'` and `control-openhands browser wait 'testid=usage-panel'` (the panel mounts a moment later; an immediate `count` is `0`). Reopen the ring, then `control-openhands browser click 'testid=context-window-compact-button' --observe '[role=status]' --observe-ms 25000`: `Context compaction started`, then `Context compacted`; `control-openhands conversation events <id> --kinds CondensationRequest,Condensation --last 5` lists both kinds.
- **Stop and resume (`F05.stop-resume`) with a queued follow-up (`F05.queued-message`).** Send `Run this shell command: sleep 25 && echo slept. Then reply with only: done` (type + Enter as above), `control-openhands conversation wait <id> --until running --timeout 30` and `control-openhands browser wait 'testid=stop-button'`; `control-openhands browser attr 'testid=stop-button' aria-label` is `Stop`. While it runs, `control-openhands browser enabled 'testid=submit-button'` is `false` with an empty field; type `Also reply with: queued-ok` (enabled turns `true`) and press Enter. Run `control-openhands browser click 'testid=stop-button'` and `control-openhands browser wait 'testid=play-button' --timeout 30000`; `control-openhands browser text 'testid=interactive-chat-box'` ends `Stopped`, `control-openhands browser attr 'testid=play-button' aria-label` is `Resume the agent task` and `control-openhands conversation status <id>` is `paused`. Run `control-openhands browser click 'testid=play-button'`, wait for `running` then the terminal state: the last agent message is `done queued-ok` and `testid=user-message >> has-text=queued-ok` counts `1`.
- **Slash menu (`F05.slash-menu`).** On a cleared composer run `control-openhands browser type 'testid=chat-input' '/'`, `control-openhands browser count 'testid=slash-command-menu >> role=option'` (37 with the bundled skills) and `control-openhands browser text 'testid=slash-command-menu'`: `Commands`, then `/btw`, `/model`, `/goal`, `/plan`, `/code` with descriptions, then skills (`/agent-creator`, …); no `/new` on the local backend. Type `mo`: the count drops (descriptions match too). `control-openhands browser attr 'testid=slash-command-menu >> role=option >> nth=0' aria-selected` is `true`; after `control-openhands browser press ArrowDown` the selected option is `/plan` (read it with `control-openhands browser text 'testid=slash-command-menu >> [aria-selected=true]'`). `control-openhands browser press Escape` closes the menu and keeps `/mo`. On `/btw`, `control-openhands browser press Tab` leaves `/btw ` in the field; on `/model ` (trailing space) the menu lists `/model deepseek-flash` and `/model deepseek-pro`, and `control-openhands browser press Enter` inserts `/model deepseek-flash ` instead of sending.
@@ -91,8 +91,9 @@ Preconditions:
- `conversation wait <id>` right after sending returns at once with the previous `finished` status. Use `conversation wait <id> --fresh`; `--until running` then a terminal wait only works for turns longer than the poll interval (a `Reply with only` turn on deepseek-flash can finish first, and `--until running` then times out).
- The slash menu stays open while you type a command; press Escape before Enter or Enter selects the highlighted option instead of sending.
- Conversation drafts are saved 500 ms after the last keystroke and not flushed on unmount (the home draft is): navigating sooner restores the older draft (fail row) (#17926). Wait a second before navigating when testing restoration.
-- Escape closes the `+` tools menu (`browser count 'testid=tools-context-menu'` is `0`) and returns focus to `chat-plus-button`, whether focus was on the trigger or on a menu item. A menu left open covers attachments in screenshots. Escape should also close the narrow-width "More input actions" overflow menu and return focus to its trigger. Known failure (reproduced 2026-10-08): `browser count 'testid=chat-input-overflow-menu'` stays `1` and the trigger's `aria-expanded` stays `true`, with focus on the trigger or on a menu item (#18172): close it with its trigger or an outside click (`browser mouse-click 160 100` at `narrow`).
-- After a client-side switch between conversations (a sidebar row, or a user-message branch in `F06.branch-from-here`) Send keeps the previous conversation's state (repro candidate, not filed yet). Switching from an empty composer to one with a saved draft restores the text next to a disabled Send; this is a race (Send came up enabled in 1 of 7 tries), so one enabled reading does not show the bug is gone. Switching from a composer with text to an empty one leaves Send enabled, and clicking it does nothing. Typing or `browser reload` fixes it, and Enter in the field sends the restored text anyway. Read `browser enabled 'testid=submit-button'` after a reload, or after typing.
+- Escape closes the `+` tools menu (`browser count 'testid=tools-context-menu'` is `0`) and returns focus to `chat-plus-button`, whether focus was on the trigger or on a menu item. A menu left open covers attachments in screenshots. Escape also closes the narrow-width "More input actions" overflow menu (`browser count 'testid=chat-input-overflow-menu'` is `0`, the trigger's `aria-expanded` is `false`) and returns focus to the trigger, whether focus was on the trigger, on `overflow-model-button` or on a profile row of the open Model submenu (#18174).
+- After a client-side switch between conversations (a sidebar row, or a user-message branch in `F06.branch-from-here`) Send keeps the previous conversation's state (Known failure, #18202). Switching from an empty composer to one with a saved draft restores the text next to a disabled Send; this is a race (Send came up enabled in 1 of 7 tries), so one enabled reading does not show the bug is gone. Switching from a composer with text to an empty one leaves Send enabled, and clicking it does nothing. Typing or `browser reload` fixes it, and Enter in the field sends the restored text anyway. Read `browser enabled 'testid=submit-button'` after a reload, or after typing.
+- Expected: a prompt that a suggestion chip (`testid=chat-suggestions`) or a `+` > **Macros** item writes into the composer enables Send at once. Known failure (reproduced 2026-10-09, #18202): on a new No workspace conversation, `browser click 'testid=chat-suggestions >> text=Fix README'` (or `browser click 'testid=chat-plus-button'`, `browser hover 'testid=macros-button'`, `browser click 'testid=fix-readme-button'`) fills the field, but `browser enabled 'testid=submit-button'` is `false` (5 of 5 tries at desktop; a race at phone) until a character is typed. Both paths call `setMessageToSend`, which writes the text without an input event.
- Uploaded chips and thumbnails have no test ids. Their remove buttons are named `Remove <file name>` (file chip) and `Remove image` (thumbnail), and Send, Stop and Play read `Send message`, `Stop` and `Resume the agent task` (#18033); select by those names, or by text/alt.
- `browser upload` feeds the hidden input directly. To drive the real path, `+` > Add Files and Images, use `control-openhands browser upload-via 'testid=add-files-and-images-button' <file>`, which answers the native file chooser.
- The home composer has no `interactive-chat-box` wrapper: scope its chips and thumbnails with `testid=home-chat-launcher`.
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 0de2633a5d..ad265a3814 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
@@ -112,7 +112,7 @@ Preconditions:
- Confirmation mode with the default Security Analyzer `LLM` uses ConfirmRisky(HIGH): harmless commands run without a prompt. Set the analyzer to `None` to confirm every action, and restore `LLM` afterwards. The composer stays editable while waiting (status "User needed"); the buttons read Cancel / Continue.
- Known failure, Canvas UI: after Cancel the event group keeps a spinner and "1/2 actions completed" forever, even after reload, although the conversation is idle and the server recorded `UserRejectObservation` (the group only counts `ObservationEvent`s) (#17920). Evidence `F06.confirmation-mode/rejected-after-reload.png`.
- Known failure, Canvas UI: the confirm/cancel shortcuts check only `metaKey`; on Linux/Windows `Control+Enter` does nothing although the composer and plan shortcuts accept Ctrl (#17928).
-- Known failure, Canvas UI (repro candidate, not filed yet): after a user-message branch the composer holds the prefilled text but Send stays disabled (`browser enabled 'testid=submit-button'` is `false`, and a click on it times out with `element is not enabled`). The prefill is a saved draft that `src/hooks/chat/use-draft-persistence.ts` writes into the field without an input event, and the composer is reused across the client-side switch, so its Send state keeps the source conversation's empty field. Editing the text, a reload, or Enter in the field (`browser press Enter --selector 'testid=chat-input'`) all work. A draft restored by a sidebar switch behaves the same (F05 Gotchas).
+- Known failure, Canvas UI (#18202): after a user-message branch the composer holds the prefilled text but Send stays disabled (`browser enabled 'testid=submit-button'` is `false`, and a click on it times out with `element is not enabled`). The prefill is a saved draft that `src/hooks/chat/use-draft-persistence.ts` writes into the field without an input event, and the composer is reused across the client-side switch, so its Send state keeps the source conversation's empty field. Editing the text, a reload, or Enter in the field (`browser press Enter --selector 'testid=chat-input'`) all work. A draft restored by a sidebar switch behaves the same (F05 Gotchas).
- Known failure, likely Agent Server: every message in a branched conversation ends with `ConversationErrorEvent` `LLMAuthenticationError` ("Your LLM API key appears to be invalid or has expired."), while the source conversation keeps working (OpenHands/software-agent-sdk#5465). Do not use branches as fixtures for other checks.
- "The terminal session was reset because the underlying tmux server/session disappeared" (clock `status-icon` on the row) appears after a `service stop agent-server` / `restart` cycle. Since `launch` gives each run its own short `TMUX_TMPDIR`, other runs no longer cause it. The underlying product bug is still open (OpenHands/OpenHands#17946) for plain `agent-canvas` instances: `bin/agent-canvas.mjs` → `scripts/dev-with-automation.mjs` passes `TMUX_TMPDIR=<state>/tmux` to the Agent Server but `ensureDirectories` never creates it (`dev-safe.mjs` does), so tmux 3.4 falls back to the shared `/tmp/tmux-<uid>/openhands` socket and another instance's restart or stop kills this run's windows. The model usually reruns the command; expect an extra action.
- The `ConversationErrorEvent` banner is not restored after navigating away and back (or a reload); read the error from the events API instead.
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 034d9317c9..36f5bad440 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
@@ -7,7 +7,7 @@ composer. From the header a user renames, stops or resumes the runtime, opens
the skills, hooks and agent-tools dialogs, exports or downloads the
conversation, jumps to usage and deletes it.
-Source: `src/routes/conversation.tsx`, `src/components/features/conversation/`, `src/components/features/conversation-panel/`, `src/components/features/controls/server-status-context-menu.tsx`, `src/hooks/use-conversation-name-context-menu.ts`.
+Source: `src/routes/conversation.tsx`, `src/components/features/conversation/`, `src/components/features/conversation-panel/`, `src/components/features/controls/server-status-context-menu.tsx`, `src/hooks/use-conversation-name-context-menu.ts`, `src/api/agent-server-adapter.ts` (the Hooks dialog's `toHooksResponse`).
## Sub-features
@@ -107,5 +107,5 @@ Preconditions:
- 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`.
- Expected: an agent hook shows its `system_prompt`, the text it runs with, even when it also carries a `prompt`. Known failure (reproduced 2026-10-09): with `{"session_start":[{"hooks":[{"type":"agent","prompt":"QA_F07 stray prompt","system_prompt":"QA_F07 agent system prompt"}]}]}` in `hooks.json`, close, reopen and Refresh the dialog, then expand `Session Start 1 hook`: it reads `Matcher * Commands QA_F07 stray prompt Type: agent Timeout: 60s`, and `QA_F07 agent system prompt` appears nowhere (`control-openhands browser count 'text=QA_F07 agent system prompt'` is `0`), although `control-openhands api POST /api/hooks --write --data '{"project_dir":"<run>/workspace/qa-f07-repo"}'` returns both fields (#18206). Write the recipe's two-event `hooks.json` (`pre_tool_use` and `stop`) back and Refresh afterwards.
