[release] v0.128.9

.agents/skills/test-design/SKILL.md

@@ -0,0 +1,57 @@
+---
+name: test-design
+description: Decide what to test and how to write the tests for a frontend change. Use before you write or review unit tests for web/mobile or the @agenta/* packages, when you fix a bug and add a regression test, and before you say a change is tested. Gives the risk list, the rules for each test, and the mutation check.
+---
+
+# Test design
+
+Tests exist to catch a change that hurts a user. Do three things in order.
+
+## 1. List the risks first
+
+Before you write a test, write a short risk list. Put it in the PR body under "What the tests cover". Ask what this change can break for a user. Order the answers by cost:
+
+1. Saved data is lost, changed, or sent wrong (commit, rename, delete, discard, import).
+2. The wrong scope or permission: another project's data, a hidden view, a role that must not see it.
+3. A main flow stops: open, edit, save, run, navigate.
+4. The screen shows a wrong value or a wrong state.
+5. Looks and copy. These need no unit test.
+
+For each risk in groups 1 to 3, check these cases:
+
+- The normal case.
+- Empty, missing, or one item.
+- More than one page of data.
+- The request fails, answers late, or answers out of order.
+- The user acts twice, or acts while a save runs.
+- The user leaves the page or discards the change.
+
+Write one test for each risk and case that matters. Skip a case only if you can say why it cannot happen.
+
+## 2. Write each test through the screen
+
+- Name the test as a user behavior, in a plain sentence.
+- Test through the screen, the hook, or the package atom. Do not test private helpers or the order of internal calls.
+- Use the real `@agenta/*` atoms, molecules, and controllers in a `createStore()` store.
+- Mock only what lives outside the code: HTTP, time, randomness, the router, and browser APIs that jsdom lacks, such as the clipboard.
+- If a test needs more than about 5 `vi.mock` calls, the level is wrong. Move down to the package state, or up to a screen with real children.
+- Never assert on a mock. An assertion that only checks what a mock returns or records tests the mock, so delete it. Keep an assertion on what the screen shows, even in a test that also mocks HTTP.
+- Find elements as a user does: by role, label, or visible text. Do not use CSS classes or test ids. In `web/mobile`, use `@testing-library/react` (`screen`, `render`).
+- Write expected values as literals. Do not compute them with the code under test.
+- For code that saves data, assert the exact request body that reaches the server.
+- A pure function with a rule over many inputs may use a table of cases.
+- Do not write tests for constants, copy, styles, type shapes, or framework behavior. Do not write a test only to raise coverage.
+
+## 3. Check your tests before you finish
+
+Run the mutation check. Break the code in each realistic way below, one at a time, and confirm that at least one test fails:
+
+- A wrong constant or argument.
+- A condition turned around, or a wrong branch.
+- A missing state change or a missing request.
+- An empty or default return.
+- A missing check for empty, null, or not allowed.
+
+If no test fails, add a test. Or write in the PR body why the behavior is not covered.
+
+For a bug fix, write the test that fails before the fix. Then ask which other cases from the list in block 1 have the same weakness. Test those cases too, in the same PR.

.claude/skills/test-design

@@ -0,0 +1 @@
+../../.agents/skills/test-design

.gitignore

@@ -120,6 +120,7 @@ clients/python/tests/results/
 !.agents/skills/review-naming/
 !.agents/skills/review-simplify/
 !.agents/skills/review-slop/
+!.agents/skills/test-design/
 !.claude/
 .claude/*
 !.claude/skills/
@@ -157,6 +158,7 @@ clients/python/tests/results/
 !.claude/skills/review-naming
 !.claude/skills/review-simplify
 !.claude/skills/review-slop
+!.claude/skills/test-design
 !.claude/settings.json
 
 # Entire (entire.com) shared config and agent hooks; each hook is a no-op

api/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "api"
-version = "0.128.8"
+version = "0.128.9"
 description = "Agenta API"
 requires-python = ">=3.11,<3.14"
 authors = [

clients/python/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agenta-client"
-version = "0.128.8"
+version = "0.128.9"
 description = "Fern-generated Python client for the Agenta API."
 requires-python = ">=3.11,<3.14"
 authors = [

docs/design/m-developer-views/registry.md

@@ -86,10 +86,23 @@ the user to /w and the views follow their stage, as before beta developer mode e
 `isBetaDeveloperModeEnabled()` in `@agenta/