-- Expected: with the Agent Server down, Refresh in the Hooks dialog shows a fetch error about hooks. Known failure (reproduced 2026-10-06): after `control-openhands service stop agent-server` on an open dialog, `refresh-hooks` and `browser wait 'testid=hooks-modal >> text=Failed to fetch' --timeout 30000`, it reads `Failed to fetch skills. Please try again later.` (the shared `COMMON$FETCH_ERROR` string) (#18088). Run it last, or `restart` afterwards.
-- After the user-message branch in `F07.branch-from-message`, Send stays disabled next to the prefilled text (`browser enabled 'testid=submit-button'` is `false`; repro candidate, not filed yet, see F05 Gotchas). Enter in the field sends it.
+- With the Agent Server down, Refresh in the Hooks dialog shows its own fetch error (#18177): after `control-openhands service stop agent-server` on an open dialog, `refresh-hooks` and `browser wait 'testid=hooks-modal >> text=Failed to fetch hooks' --timeout 30000`, the dialog reads `Failed to fetch hooks. Please try again later.` under its explanation (`browser count 'testid=hooks-modal >> text=Failed to fetch skills'` is `0`). The 502s and the closed WebSocket that `browser errors --app-only` then lists are the outage itself. Run it last, or `restart` afterwards.
+- After the user-message branch in `F07.branch-from-message`, Send stays disabled next to the prefilled text (`browser enabled 'testid=submit-button'` is `false`; Known failure #18202, see F05 Gotchas). Enter in the field sends it.
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 41e8bb8fe2..b68134db07 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
@@ -9,7 +9,7 @@ call, renames, duplicates, deletes and sets one as **Default**, and bulk-adds
models from a provider. The Default profile is the model new conversations run
on. Provider connections themselves (add, edit, delete) are F11.
-Source: `src/routes/llm-settings.tsx`, `src/components/features/settings/llm-profiles/`, `src/components/shared/modals/settings/model-selector.tsx`, `src/components/features/settings/llm-settings/openai-subscription-auth-card.tsx`, `src/components/features/settings/sdk-settings/`, `src/api/profiles-service/profiles-service.api.ts`.
+Source: `src/routes/llm-settings.tsx`, `src/components/features/settings/llm-profiles/`, `src/components/shared/modals/settings/model-selector.tsx`, `src/components/features/settings/settings-input.tsx`, `src/components/features/settings/llm-settings/openai-subscription-auth-card.tsx`, `src/components/features/settings/sdk-settings/`, `src/api/profiles-service/profiles-service.api.ts`.
## Sub-features
@@ -75,14 +75,14 @@ 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 red. An existing name (`QA_flash`) also gives `enabled` `false`, but `aria-invalid` stays `false` and nothing says why (see Gotchas). `QA_flash2` gives `true`.
+- **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`, but `aria-invalid` stays `false` and nothing says why (see Gotchas). `QA_flash2` gives `true`.
- **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.
- **Row menu (`F10.actions-menu`).** Run `control-openhands browser click '[data-testid=profile-row]:has([title="QA_flash"]) >> testid=profile-menu-trigger'` and `control-openhands browser snapshot 'testid=profile-actions-menu'`: menuitems Edit, Rename, Duplicate, `Set as default [disabled]`, Delete on the Default row (all enabled on other rows). `control-openhands browser eval "document.activeElement.dataset.testid"` is `profile-edit`: focus moves into the menu on open. Run `control-openhands browser press ArrowDown` and that eval four times: `profile-rename`, `profile-duplicate`, `profile-delete` (the disabled Set as default is skipped), `profile-edit` (wraps); `control-openhands browser press ArrowUp` twice gives `profile-delete`, `profile-duplicate`. `control-openhands browser press Tab` closes the menu (`browser count 'testid=profile-actions-menu'` is `0`) and moves focus on to the next row's trigger: the row-aware eval `control-openhands browser eval "(()=>{const a=document.activeElement;const r=a.closest('[data-testid=profile-row]');return a.dataset.testid+'@'+(r?r.querySelector('[title]').title:'-')})()"` is `profile-menu-trigger@QA_pro`. Reopen the `QA_flash` menu and `control-openhands browser press Escape`: count `0`, and the row-aware eval is `profile-menu-trigger@QA_flash` (focus is back on its own trigger). Reopen it and `control-openhands browser click 'role=heading[name="Available Profiles"]'` closes it too. Keyboard open: `control-openhands browser focus '[data-testid=profile-row]:has([title="QA_pro"]) >> testid=profile-menu-trigger'` and `control-openhands browser press Enter` put focus on `profile-edit`; on this non-default row `control-openhands browser snapshot 'testid=profile-actions-menu'` lists the same five menuitems with none `[disabled]`, and `control-openhands browser screenshot --feature F10.actions-menu --name menu-nondefault` shows them with the focus ring on Edit. Five `ArrowDown`s walk `profile-rename`, `profile-duplicate`, `profile-set-active`, `profile-delete`, `profile-edit`. `control-openhands browser press Escape` closes it, and the row-aware eval is `profile-menu-trigger@QA_pro`.
- **Set as default (`F10.set-default`).** Run `control-openhands browser click '[data-testid=profile-row]:has([title="QA_pro"]) >> testid=profile-menu-trigger'`, `control-openhands browser click 'testid=profile-actions-menu >> testid=profile-set-active'` and `control-openhands browser wait-text 'Switched to profile "QA_pro"'`. After `control-openhands browser reload`, `control-openhands browser count 'testid=profile-active-badge'` is `1` and `control-openhands browser count '[data-testid=profile-row]:has([title="QA_pro"]) >> testid=profile-active-badge'` is `1`; `control-openhands api GET /api/profiles` has `active_profile` `QA_pro` and `control-openhands llm show` reports `active.model` `deepseek/deepseek-v4-pro`. A profile's Base URL follows it into the agent settings: `QA_adv` (Custom model above) carries one: Set as active on its row makes `control-openhands llm show` report `active.base_url` `https://api.deepseek.com`, also under `agent_settings.llm.base_url` in `control-openhands api GET /api/settings`. With a key, `control-openhands conversation start --prompt "Reply with only: adv-ok. Do not run any tools." --wait --timeout 240` then shows the pill `QA_adv` and the reply `adv-ok` (driven 2026-10-08; `QA_pro` likewise replied `pro-ok` on `deepseek/deepseek-v4-pro`). Make `QA_pro` the Default again before Duplicate.
- **Duplicate (`F10.duplicate`).** From the `QA_pro` row menu click `testid=profile-actions-menu >> testid=profile-duplicate` and `control-openhands browser wait-text 'Profile duplicated as "QA_pro-copy"'`; repeat on `QA_pro` for `Profile duplicated as "QA_pro-copy-1"`, and on `QA_pro-copy` for `Profile duplicated as "QA_pro-copy-copy"`. After a reload `control-openhands browser text 'testid=profile-row >> has-text=QA_pro-copy-1'` reads `QA_pro-copy-1 deepseek/deepseek-v4-pro`, the badge stays on `QA_pro`, and `api GET /api/profiles` has `api_key_set: true` on every copy.
-- **Rename (`F10.rename`).** Open the `QA_pro-copy-copy` row menu and click `testid=profile-actions-menu >> testid=profile-rename`. `control-openhands browser value 'testid=rename-profile-input'` is `QA_pro-copy-copy` and `browser eval "document.activeElement.dataset.testid"` is `rename-profile-input`. `control-openhands browser fill 'testid=rename-profile-input' '-bad'` makes `control-openhands browser enabled 'role=dialog[name="Rename Profile"] >> role=button[name="Rename"]'` `false` and `control-openhands browser attr 'testid=rename-profile-rule' class` contains `text-red-400`. Filling the old name and clicking Rename closes the modal with no toast (`browser count 'testid=rename-profile-modal'` is `0`); `control-openhands browser click 'role=dialog[name="Rename Profile"] >> role=button[name="Cancel"]'` also closes it. To rename, reopen, fill `QA_renamed`, `control-openhands browser press Enter`, `control-openhands browser wait-text 'Profile renamed to "QA_renamed"'`, reload, and the count of `'[data-testid=profile-row]:has([title="QA_renamed"])'` is `1` while the old name's is `0`. Renaming the Default row (`QA_pro` → `QA_pro2`, submit with `testid=rename-profile-submit`) keeps its badge (`browser text '[data-testid=profile-row]:has([data-testid=profile-active-badge])'` reads `QA_pro2 … Default`) and `active_profile` follows.
+- **Rename (`F10.rename`).** Open the `QA_pro-copy-copy` row menu and click `testid=profile-actions-menu >> testid=profile-rename`. `control-openhands browser value 'testid=rename-profile-input'` is `QA_pro-copy-copy` and `browser eval "document.activeElement.dataset.testid"` is `rename-profile-input`. `control-openhands browser fill 'testid=rename-profile-input' '-bad'` makes `control-openhands browser enabled 'role=dialog[name="Rename Profile"] >> role=button[name="Rename"]'` `false` and `control-openhands browser attr 'testid=rename-profile-rule' class` contains `text-danger` (the theme danger colour, #18199). Filling the old name and clicking Rename closes the modal with no toast (`browser count 'testid=rename-profile-modal'` is `0`); `control-openhands browser click 'role=dialog[name="Rename Profile"] >> role=button[name="Cancel"]'` also closes it. To rename, reopen, fill `QA_renamed`, `control-openhands browser press Enter`, `control-openhands browser wait-text 'Profile renamed to "QA_renamed"'`, reload, and the count of `'[data-testid=profile-row]:has([title="QA_renamed"])'` is `1` while the old name's is `0`. Renaming the Default row (`QA_pro` → `QA_pro2`, submit with `testid=rename-profile-submit`) keeps its badge (`browser text '[data-testid=profile-row]:has([data-testid=profile-active-badge])'` reads `QA_pro2 … Default`) and `active_profile` follows.
- **Rename onto an existing name (`F10.rename-conflict`).** The Rename step above already renamed `QA_pro-copy-copy` and `QA_pro`, so use the remaining fixtures: open the `QA_pro-copy` row menu, click `testid=profile-actions-menu >> testid=profile-rename`, fill `testid=rename-profile-input` with `QA_pro2` and click `testid=rename-profile-submit`. The modal stays open (`count 'testid=rename-profile-modal'` is `1`), `control-openhands browser toasts` has the server's message `Profile 'QA_pro2' already exists` (no `HTTP request failed` prefix, no JSON), and `browser screenshot --feature F10.rename-conflict --name toast` shows that toast above the open modal. Close it with `control-openhands browser click 'role=dialog[name="Rename Profile"] >> role=button[name="Cancel"]'`.
- **Edit (`F10.edit`).** Open the `QA_renamed` row menu and click `testid=profile-actions-menu >> testid=profile-edit`. `browser text 'testid=profile-editor-title'` is `Edit LLM Profile`, `browser text 'testid=profile-editor-description'` is `Editing profile "QA_renamed" - save to apply changes`, `browser attr 'testid=llm-api-key-input' placeholder` is `<hidden>`, `browser count 'testid=set-indicator'` is `1` and `browser enabled 'testid=save-profile-btn'` is `false` (screenshot `--feature F10.edit --name loaded`). Run `control-openhands browser click 'testid=llm-model-input'`, `control-openhands browser click 'testid=model-item-deepseek-flash'` (Save turns `true`), `control-openhands browser fill 'testid=profile-name-input' QA_edited`, Save, and `control-openhands browser wait-text 'Profile "QA_edited" updated' --timeout 60000`. After a reload `browser text '[data-testid=profile-row]:has([title="QA_edited"])'` reads `QA_edited deepseek/deepseek-flash` and `QA_renamed` is gone. The save passed validation without retyping the key, which proves the stored key was kept.
- **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).
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 734c02c32b..dd32bddce2 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
@@ -1,7 +1,7 @@
# F13 — Agent profiles
Settings → Agent is a library of named agent profiles. Each profile is either an
-OpenHands agent (an LLM profile plus sub-agents, the LLM-switch tool, parallel
+OpenHands agent (an LLM profile, a system prompt, a tool set, parallel
tool calls, and MCP-server and secret scopes) or an ACP external agent (a preset
such as Claude Code, Codex or Gemini CLI, its launch command, a model and
provider credentials). One profile is active; new conversations started from
@@ -15,7 +15,7 @@ Source: `src/routes/agent-profiles-settings.tsx`, `src/routes/agent-settings.tsx
- `F13.create`: **Add agent profile** opens the editor; Save toasts `Profile "<name>" created` and the row is listed after a reload.
- `F13.name-validation`: an empty or malformed name marks the field invalid (red rule) and disables Save; a duplicate name disables Save without a message.
- `F13.editor-no-llm`: with no LLM profiles, an OpenHands profile shows `Create an LLM profile first to define an OpenHands agent.` and Save toasts `Select an LLM profile for this agent.` without creating anything.
-- `F13.openhands-options`: OpenHands profiles offer an LLM profile dropdown, **Enable sub-agents** (off by default), **Let the agent switch LLM profiles** (on by default) and **Parallel tool calls** (default 1, minimum 1); values persist and reopen.
+- `F13.openhands-options`: OpenHands profiles offer an LLM profile dropdown (the active LLM profile by default), a **System prompt** choice (`OpenHands default`), **Parallel tool calls** (default 1, minimum 1) and **Tools**: `Standard tools` (default) lists the server's standard set read-only, `Choose tools` turns the server's tool catalog into switches; values persist and reopen.
- `F13.edit`: Edit reopens the stored values with Save disabled until something changes; renaming keeps the profile id; Back and Cancel leave without saving.
- `F13.stale-llm-ref`: editing a profile whose LLM profile no longer exists preselects the active LLM profile and enables Save, so saving heals the reference.
- `F13.set-active`: **Set as active** toasts `Switched to profile "<name>"`, moves the Default badge and is disabled on the active row.
@@ -53,14 +53,14 @@ 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) → `false`/`false`, and nothing explains why; `'bad name!'` → `true`/`false`; `QA_ok` → `false`/`true`. The rule under the field is always shown (grey) and turns red (`text-red-400`) 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)`. Take `control-openhands browser screenshot --feature F13.name-validation --name duplicate` with `default` filled.
+- **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) → `false`/`false`, and nothing explains why; `'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)`. Take `control-openhands browser screenshot --feature F13.name-validation --name duplicate` with `default` filled.
- **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 `Claude Opus (1M)`; `agent-profile-no-llm` is gone (ACP needs no LLM profile). 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 `GPT-5.5` / `Gemini 2.5 Pro`. 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`.
- **Edit and rename (`F13.edit`).** Run `control-openhands browser click 'testid=agent-profile-row >> has-text=QA_acp_custom >> testid=agent-profile-menu-trigger'`; `control-openhands browser snapshot 'testid=agent-profile-actions-menu'` lists `Edit`, `Set as active`, `Delete`. Click `testid=agent-profile-edit`: the title is `Edit agent profile`, the description `Editing profile "QA_acp_custom" - save to apply changes`, the type/preset/command/model values are the stored ones, and Save is disabled. Note the `id` from `api GET /api/agent-profiles`, fill the name with `QA_acp_renamed`, click Save, `control-openhands browser wait-text 'Profile "QA_acp_renamed" updated' --timeout 10000` and `browser reload`: `browser count 'testid=agent-profile-row >> has-text=QA_acp_custom'` is `0`, the renamed row is `1` and `api GET /api/agent-profiles` shows the same id under the new name. Back/Cancel: open any editor, change a field, click `testid=back-to-agent-profiles` (or `testid=cancel-agent-profile-btn`); `browser count 'testid=add-agent-profile'` is `1` and the API is unchanged.
- **Credentials (`F13.acp-credentials`).** In a new ACP editor with preset `Claude Code`, `control-openhands browser testids --hidden --filter acp` lists `settings-acp-secret-CLAUDE_CODE_OAUTH_TOKEN`, `-ANTHROPIC_API_KEY`, `-ANTHROPIC_BASE_URL` and one banner (`settings-acp-auth-detected` when the agent-server machine has a Claude CLI login, as this sandbox does: `You're already signed in to Claude Code — you can leave the fields below blank.`). `control-openhands browser attr 'testid=settings-acp-secret-ANTHROPIC_API_KEY' type` is `password`. Fill `settings-acp-secret-CLAUDE_CODE_OAUTH_TOKEN` with `qa-dummy-oauth` and `settings-acp-secret-ANTHROPIC_API_KEY` with `qa-dummy-key`; `control-openhands browser text 'testid=acp-credential-conflict-warning'` is `Setting ANTHROPIC_API_KEY alongside CLAUDE_CODE_OAUTH_TOKEN breaks its authentication — leave one of them blank.` (screenshot `--feature F13.acp-credentials --name conflict` after `browser scroll 'testid=acp-credential-conflict-warning'`). Cancel; `api GET /api/settings/secrets` gains nothing. Saved credential: new profile `QA_codex`, ACP, preset `Codex`, fill `testid=settings-acp-secret-OPENAI_API_KEY` with `qa-dummy-openai`, Save (`Profile "QA_codex" created`); `api GET /api/settings/secrets` now lists `OPENAI_API_KEY` and the profile has `acp_server` `codex`, `acp_command` `null`, `acp_model` `gpt-5.5`. Reopen it: `control-openhands browser testids --hidden --filter settings-acp-auth` shows `settings-acp-auth-configured` (`Credentials for Codex are configured — you can leave the fields below blank.`) and `control-openhands browser attr 'testid=settings-acp-secret-OPENAI_API_KEY' placeholder` is `Already saved — leave blank to keep` with an empty value.
- **Secrets scope (`F13.secrets-scope`).** In the `QA_codex` editor, `control-openhands browser eval "[...document.querySelectorAll('[data-testid=agent-settings-secret-list] input')].map(i=>i.closest('li').innerText.split('\n')[0]+'='+i.checked+(i.disabled?'(disabled)':''))"` shows every name `=true(disabled)` under `No profile restriction`. Open `testid=agent-settings-secrets-mode`, `control-openhands browser click 'role=option[name="Choose secrets"]'` and rerun the eval: `OPENHANDS_AUTOMATION_API_KEY=false`, `OPENAI_API_KEY=true`, `CODEX_AUTH_JSON=true`, `OPENAI_BASE_URL=true`. Save (`Profile "QA_codex" updated`); `api GET /api/agent-profiles/QA_codex` has those three in `secret_refs`, and after `browser reload` and Edit the mode and switches are the same. On an OpenHands profile, Choose starts with nothing selected; a switch toggles by its name: in the Add editor pick `Choose secrets`, then `control-openhands browser click 'testid=agent-settings-secret-list >> text=OPENHANDS_AUTOMATION_API_KEY'` turns `document.querySelector('[data-testid=agent-settings-secret-OPENHANDS_AUTOMATION_API_KEY]').checked` from `false` to `true`.
- **Stale LLM reference (`F13.stale-llm-ref`).** After `control-openhands llm preset deepseek`, run `control-openhands browser reload`, open `testid=agent-profile-row >> has-text=default >> testid=agent-profile-menu-trigger` → `testid=agent-profile-edit`, `control-openhands browser wait 'testid=agent-profile-llm-selector'`. `browser value 'testid=agent-profile-llm-selector'` is `deepseek-flash (deepseek/deepseek-flash)` and Save is enabled although nothing was touched. Save (`Profile "default" updated`), `browser reload`: the row reads `default` / `deepseek-flash` / `Default` and `api GET /api/agent-profiles/default` has `llm_profile_ref` `deepseek-flash` with the same id. Screenshot `--feature F13.stale-llm-ref --name fallback` before saving.
-- **OpenHands options (`F13.openhands-options`, `F13.create`).** Click `testid=add-agent-profile`, fill the name `QA_oh`. Defaults: `browser value 'testid=agent-profile-llm-selector'` is the active LLM profile, `control-openhands browser eval "document.querySelector('[data-testid=agent-settings-enable-sub-agents]').checked"` is `false`, the same for `agent-settings-enable-switch-llm-tool` is `true`, `browser value 'testid=sdk-settings-tool_concurrency_limit'` is `1`. Click `'testid=agent-settings-screen >> text=Enable sub-agents'` and `'testid=agent-settings-screen >> text=Let the agent switch LLM profiles'` (the checks flip). Fill `testid=sdk-settings-tool_concurrency_limit` with `0` and click Save: `browser toasts` shows `Parallel tool calls must be at least 1` and `browser eval "document.querySelector('[data-testid=sdk-settings-tool_concurrency_limit]').validationMessage"` is `Value must be greater than or equal to 1.` Fill `2`, `control-openhands browser choose 'testid=agent-profile-llm-selector' 'deepseek-pro (deepseek/deepseek-v4-pro)'`, Save (`Profile "QA_oh" created`), `browser reload`: the row reads `QA_oh` / `deepseek-pro`, and `api GET /api/agent-profiles/QA_oh` has `llm_profile_ref` `deepseek-pro`, `enable_sub_agents` `true`, `enable_switch_llm_tool` `false`, `tool_concurrency_limit` `2`. Reopening shows the same values with Save disabled.
+- **OpenHands options (`F13.openhands-options`, `F13.create`).** Click `testid=add-agent-profile`, fill the name `QA_oh`. Defaults: `browser value` on `testid=agent-profile-llm-selector` is the active LLM profile (`deepseek-flash (deepseek/deepseek-flash)`), on `testid=agent-settings-system-prompt-mode` `OpenHands default`, on `testid=agent-settings-tools-mode` `Standard tools` and on `testid=sdk-settings-tool_concurrency_limit` `1`; `control-openhands browser text 'testid=agent-settings-tool-list'` lists `file_editor`, `task_tracker`, `terminal` and `switch_llm` with their descriptions (the list follows the Agent Server; 1.53.0 here). Run `control-openhands browser choose 'testid=agent-settings-tools-mode' 'Choose tools'`; `control-openhands browser eval "[...document.querySelectorAll('[data-testid=agent-settings-tool-list] input')].map(i=>i.dataset.testid.replace('agent-settings-tool-','')+'='+i.checked)"` is `file_editor=true`, `task_tool_set=false`, `task_tracker=true`, `terminal=true`, `glob=false`, `grep=false`, `switch_llm=true`. Click `'testid=agent-settings-tool-list >> text=switch_llm'` and `'testid=agent-settings-tool-list >> text=task_tool_set'` (the eval flips both). Fill `testid=sdk-settings-tool_concurrency_limit` with `0` and click Save: `browser toasts` shows `Parallel tool calls must be at least 1` and `browser eval "document.querySelector('[data-testid=sdk-settings-tool_concurrency_limit]').validationMessage"` is `Value must be greater than or equal to 1.` Fill `2`, `control-openhands browser choose 'testid=agent-profile-llm-selector' 'deepseek-pro (deepseek/deepseek-v4-pro)'`, Save (`Profile "QA_oh" created`), `browser reload`: the row reads `QA_oh` / `deepseek-pro`, and `api GET /api/agent-profiles/QA_oh` has `llm_profile_ref` `deepseek-pro`, `tools` named `file_editor`, `task_tool_set`, `task_tracker`, `terminal`, `enable_classify_and_switch_llm_tool` `false` and `tool_concurrency_limit` `2`. Reopening shows `Choose tools` with the same switches, `2` and `deepseek-pro`, with Save disabled. (There are no **Enable sub-agents** or **Let the agent switch LLM profiles** switches: `browser count 'text=Enable sub-agents'` is `0`.)
- **Set as active (`F13.set-active`).** Run `control-openhands browser click 'testid=agent-profile-row >> has-text=QA_oh >> testid=agent-profile-menu-trigger'`, `control-openhands browser click 'testid=agent-profile-set-active'` and `control-openhands browser wait-text 'Switched to profile "QA_oh"' --timeout 10000`. After `browser reload`, `browser count 'testid=agent-profile-row >> has-text=QA_oh >> testid=agent-profile-active-badge'` and `browser count 'testid=agent-profile-active-badge'` are both `1`. Reopen the QA_oh menu: `browser enabled 'testid=agent-profile-set-active'` is `false` (screenshot `--feature F13.set-active --name menu`); `control-openhands browser press Escape` closes it (`browser count 'testid=agent-profile-actions-menu'` is `0`).
- **Pill wins over a pinned LLM (`F13.llm-pill-precedence`).** With `QA_oh` active (pinned to `deepseek-pro`) and the Home pill on `deepseek-flash` (`browser text 'testid=chat-input-llm-profile'` on `/`), run `control-openhands conversation start --prompt "Reply with only: qa-profile-ok" --wait --timeout 240`. The output's `model` is `deepseek/deepseek-flash`, and `control-openhands api GET /api/conversations/<id>` (id from that output) has `"launched_agent_profile": null` and `tool_concurrency_limit` `1`: the profile was not applied.
- **Active profile drives the launch (`F13.active-drives-conversation`).** Run `control-openhands browser goto /settings/agents` (the previous `conversation start` left the browser on the conversation), edit `QA_oh`, run `control-openhands browser choose 'testid=agent-profile-llm-selector' 'deepseek-flash (deepseek/deepseek-flash)'` (the toggle-button opener failed 5 of 5 times here: see Gotchas), check `browser value 'testid=agent-profile-llm-selector'`, Save (`Profile "QA_oh" updated`). Run the same `conversation start` again; `api GET /api/conversations/<id>` now has `launched_agent_profile.agent_profile_id` equal to QA_oh's id from `api GET /api/agent-profiles` and `tool_concurrency_limit` `2`; `control-openhands conversation events <id> --kinds MessageEvent` contains `qa-profile-ok`.
@@ -78,10 +78,10 @@ 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.
- `Save` is disabled for a duplicate name with no message and no red rule (`aria-invalid` stays `false`) (#17938); check the name before suspecting the form.
-- Switch rows are `<input>`s inside labels (#17900): click the label text (`'testid=agent-settings-screen >> text=Enable sub-agents'`, `'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 `<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.
- 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 sub-agent, 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.
+- 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.
- The ACP auth probe runs the provider CLI on the agent-server machine. On this sandbox the host's Claude CLI is logged in, so Claude Code shows "already signed in". Never launch an ACP conversation on that login.
- ACP credentials are global secrets, not profile fields: cancelling the editor saves nothing, but saving any ACP profile with a typed credential writes it to Settings → Secrets (F14) for every conversation.
- **Choose secrets** on an ACP profile also selects provider secret names that are not saved (`CODEX_AUTH_JSON`, `OPENAI_BASE_URL`); they show as switches without descriptions.
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 62a06b4d52..2ac38b62df 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F14-secrets.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F14-secrets.md
@@ -37,7 +37,7 @@ Preconditions:
- **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`.
- **Create (`F14.create`).** Run `control-openhands browser click 'testid=add-secret-button'`, `control-openhands browser fill 'testid=add-secret-form >> testid=name-input' QA_TMP_SECRET`, `control-openhands browser fill 'testid=add-secret-form >> testid=value-input' dummy-value-123`, `control-openhands browser fill 'testid=add-secret-form >> testid=description-input' 'QA dummy'`, `control-openhands browser click 'testid=add-secret-form >> testid=submit-button'`, then `control-openhands browser reload` and `control-openhands browser count 'testid=secret-item >> has-text=QA_TMP_SECRET'`. The count is `1` after the reload.
-- **Duplicate name (`F14.create-validation`).** Run `control-openhands browser click 'testid=add-secret-button'`, fill `name-input` with `QA_TMP_SECRET` and `value-input` with `other` as above, click `testid=add-secret-form >> testid=submit-button`, then `control-openhands browser snapshot 'testid=add-secret-form'`. The snapshot shows the paragraph `Secret already exists` and the form stays open.
+- **Duplicate name (`F14.create-validation`).** Run `control-openhands browser click 'testid=add-secret-button'`, fill `name-input` with `QA_TMP_SECRET` and `value-input` with `other` as above, click `testid=add-secret-form >> testid=submit-button`, then `control-openhands browser snapshot 'testid=add-secret-form'`. The snapshot shows the paragraph `Secret already exists` and the form stays open. The paragraph uses the theme's danger colour (#18199): `control-openhands browser eval "getComputedStyle([...document.querySelectorAll('[data-testid=add-secret-form] p')].find(e=>e.innerText==='Secret already exists')).color"` is `rgb(231, 106, 94)` on OpenHands-Neutral and `rgb(209, 52, 56)` on Light+ (`F16.color-theme`).
- **Invalid name (`F14.create-validation`).** In the same form run `control-openhands browser fill 'testid=add-secret-form >> testid=name-input' 'bad name!'`, click submit, then `control-openhands browser eval "document.querySelector('[data-testid=add-secret-form] [data-testid=name-input]').validationMessage"`. The value is the browser's pattern message (`Please match the requested format.` in Chromium) and no secret is created (`api GET /api/settings/secrets`). Close with `control-openhands browser click 'testid=add-secret-form >> testid=cancel-button'`.
- **Empty value (`F14.create-validation`).** Run `control-openhands browser click 'testid=add-secret-button'`, fill `testid=add-secret-form >> testid=name-input` with `QA_EMPTY_VALUE` and leave `value-input` empty. `control-openhands browser enabled 'testid=add-secret-form >> testid=submit-button'` is `false`, and `control-openhands browser eval "document.querySelector('[data-testid=add-secret-form] [data-testid=value-input]').validationMessage"` is `Please fill out this field.` Close with `testid=add-secret-form >> testid=cancel-button`; `api GET /api/settings/secrets` lists no `QA_EMPTY_VALUE`.
- **Edit keeps the value (`F14.edit`).** Run `control-openhands browser click 'testid=secret-item >> has-text=QA_TMP_SECRET >> testid=edit-secret-button'`, `control-openhands browser value 'testid=edit-secret-form >> testid=value-input'` (empty) and `control-openhands browser enabled 'testid=edit-secret-form >> testid=submit-button'` (`false`). Fill `testid=edit-secret-form >> testid=description-input` with `QA dummy edited`; `enabled` turns `true`. Click `testid=edit-secret-form >> testid=submit-button`, `control-openhands browser reload`, then `control-openhands browser text 'testid=secret-item >> has-text=QA_TMP_SECRET'`. The row text is `QA_TMP_SECRET` and `QA dummy edited` separated by a tab (`"QA_TMP_SECRET\tQA dummy edited"` in the JSON output).
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 9b6cdc3de5..666aa8b1ff 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
@@ -8,7 +8,7 @@ page mixes two save models: language, analytics, sound, title model and git
identity wait for **Save Changes** and are stored on the backend; theme,
checklist and voice input apply instantly and live in this browser only.
-Source: `src/routes/app-settings.tsx`, `src/components/features/settings/app-settings/`, `src/components/features/settings/settings-switch.tsx`, `src/themes/color-theme/`, `src/utils/transcription-endpoint-storage.ts`, `src/api/settings-service/settings-service.api.ts`.
+Source: `src/routes/app-settings.tsx`, `src/components/features/settings/app-settings/`, `src/components/features/settings/settings-switch.tsx`, `src/themes/color-theme/`, `src/utils/transcription-endpoint-storage.ts`, `src/api/settings-service/settings-service.api.ts`, `src/services/telemetry.ts`.
## Sub-features
diff --git a/.agents/skills/verify-openhands/references/feature-map/F17-mcp-servers.md b/.agents/skills/verify-openhands/references/feature-map/F17-mcp-servers.md
index 01c992287a..b5202fdb0e 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F17-mcp-servers.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F17-mcp-servers.md
@@ -9,7 +9,7 @@ test connections, check each installed server's health, enable or disable,
edit and delete servers. Enabled servers are offered to every new
conversation's agent.
-Source: `src/routes/extensions-hub.tsx`, `src/routes/mcp.tsx`, `src/components/features/skills/extensions-navigation.tsx`, `src/components/features/skills/extensions-mobile-hub.tsx`, `src/components/features/mcp-page/`, `src/components/features/settings/mcp-settings/mcp-server-form.tsx`, `src/utils/mcp-marketplace-utils.ts`, `src/api/mcp-health/`, `src/utils/mobile-section-nav.ts`.
+Source: `src/routes/extensions-hub.tsx`, `src/routes/mcp.tsx`, `src/components/features/skills/extensions-navigation.tsx`, `src/components/features/skills/extensions-mobile-hub.tsx`, `src/components/features/mcp-page/`, `src/components/features/settings/mcp-settings/mcp-server-form.tsx`, `src/utils/mcp-marketplace-utils.ts`, `src/api/mcp-health/`, `src/hooks/mutation/use-add-mcp-server.ts`, `src/utils/mobile-section-nav.ts`.
## Sub-features
@@ -99,7 +99,7 @@ Preconditions:
## Gotchas
-- Server errors on MCP mutations (a 404 on a server that is already removed, a 502 while the Agent Server is down) toast the server's message (`MCP server '<id>' was not found`), or `An error occurred` when the body has none (#18051). Expected: one toast per failed action. Known failure (reproduced 2026-10-08): a failed delete from the editor or a failed card toggle shows two identical toasts, one from the caller and one from the global mutation handler (#18181). Count them with `click ... --observe 'role=status'` or `browser toasts --history`.
+- Server errors on MCP mutations (a 404 on a server that is already removed, a 502 while the Agent Server is down) toast the server's message (`MCP server '<id>' was not found`), or `An error occurred` when the body has none (#18051). Expected: one toast per failed action. Known failure (reproduced 2026-10-09): a failed delete from the editor or a failed card toggle shows two identical toasts, one from the caller and one from the global mutation handler (#18181). Count them with `click ... --observe 'role=status'` or `browser toasts --history`.
- Desktop `/customize` is a client-side redirect: `browser url` may still read `/customize` for a moment; use `wait-url '/mcp(\?|$)'`.
- From 768 to 1023 px the sidebar rail stays and the top bar is gone, so a Customize detail page shows neither the aside nor a Back chevron; the rail's **Customize** link (`sidebar-skills-link`, unscoped at this width) returns to the hub.
- On a phone the drawer and the hidden desktop rail both render `sidebar-skills-link`; scope it with `testid=sidebar-mobile-drawer >> ...` or the click fails with a strict-mode error, even though `browser testids` lists it once.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F19-plugins.md b/.agents/skills/verify-openhands/references/feature-map/F19-plugins.md
index b7f50787fc..d42026f99e 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F19-plugins.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F19-plugins.md
@@ -26,7 +26,7 @@ Source: `src/routes/skills-plugins.tsx`, `src/routes/launch.tsx`, `src/component
- `F19.uninstall`: **Uninstall** removes the plugin at once (no confirmation), shows `Plugin uninstalled.`, closes the modal and returns the card to **Install**.
- `F19.add-plugin`: **Add plugin** opens **Add a plugin** with Source (required), Ref and Path (optional); Install stays disabled until Source has text; a valid source installs and closes the modal.
- `F19.add-plugin-cancel`: the Add modal's X, **Close**, Escape and a backdrop click close it without installing; reopening starts with empty fields.
-- `F19.add-plugin-error`: an unusable source shows one error toast with the server's reason and keeps the modal open with the typed values.
+- `F19.add-plugin-error`: an unusable source shows one error toast with the server's reason and keeps the modal open with the typed values; a plugin that is already installed (409) toasts `This plugin is already installed. Open its card to Update or Uninstall it.` instead (#18110).
- `F19.installed-coordinates`: a plugin installed with a source, ref and path shows that source on its card and in the modal header, and modal pills `v<version>`, `Ref <resolved commit>` and `Path <repo path>` next to the Enabled switch.
- `F19.local-plugin`: a plugin found in `~/.agents/plugins` shows a **Local** badge (no Install or switch), is listed under the Local filter, and its modal has a Local pill, Files, no Install/Uninstall and no Start Conversation.
- `F19.start-conversation`: **Start Conversation** in the detail modal opens `/launch?plugins=<base64 JSON>` for that plugin.
@@ -59,7 +59,7 @@ Preconditions:
- Right before the Local plugin bullet (not earlier: it changes card counts and the Local filter), put a local plugin in the Agent Server's `~/.agents/plugins`. No CLI verb does this (harness gap); from the checkout run `H="$OH_VERIFY_RUN/private/home/.agents/plugins"; mkdir -p "$H" && cp -r "$OH_VERIFY_RUN/private/cache/skills/public-skills/plugins/magic-test" "$H/qa-local-plugin" && sed -i 's/"name": "magic-test"/"name": "qa-local-plugin"/' "$H/qa-local-plugin/.plugin/plugin.json"`. The rename matters: a local plugin whose name is already in the catalog is hidden.
- The Add plugin bullet installs from `github:OpenHands/extensions`, which needs outbound access to GitHub.
-- **Open the page (`F19.page`).** Run `control-openhands browser goto /conversations`, `control-openhands browser click 'testid=sidebar-skills-link' --expect-url '/mcp(\?|$)'`, `control-openhands browser click 'testid=sidebar-extensions-/plugins' --expect-url '/plugins(\?|$)'`, then `control-openhands browser text 'testid=skills-plugins-screen >> h2'` (`Plugins`), `control-openhands browser text 'testid=plugins-settings-description'` (the description above), `control-openhands browser enabled 'testid=plugins-add-plugin-button'` (`true`) and `control-openhands browser count '[data-testid^="plugin-card-"]'` (9 with today's catalog). `control-openhands browser screenshot --feature F19.page --name catalog` shows two columns of cards, each with name, truncated source path, a two-line description and **Install**.
+- **Open the page (`F19.page`).** Run `control-openhands browser goto /conversations`, `control-openhands browser click 'testid=sidebar-skills-link' --expect-url '/mcp(\?|$)'`, `control-openhands browser click 'testid=sidebar-extensions-/plugins' --expect-url '/plugins(\?|$)'`, then `control-openhands browser text 'testid=skills-plugins-screen >> h2'` (`Plugins`), `control-openhands browser text 'testid=plugins-settings-description'` (the description above), `control-openhands browser enabled 'testid=plugins-add-plugin-button'` (`true`), `control-openhands browser wait '[data-testid^="plugin-card-"]' --timeout 60000` (the catalog loads after the page: an immediate count can read `0`) and `control-openhands browser count '[data-testid^="plugin-card-"]'` (9 with today's catalog). `control-openhands browser screenshot --feature F19.page --name catalog` shows two columns of cards, each with name, truncated source path, a two-line description and **Install**.
- **Command-menu entry (`F19.page`).** From `/conversations` run `control-openhands browser press Control+k`, `control-openhands browser type 'testid=command-menu >> role=combobox' Plugins`, `control-openhands browser snapshot 'testid=command-menu'` (one option, `Customize Browse skills, plugins, and integrations. Go`), `control-openhands browser press Enter`, `control-openhands browser wait-url '/mcp(\?|$)'`, then `control-openhands browser click 'testid=sidebar-extensions-/plugins' --expect-url '/plugins(\?|$)'`.
- **Search (`F19.search`).** Run `control-openhands browser fill 'testid=plugins-search-input' weather` and `control-openhands browser count '[data-testid^="plugin-card-"]'` (`1`). `control-openhands browser fill 'testid=plugins-search-input' alakazam` also leaves `1` (`magic-test`, matched on its description). `control-openhands browser fill 'testid=plugins-search-input' zzz-none`, then `control-openhands browser text 'testid=plugins-no-match'`: `No plugins match your search.` Run `control-openhands browser click 'testid=plugins-toolbar >> role=button[name="Clear search"]'`; `control-openhands browser value 'testid=plugins-search-input'` is empty and the count is back to 9.
- **Status filter (`F19.status-filter`).** `control-openhands browser attr 'testid=plugins-filter-all' aria-pressed` is `true`. Run `control-openhands browser click 'testid=plugins-filter-installed'`: its `aria-pressed` is `true`, All's is `false`, and with nothing installed `control-openhands browser count 'testid=plugins-no-match'` is `1`. `control-openhands browser click 'testid=plugins-filter-available'` shows all 9 cards; `control-openhands browser click 'testid=plugins-filter-local'` shows `plugins-no-match` (no local plugin yet). After `control-openhands browser reload`, `plugins-filter-all` is pressed again.
@@ -68,7 +68,7 @@ Preconditions:
- **Install from a card (`F19.install-card`).** Run `control-openhands browser toasts --clear`, `control-openhands browser click 'testid=plugin-install-magic-test'`, `control-openhands browser wait 'testid=plugin-toggle-magic-test' --timeout 60000` and `control-openhands browser toasts` (`Plugin installed.`). `control-openhands browser text 'testid=plugin-version-magic-test'` is `v1.0.0` and `control-openhands browser eval "[...document.querySelectorAll('[data-testid^=plugin-card-]')].map(e=>e.dataset.testid)[0]"` is `plugin-card-magic-test`. After `control-openhands browser reload` and `control-openhands browser click 'testid=plugins-filter-installed'`, `browser count '[data-testid^="plugin-card-"]'` is `1`, and `control-openhands api GET /api/plugins/installed` lists `magic-test` with `"enabled": true`. Click `testid=plugins-filter-all`.
- **Enable switch (`F19.enable-toggle`).** Run `control-openhands browser attr 'testid=plugin-toggle-magic-test' aria-checked` (`true`) and `control-openhands browser attr 'testid=plugin-toggle-magic-test' aria-label` (`Disable plugin`), `control-openhands browser hover 'testid=plugin-toggle-magic-test'`, `control-openhands browser click 'testid=plugin-toggle-magic-test'`, wait three seconds, then `browser attr ... aria-checked` again (`false`, with `aria-label` `Enable plugin`); after `control-openhands browser reload` it is still `false` and `api GET /api/plugins/installed` shows `"enabled": false`. Today the disable is often lost: the reload's concurrent reads hit the metadata race (Gotchas) and the API, then the switch, read `"enabled": true` again (2 of 3 trials; the PATCH itself returns 200). If that happens, disable it once more before the modal step, which needs the plugin disabled. In the modal: `control-openhands browser click 'testid=plugin-card-magic-test'`, `control-openhands browser click 'testid=plugin-detail-modal >> text="Disabled"'`, then `control-openhands browser eval "document.querySelector('[data-testid=plugin-modal-toggle-magic-test]').checked"` (`true`); the API shows `"enabled": true` and, after closing the modal, the card switch's `aria-checked` is `true`.
- **Enabled plugins auto-load (`F19.enabled-autoload`).** With `magic-test` enabled, run `control-openhands conversation start --prompt "alakazam" --wait --timeout 240` and `control-openhands browser text 'testid=agent-message >> nth=-1'`: `The magic worked! Plugin loaded successfully. 🎩✨`. Then `control-openhands browser goto /plugins`, disable it with the card switch (hover, click; API `"enabled": false`) and run the same `conversation start`: the reply is an ordinary request for a task, without the phrase. `browser goto /plugins` again; `magic-test` stays installed and disabled. Today the revisit often re-enables it (API `"enabled": true`, `installed_at` reset; see Gotchas), so re-check the API before relying on the disabled state.
-- **Update (`F19.refresh`).** Run `control-openhands browser click 'testid=plugin-card-magic-test'`, `control-openhands browser toasts --clear`, `control-openhands browser click 'testid=plugin-detail-refresh-magic-test'`, wait a few seconds, then `control-openhands browser toasts`. Expected `Plugin updated.` (the toast fades fast: read `control-openhands browser toasts --history`). Whether it works depends on the recorded source (`control-openhands browser text 'testid=plugin-source-magic-test'`): with the catalog path it does. Known failure (reproduced 2026-10-06): a page revisit after a catalog install usually rewrites the source to `local` (OpenHands/software-agent-sdk#5496, see Gotchas), and Update then answers 500 with one toast carrying the server's reason, `Unable to parse extension source: local. Expected formats: 'github:owner/repo', git URL, or local path` (`browser screenshot --feature F19.refresh --name local-source-error`). On a plugin whose source survived (the GitHub install in **Add plugin from GitHub**, `browser text 'testid=plugin-source-magic-test'` = `github:OpenHands/extensions`), the same click logs `Updating extension magic-test from github:OpenHands/extensions` and `toasts --history` shows `Plugin updated.` Close the modal.
+- **Update (`F19.refresh`).** Run `control-openhands browser click 'testid=plugin-card-magic-test'`, `control-openhands browser toasts --clear`, `control-openhands browser click 'testid=plugin-detail-refresh-magic-test'`, wait a few seconds, then `control-openhands browser toasts`. Expected `Plugin updated.` (the toast fades fast: read `control-openhands browser toasts --history`). Whether it works depends on the recorded source (`control-openhands browser text 'testid=plugin-source-magic-test'`): with the catalog path it does. Known failure (reproduced 2026-10-09): a page revisit after a catalog install usually rewrites the source to `local` (OpenHands/software-agent-sdk#5496, see Gotchas), and Update then answers 500 with one toast carrying the server's reason, `Unable to parse extension source: local. Expected formats: 'github:owner/repo', git URL, or local path` (`browser screenshot --feature F19.refresh --name local-source-error`). On a plugin whose source survived (the GitHub install in **Add plugin from GitHub**, `browser text 'testid=plugin-source-magic-test'` = `github:OpenHands/extensions`), the same click logs `Updating extension magic-test from github:OpenHands/extensions` and `toasts --history` shows `Plugin updated.` Close the modal.
- **Start a conversation from an installed plugin (`F19.start-conversation`).** Run `control-openhands browser click 'testid=plugin-card-magic-test'`, `control-openhands browser click 'testid=plugin-detail-start-conversation-magic-test' --expect-url '/launch\?plugins='` and `control-openhands browser eval "JSON.parse(atob(new URLSearchParams(location.search).get('plugins')))"`. Expected: the plugin's real source. Today it is `[{"source":"local","ref":null,"repo_path":null}]` and the modal title is `Launch local`; after `control-openhands browser check 'testid=trust-checkbox'` and `control-openhands browser click 'testid=start-conversation-button' --expect-url '/conversations/[0-9a-f-]+'`, `control-openhands browser type 'testid=chat-input' 'alakazam'` and `control-openhands browser press Enter --selector 'testid=chat-input'` leave the message at `Sending...` (still after 25 s) with the banner `Unable to parse plugin source: local. Expected formats: 'github:owner/repo', git URL, or local path` (`control-openhands browser screenshot --feature F19.start-conversation --name local-source-conversation`). When the installed plugin kept its source (the GitHub install below), the link carries it: `[{"source":"github:OpenHands/extensions","ref":"<resolved commit>","repo_path":"plugins/magic-test"}]` and the title is `Launch magic-test`.
- **Uninstall (`F19.uninstall`).** Run `control-openhands browser goto /plugins`, `control-openhands browser toasts --clear`, `control-openhands browser click 'testid=plugin-card-magic-test'`, `control-openhands browser click 'testid=plugin-detail-uninstall-magic-test'`, then `control-openhands browser count 'testid=plugin-detail-modal'` (`0`), `control-openhands browser count 'testid=confirmation-modal'` (`0`: no confirmation), `control-openhands browser toasts` (`Plugin uninstalled.`) and `control-openhands browser count 'testid=plugin-install-magic-test'` (`1`). After `control-openhands browser reload` the Install button is still there, `browser text 'testid=plugin-source-magic-test'` is the catalog path again and `api GET /api/plugins/installed` returns `{"plugins": []}`.
- **Install from the modal (`F19.install-modal`).** Run `control-openhands browser toasts --clear`, `control-openhands browser click 'testid=plugin-card-city-weather'`, `control-openhands browser click 'testid=plugin-detail-install-city-weather'`, `control-openhands browser wait 'testid=plugin-detail-uninstall-city-weather' --timeout 60000` and `control-openhands browser toasts` (`Plugin installed.`). The modal stays open (`browser count 'testid=plugin-detail-modal'` is `1`), `control-openhands browser eval "document.querySelector('[data-testid=plugin-modal-toggle-city-weather]').checked"` is `true`, and `control-openhands browser snapshot 'testid=plugin-detail-modal >> footer'` lists **Update**, **Uninstall**, **Start Conversation**, **Close**. Clean up with `control-openhands browser click 'testid=plugin-detail-uninstall-city-weather'`; `browser count 'testid=plugin-install-city-weather'` is `1`.
@@ -92,7 +92,7 @@ Preconditions:
- The card switch swaps its icon on hover. A plain `browser click` that moves the pointer onto it from elsewhere (for example right after clicking a filter segment) is often swallowed (#17941): no `PATCH /api/plugins/installed/<name>` shows in `browser network`. Run `browser hover` on the switch first, then click, and assert `aria-checked` and the API, not the click result.
- Installed plugins lose their source and their enabled state, intermittently (OpenHands/software-agent-sdk#5496). The Agent Server rewrites `<run>/private/plugins/installed/.installed.json` with a plain truncate-and-write on every load (SDK `InstallationMetadata.save_to_dir`), so a concurrent `GET /api/plugins/installed` (page loads fire several) can read an empty file: it logs `Failed to load installed extension metadata … Invalid JSON: EOF while parsing` and `Discovered untracked extension`, then saves every plugin with `"source": "local"`, `"enabled": true` and a fresh `installed_at`. It hit catalog installs in most trials and a disabled plugin on 2 of 3 page revisits; GitHub installs kept their source in this run. Once it hits, the card shows `local`, **Update** fails with a 500, **Start Conversation** builds an unusable `/launch` link, and a disabled plugin silently auto-loads again. Prove the launch path from a card that is not installed, and re-read the API after every navigation.
- A failed plugin install, update or uninstall shows one toast with the server's reason (enable uses the same handler; its failure is not driven here). A response with no reason in it, such as the ingress's 502 while the Agent Server is stopped, shows the generic `An error occurred`; read the cause in the run's `private/stack.log`.
-- Installing a plugin whose name is already installed should point the user to **Update** or **Uninstall**, as Apps do (`F20.install-duplicate`). Known failure (reproduced 2026-10-06): with the catalog `magic-test` installed, **Add plugin** with `github:OpenHands/extensions` and Path `plugins/magic-test` answers 409 and toasts the server's `Plugin already installed. Use force=true to overwrite.`, which the form cannot do (`browser screenshot --feature F19.add-plugin-error --name duplicate-409`) (#18090).
+- Installing a plugin whose name is already installed points the user to **Update** or **Uninstall**, as Apps do (`F20.install-duplicate`): with the catalog `magic-test` installed, **Add plugin** with `github:OpenHands/extensions` and Path `plugins/magic-test` answers 409 (`browser network --filter 'plugins/install'`) and `browser toasts --history` holds one toast, `This plugin is already installed. Open its card to Update or Uninstall it.`, instead of the server's `force=true` hint; the modal stays open with the typed values (`browser screenshot --feature F19.add-plugin-error --name duplicate-409`). Other install errors still toast the server's reason (`F19.add-plugin-error`). The 409 adds one `http-error` and one `console.error` to `browser errors --app-only`.
- **Uninstall** acts immediately, with no confirmation (MCP server delete and App uninstall confirm).
- `/launch` with a `message` starts the agent as soon as **Start Conversation** is clicked. Prefer links without a message and send `alakazam` yourself; a `city-weather` launch with a message makes the agent call an external weather API (stop it with `testid=stop-button`).
- Edited launch parameters cannot be observed after creation: the conversation's client metadata (`localStorage['openhands-agent-server-conversation-metadata']`) strips them and the Agent Server does not return a conversation's plugins. Proving delivery needs request-body capture (harness gap).
diff --git a/.agents/skills/verify-openhands/references/feature-map/F20-canvas-apps.md b/.agents/skills/verify-openhands/references/feature-map/F20-canvas-apps.md
index 6be72840fc..aa5bfc173e 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F20-canvas-apps.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F20-canvas-apps.md
@@ -90,7 +90,7 @@ Preconditions:
- The page states (empty, error, unsupported) and the `/extensions/...` unavailable card have no test ids; assert on text with `browser snapshot`.
- The Agent Server fetches Git sources with a shallow clone and appends `.git` to the repo URL: a test Git server must speak smart HTTP and answer both `qa-repo` and `qa-repo.git`.
- The Git bullets read the public upstream repo, so they depend on its `main` still holding the identical fixture; the precondition's `diff -r` guards that. A Git install or Update clones in about 0.4 to 2 s here (Updates measured about 0.4 to 1.5 s on 2026-10-06), long enough for the busy-lock checks when they run in the same shell line as the click and read both controls in one `eval`; on a slower network raise the `wait` timeouts, not the expectations.
-- A duplicate install answers `409` with the Agent Server's API hint (`Canvas extension already installed. Use force=true to overwrite.`); the form has no force option, so the toast replaces it with the Update/Uninstall advice. Plugins still show their 409 hint verbatim (F19 Gotchas).
+- A duplicate install answers `409` with the Agent Server's API hint (`Canvas extension already installed. Use force=true to overwrite.`); the form has no force option, so the toast replaces it with the Update/Uninstall advice. Plugins give the same advice since #18110 (F19 Gotchas).
- Busy states with a local-path source last about 50 ms: `--observe` catches `Installing…`, but checking that controls are disabled during Update needs a Git source (an upstream Update lasts about 0.4 to 2 s).
- `control-openhands launch` passes the launcher only a fixed list of environment variables (path, locale, proxy) plus its own flags: `OH_AGENT_SERVER_VERSION=<old>` in the environment is ignored (the run still gets the pinned server); use `--sdk-ref v<version>`. Any server older than 1.47.0 is gated by Canvas before `/apps` renders.
- Stopping the Agent Server makes the next full page load show the **Manage backends** gate; always `restart` and `doctor` afterwards.
diff --git a/.agents/skills/verify-openhands/references/feature-map/F24-git-sync.md b/.agents/skills/verify-openhands/references/feature-map/F24-git-sync.md
index 7db86c6484..2f2b0e0fe1 100644
--- a/.agents/skills/verify-openhands/references/feature-map/F24-git-sync.md
+++ b/.agents/skills/verify-openhands/references/feature-map/F24-git-sync.md
@@ -61,10 +61,10 @@ Preconditions:
- **Dirty tracking (`F24.form-dirty`).** `control-openhands browser enabled 'testid=git-sync-save-button'` and `enabled 'testid=git-sync-save-and-sync-button'` are both `false`. Run `control-openhands browser fill 'testid=git-sync-interval-input' 300`: Save is `true`, Save and sync stays `false` (sync is off). `control-openhands browser fill 'testid=git-sync-interval-input' 0`: Save is `false` again. Run `control-openhands browser click 'testid=git-sync-enabled-switch >> xpath=ancestor::label'`; `control-openhands browser eval "document.querySelector('[data-testid=git-sync-enabled-switch]').checked"` is `true` and both buttons are `true`. Click the label again; Save is `false`.
- **Unreachable repository (`F24.check-failure`).** Run `control-openhands browser fill 'testid=git-sync-repo-url-input' /nonexistent/qa-missing.git` and `control-openhands browser click 'testid=git-sync-save-button' --observe 'testid=git-sync-save-button' --observe-ms 3000`; the observation includes `Checking repository...`. Run `control-openhands browser wait 'testid=git-sync-check-failure' --timeout 25000` and `control-openhands browser text 'testid=git-sync-check-failure'`: `Could not reach the repository with these settings`, git's `fatal: '/nonexistent/qa-missing.git' does not appear to be a git repository`, then `Fix the settings above, or press Save again to store them anyway.` Take `control-openhands browser screenshot 'testid=git-sync-check-failure' --feature F24.check-failure --name failure-block`. `control-openhands api GET /api/automation/v1/git-sync/status` still has `"repo_url": ""`. Click `testid=git-sync-save-button` again, then `control-openhands browser wait-text 'Git Sync settings saved.' --timeout 10000`; `browser count 'testid=git-sync-check-failure'` is `0`. After `control-openhands browser reload`, `control-openhands browser text 'testid=git-sync-enabled-pill'` is `Enabled` (configuring a repo turns sync on) and the card shows `/nonexistent/qa-missing.git` as plain text (`browser count 'testid=git-sync-repo-link'` is `0`).
- **Failing cycle (`F24.sync-failure`).** Run `control-openhands browser click 'testid=git-sync-now-button' --observe 'testid=git-sync-activity-row' --observe-ms 8000`. The observation goes `Syncing...` / `started 0s ago` (counting up each second) and ends `Sync failed`; `control-openhands browser attr 'testid=git-sync-activity-row' data-state` is `failed`. Run `control-openhands browser wait 'testid=git-sync-error-banner' --timeout 20000` and `control-openhands browser text 'testid=git-sync-error-banner'`: `Last sync error`, `git command failed (128): git clone --origin origin /nonexistent/qa-missing.git .`, `fatal: repository '/nonexistent/qa-missing.git' does not exist`, `Ns ago`. Screenshot with `control-openhands browser screenshot --feature F24.sync-failure --name failed`. After `control-openhands browser reload` the banner count is `1` and the activity row count is `0` (the row is page-local).
-- **Configure and Save and sync (`F24.save-and-sync`).** Run `control-openhands browser fill 'testid=git-sync-repo-url-input' "$OH_VERIFY_RUN/workspace/qa-sync-remote"`, `control-openhands browser fill 'testid=git-sync-branch-input' qa-sync`, `control-openhands browser fill 'testid=git-sync-path-input' qa-automations`, then `control-openhands browser click 'testid=git-sync-save-and-sync-button' --observe 'testid=git-sync-activity-row' --observe-ms 10000`. The observation goes `Syncing... started 0s ago` → `Sync complete` within a few seconds; `browser count 'testid=git-sync-check-failure'` and `browser count 'testid=git-sync-error-banner'` are `0` (a successful cycle clears the last error). The toast `Git Sync settings saved.` also fires, but it is gone before a 10 s observation ends; to see it, run `control-openhands browser wait-text 'Git Sync settings saved.' --timeout 5000` right after the click instead of `--observe`. The card text (as in Unconfigured card) shows `<repo-path>`, `qa-sync`, `qa-automations`, a 7-character commit and `Last synced / Ns ago`. `git -C <repo-path> log --oneline qa-sync` shows the fixture's `Initial fixture commit` (the cycle created the branch).
+- **Configure and Save and sync (`F24.save-and-sync`).** **Save and sync now** stays disabled until **Enable Git Sync** is on (`F24.check-failure` turns it on; on a fresh stack turn it on first). Run `control-openhands browser fill 'testid=git-sync-repo-url-input' "$OH_VERIFY_RUN/workspace/qa-sync-remote"`, `control-openhands browser fill 'testid=git-sync-branch-input' qa-sync`, `control-openhands browser fill 'testid=git-sync-path-input' qa-automations`, then `control-openhands browser click 'testid=git-sync-save-and-sync-button' --observe 'testid=git-sync-activity-row' --observe-ms 10000`. The observation goes `Syncing... started 0s ago` → `Sync complete` within a few seconds; `browser count 'testid=git-sync-check-failure'` and `browser count 'testid=git-sync-error-banner'` are `0` (a successful cycle clears the last error). The toast `Git Sync settings saved.` also fires, but it is gone before a 10 s observation ends; to see it, run `control-openhands browser wait-text 'Git Sync settings saved.' --timeout 5000` right after the click instead of `--observe`. The card text (as in Unconfigured card) shows `<repo-path>`, `qa-sync`, `qa-automations`, a 7-character commit and `Last synced / Ns ago`. `git -C <repo-path> log --oneline qa-sync` shows the fixture's `Initial fixture commit` (the cycle created the branch).
- **Sync a changed automation (`F24.sync-now`).** Create `QA_F24 Sync` (Preconditions), `control-openhands browser reload`, then `control-openhands browser text 'text=Pending changes >> xpath=ancestor::div[1]/..'` (`Pending changes / 1`) and `control-openhands browser attr 'text=Pending changes >> xpath=ancestor::div[1]/.. >> span.text-warning' class` (`text-warning`). Screenshot with `control-openhands browser screenshot --feature F24.sync-now --name pending`. Run `control-openhands browser click 'testid=git-sync-now-button' --observe 'testid=git-sync-activity-row' --observe-ms 8000`: `Syncing... / started 0s ago / 1 pending`, then `control-openhands browser text 'testid=git-sync-activity-row'` is `Sync complete` and Pending changes reads `0`. `git -C <repo-path> log --oneline qa-sync` has a new `Sync automations from agent server` commit whose short hash matches the card's Last synced commit, and `git -C <repo-path> ls-tree -r --name-only qa-sync` lists `qa-automations/qa-f24-sync/automation.yaml` and `qa-automations/qa-f24-sync/tarball/...`.
- **Commit author (`F24.author`).** Run `control-openhands browser fill 'testid=git-sync-author-email-input' not-an-email`, `control-openhands browser network --clear`, `control-openhands browser click 'testid=git-sync-save-button'`, then `control-openhands browser eval "document.querySelector('[data-testid=git-sync-author-email-input]').validationMessage"` (`Please include an '@' in the email address. 'not-an-email' is missing an '@'.`); `control-openhands browser network` lists no `git-sync/config` request. Fill `testid=git-sync-author-name-input` with `'QA F24 Bot'` and `testid=git-sync-author-email-input` with `[email protected]`, click Save, `control-openhands browser wait-text 'Git Sync settings saved.' --timeout 10000`, `control-openhands browser reload`; `control-openhands browser value 'testid=git-sync-author-name-input'` is `""` (the fields never show the stored author). Mark the automation changed (PATCH `{"enabled":false}`), `control-openhands browser reload`, click `testid=git-sync-now-button`, `control-openhands browser wait '[data-testid=git-sync-activity-row]:not([data-state=running])' --timeout 30000`; then `git -C <repo-path> log -1 --format='%an <%ae>' qa-sync` is `QA F24 Bot <[email protected]>`.
-- **Encryption key (`F24.encryption`).** `control-openhands browser attr 'testid=git-sync-encryption-key-input' placeholder` is `No encryption key set`. Fill it with `qa-f24-dummy-key`, click Save (no `Checking repository...`: the key is not checked), wait for `Git Sync settings saved.`, `control-openhands browser reload`: `browser text 'testid=git-sync-encryption-pill'` is `Encrypted`, the placeholder is `An encryption key is currently set` and the input value is `""`. Before changing the automation, run `control-openhands api GET /api/automation/v1/git-sync/status`, record `git -C <repo-path> rev-parse qa-sync`, click **Sync now** and wait as in Commit author, then read the head again and `git -C <repo-path> show qa-sync:qa-automations/qa-f24-sync/automation.yaml | head -c 20`. Expected: the key save alone causes a new commit and ciphertext (`gAAAAA`); known failure #551: `dirty_count` is `0`, the head is unchanged and the export stays plaintext. Record this outcome before any PATCH. Continue the separate dirty-export check: mark the automation changed (PATCH `{"enabled":true}` if the Commit author bullet left it disabled, otherwise `{"enabled":false}`; `api GET /api/automation/v1/git-sync/status` must show `dirty_count` ≥ 1), reload, click **Sync now** and wait as in Commit author; the same git read starts with `gAAAAA`. This PATCH arranges a dirty automation; its passing export does not prove the key-save re-export contract. Record the encrypted head, fill the key input with a different dummy key, Save, wait for the toast, reload and **Sync now** without changing the automation. Expected: the head and ciphertext change for the new key; known failure #551: `dirty_count` stays `0` and the head and ciphertext remain unchanged. If that failure occurs, arrange ciphertext under the new key with a separate enabled-state PATCH and sync before the clear test; do not count that repair as a key-rotation pass. Then fill the key input with any text, run `control-openhands browser click 'testid=git-sync-clear-encryption-key-switch >> xpath=ancestor::label'`: `control-openhands browser enabled 'testid=git-sync-encryption-key-input'` is `false` and its value is `""`. Click Save, wait for the toast, reload: the pill is `Not encrypted`, the placeholder `No encryption key set`, and the clear switch is unchecked again. Record the encrypted head, read `git-sync/status`, click **Sync now** and wait without changing the automation, then compare the head and exported file. Expected: a new commit contains plaintext; known failure #551: `dirty_count` is `0`, the head is unchanged and ciphertext remains despite the `Not encrypted` pill. Repeat the non-dirty set, rotate and clear checks at `browser viewport phone`, arranging plaintext before set and ciphertext before rotate/clear; record desktop and phone separately with the actual selected backend in each evidence entry.
+- **Encryption key (`F24.encryption`).** `control-openhands browser attr 'testid=git-sync-encryption-key-input' placeholder` is `No encryption key set`. Fill it with `qa-f24-dummy-key`, click Save (no `Checking repository...`: the key is not checked), wait for `Git Sync settings saved.`, `control-openhands browser reload`: `browser text 'testid=git-sync-encryption-pill'` is `Encrypted`, the placeholder is `An encryption key is currently set` and the input value is `""`. Before changing the automation, run `control-openhands api GET /api/automation/v1/git-sync/status`, record `git -C <repo-path> rev-parse qa-sync`, click **Sync now** and wait as in Commit author, then read the head again and `git -C <repo-path> show qa-sync:qa-automations/qa-f24-sync/automation.yaml | head -c 20`. Expected: the key save alone causes a new commit and ciphertext (`gAAAAA`); known failure on automation 1.19.0 (reproduced 2026-10-09, #551, see Gotchas; passes on automation `main`): `dirty_count` is `0`, the head is unchanged and the export stays plaintext. Record this outcome before any PATCH. Continue the separate dirty-export check: mark the automation changed (PATCH `{"enabled":true}` if the Commit author bullet left it disabled, otherwise `{"enabled":false}`; `api GET /api/automation/v1/git-sync/status` must show `dirty_count` ≥ 1), reload, click **Sync now** and wait as in Commit author; the same git read starts with `gAAAAA`. This PATCH arranges a dirty automation; its passing export does not prove the key-save re-export contract. Record the encrypted head, fill the key input with a different dummy key, Save, wait for the toast, reload and **Sync now** without changing the automation. Expected: the head and ciphertext change for the new key; known failure #551: `dirty_count` stays `0` and the head and ciphertext remain unchanged. If that failure occurs, arrange ciphertext under the new key with a separate enabled-state PATCH and sync before the clear test; do not count that repair as a key-rotation pass. Then fill the key input with any text, run `control-openhands browser click 'testid=git-sync-clear-encryption-key-switch >> xpath=ancestor::label'`: `control-openhands browser enabled 'testid=git-sync-encryption-key-input'` is `false` and its value is `""`. Click Save, wait for the toast, reload: the pill is `Not encrypted`, the placeholder `No encryption key set`, and the clear switch is unchecked again. Record the encrypted head, read `git-sync/status`, click **Sync now** and wait without changing the automation, then compare the head and exported file. Expected: a new commit contains plaintext; known failure #551: `dirty_count` is `0`, the head is unchanged and ciphertext remains despite the `Not encrypted` pill. Repeat the non-dirty set, rotate and clear checks at `browser viewport phone`, arranging plaintext before set and ciphertext before rotate/clear; record desktop and phone separately with the actual selected backend in each evidence entry.
- **Access token (`F24.token`).** `control-openhands browser attr 'testid=git-sync-token-input' type` is `password` and the placeholder `Leave blank to keep the current token`. Fill it with `qa-dummy-token`, run `control-openhands browser network --clear`, click Save with `--observe 'testid=git-sync-save-button' --observe-ms 2000` (shows `Checking repository...`), wait for the toast; `control-openhands browser network --last 6` lists `/api/automation/v1/git-sync/check` then `/api/automation/v1/git-sync/config`. After reload the token input is `""`. Run `control-openhands browser click 'testid=git-sync-clear-token-switch >> xpath=ancestor::label'`; `browser enabled 'testid=git-sync-token-input'` is `false`. Click Save, wait for the toast, reload; `browser eval "document.querySelector('[data-testid=git-sync-clear-token-switch]').checked"` is `false`.
- **Interval (`F24.interval`).** Fill `testid=git-sync-interval-input` with `300`, Save, wait for the toast, reload: the card's `Sync every (seconds)` field reads `Every 300s` (`control-openhands browser text 'role=heading[name="Sync Status"] >> xpath=ancestor::div[2]'`) and the input value is `300`. Fill it with `''`, Save, reload: `Manual only` and input `0`. Fill `-5` and click Save: `browser eval "document.querySelector('[data-testid=git-sync-interval-input]').validationMessage"` is `Value must be greater than or equal to 0.`; reload to drop the draft.
- **Pause and resume (`F24.pause`).** Click `testid=git-sync-enabled-switch >> xpath=ancestor::label`; `browser enabled 'testid=git-sync-save-and-sync-button'` is `false`. Run `control-openhands browser network --clear`, click Save, wait for the toast; `control-openhands browser network` lists no `git-sync/check`. After reload the pill is `Disabled`, `enabled 'testid=git-sync-now-button'` is `false` and the card still shows `<repo-path>`; screenshot `--feature F24.pause --name disabled`. The switch's help text (`control-openhands browser text 'testid=git-sync-enabled-switch >> xpath=ancestor::label/following-sibling::p'`) is `Sync is on as soon as a repository is configured. Turn this off to pause syncing without losing the configuration.` Click the switch label again, Save, wait for the toast: the pill is `Enabled`.
@@ -92,7 +92,7 @@ Preconditions:
- The activity row (`git-sync-activity-row`, `data-state` `running`/`succeeded`/`failed`) is page state: it disappears on reload and does not appear for a background cycle that starts and ends between two idle polls (15 s). Assert background cycles through the commit hash, not the row.
- Toasts (`Git Sync settings saved.`, `Enable Git Sync before triggering a sync.`) last a few seconds: read them with `browser wait-text` immediately after the click, never after a long `--observe` window.
- Save's label cycles `Save Changes` → `Checking repository...` (only when a checked field changed) → `Saving...` → `Save Changes` in about 100 ms against a local repo; `--observe 'testid=git-sync-save-button'` records it.
-- Expected: setting, rotating or clearing an encryption key re-exports every automation: the next sync rewrites plaintext as ciphertext (`gAAAAA…`), rewrites ciphertext with the new key, or restores plaintext after clearing. Known failure (reproduced 2026-10-08, automation 1.19.0, desktop and phone, selected backend Local): each key-only save leaves `dirty_count` at `0`; **Sync now** reports `Sync complete` without a new commit. Setting a key leaves existing files plaintext, rotating retains the old ciphertext, and clearing retains ciphertext despite the `Not encrypted` pill. A separate automation change does export under the current key setting, but does not satisfy key-only re-export (OpenHands/automation#551, fixed by OpenHands/automation#563 after the 1.19.0 release; re-drive once `config/defaults.json` `versions.automation` includes it).
+- Expected: setting, rotating or clearing an encryption key re-exports every automation: the next sync rewrites plaintext as ciphertext (`gAAAAA…`), rewrites ciphertext with the new key, or restores plaintext after clearing. Known failure (reproduced 2026-10-09, automation 1.19.0, desktop and phone, selected backend Local): each key-only save leaves `dirty_count` at `0`; **Sync now** reports `Sync complete` without a new commit. Setting a key leaves existing files plaintext, rotating retains the old ciphertext, and clearing retains ciphertext despite the `Not encrypted` pill. A separate automation change does export under the current key setting, but does not satisfy key-only re-export ([OpenHands/automation#551](https://github.com/OpenHands/automation/issues/551), closed). The fix ([OpenHands/automation#563](https://github.com/OpenHands/automation/pull/563)) merged after the 1.19.0 release and is not yet in a released automation pinned by Canvas (`config/defaults.json` `versions.automation` is `1.19.0`). On automation `main` (`control-openhands launch --new --automation-ref main`, commit `b04ef24`, 2026-10-09) the same recipe passes at desktop and phone: each key-only save sets `dirty_count` to `1`, and the next **Sync now** commits ciphertext after set, new ciphertext after rotate and plaintext after clear. Re-drive on the pinned stack once `versions.automation` names a release that includes #563, then drop this Known failure.
- Nothing on the page says whether a token is stored: the token placeholder is always `Leave blank to keep the current token`, and the author fields always reload blank. Prove author changes from the commit (`git log --format='%an <%ae>'`).
- The card's Repository value drops embedded credentials, as link text and as plain text: an http(s) URL loses its whole `user:token@` (a token alone in the user slot too), other schemes keep the user and lose only the password (`ssh://git@host:2222/...`), and the rest stays as configured. The Repository URL input and `GET /api/automation/v1/git-sync/status` still carry the credentials, so a screenshot of the form shows them. Never type a real token into the URL.
- The page has no Automate sub-navigation (Dashboard / Templates) and no command-menu entry; leave through **Back to Automations** (`role=link`, not a button).
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 e652ffcfec..7b6601588c 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
@@ -112,7 +112,8 @@ Preconditions:
- 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.
- 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.
-- Local and remote backends are probed every 30 s (settings + server_info), Cloud backends every 5 min (#18128), so allow up to 40 s for a local dot to change. While the active backend is down, any reload lands on the recovery gate: drive the disconnected rows on the page that was already loaded.
+- Expected: Escape never clears the backend selector. Known failure (reproduced 2026-10-09, #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; fix PR #18235).
+- Local and remote backends are probed every 30 s (settings + server_info), Cloud backends every 5 min (#18128), so allow up to 40 s for a local dot to change. `browser network --clear`, wait about 70 s, then `browser network --filter server_info` shows two probes per local origin 30 s apart. While the active backend is down, any reload lands on the recovery gate: drive the disconnected rows on the page that was already loaded.
- `/oauth/device/verify` is a Cloud page that also ships in Canvas: on a local stack its POST target does not exist, so Authorize always ends in `Error`. The "Authentication Required" state is unreachable (`useIsAuthed` always resolves `true`).
- `/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).
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 73d7e9171a..7fa676c4d3 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
@@ -39,7 +39,7 @@ Source: `bin/agent-canvas.mjs`, `scripts/dev-with-automation.mjs`, `scripts/dev-
- `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.
- `F26.lib-style-scope`: all bundled CSS is scoped under `[data-agent-server-ui]`, and theme tokens such as `--oh-color-base` live on that scope root (observable in the standalone app).
- `F26.lib-host-app`: mounted in a separate host app, Canvas styles stay inside the scope and `styleOverrides` restyle it. Blocked: no host-app example exists.
-- `F26.docker-image`: `docker run ghcr.io/openhands/agent-canvas` serves Canvas at `/canvas`; without `AGENT_CANVAS_ALLOW_LAN_SESSION_KEY=true` the UI asks for the key. Blocked without a Docker daemon. Known failure (reproduced 2026-10-08): on a host whose kernel has no IPv6 (no `/proc/net/if_inet6`), `docker/entrypoint.sh` starts the static server with `--host ::`, which exits with `listen EAFNOSUPPORT: address family not supported :::8000`, and the container stops within about half a minute of start, with exit code `0` (#18178).
+- `F26.docker-image`: `docker run ghcr.io/openhands/agent-canvas` serves Canvas at `/canvas`; without `AGENT_CANVAS_ALLOW_LAN_SESSION_KEY=true` the UI asks for the key. Blocked without a Docker daemon. Known failure (reproduced 2026-10-09 with an image built from this checkout): on a host whose kernel has no IPv6 (no `/proc/net/if_inet6`), `docker/entrypoint.sh` starts the static server with `--host ::`, which exits with `listen EAFNOSUPPORT: address family not supported :::8000`, and the container stops within about half a minute of start, with exit code `0` (#18178).
- `F26.docker-conversation-runtime`: `OH_CONVERSATION_RUNTIME=docker` runs each new conversation in its own container. Blocked without a Docker daemon.
- `F26.helm-chart`: `helm install agent-canvas ./helm/agent-canvas` runs the Docker image as a Kubernetes StatefulSet with a PVC and an Ingress. Blocked without `helm` and a cluster.
- `F26.desktop-boot-splash`: the Electron app first shows a dark splash (logo, **OpenHands Agent Canvas**, spinner, `Starting backend services…`, then live service-log lines, a first-launch note and **Show details**).
@@ -94,7 +94,7 @@ 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, and Docker Hub may answer 429. 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`.
+- **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). On a kernel without IPv6 the unmodified image hits #18178: record that run `fail` with `docker logs` (`EAFNOSUPPORT ... :::8000`) and `docker inspect -f '{{.State.Status}} {{.State.ExitCode}}'` (`exited 0`). To drive the rest of the row anyway (arrange only, say so in the note), start it with the static server bound to IPv4: `docker run -d --name <qa-name> -p 127.0.0.1:<free>:8000 --entrypoint tini qa-agent-canvas:target -- bash -c "sed 's/--host ::/--host 0.0.0.0/' /opt/agent-canvas/entrypoint.sh > /tmp/ep.sh && exec bash /tmp/ep.sh"`; `/canvas/` answers `200` within about 20 s. 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.
- **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).
diff --git a/.agents/skills/verify-openhands/references/feature-map/README.md b/.agents/skills/verify-openhands/references/feature-map/README.md
index f5105d5162..7ee56e0cce 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@ed815e141409c7b52991ab8d13c553b595da9165 (2026-10-06). 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@553d4519113192a80fb18376fdb6c1e39fa06387 (2026-10-09). 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
@@ -45,7 +45,7 @@ Maintenance baseline: main@ed815e141409c7b52991ab8d13c553b595da9165 (2026-10-06)
- Verbs that are easy to miss: `browser mouse-click X Y` (backdrops and
overlays), `click --modifiers Control,Meta` (open in new tab, multi-select),
`click --hover-first` (hover-driven toggles), `click --expect-new-url`,
- `browser tooltip <sel>`, `browser wait <sel> --state hidden|visible|detached`
+ `browser tooltip <sel>` (it returns a tooltip that is still open: before reading another one, hover elsewhere and `browser wait 'role=tooltip' --state detached`), `browser wait <sel> --state hidden|visible|detached`
(use it after toggles: animations make an immediate `count` or `visible`
lie), `browser media --clear` (forget earlier sounds before a negative sound
check). `browser eval` takes one expression; wrap statements in an IIFE.
@@ -230,8 +230,8 @@ 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 | 26 |
-| 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 | 39 |
+| 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 |
+| 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 | 40 |
| 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 | 40 |
| 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 |
| F06 | [Agent activity](F06-agent-activity.md) | messages, tool events, thinking, empty state, confirmation mode, stop/resume, errors, branching | `/conversations/<id>` | LLM | 35 |
@@ -257,7 +257,7 @@ How to write and prove new entries: [../mapping.md](../mapping.md).
| 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 | 32 |
| F27 | [Workspace tools](F27-workspace-tools.md) | terminal, browser, planner, task list, usage, `canvas_ui_control` | drawer tabs | LLM | 30 |
-27 families, 743 sub-features. `control-openhands map ids` lists every ID with its file.
+27 families, 745 sub-features. `control-openhands map ids` lists every ID with its file.
### Neighbouring families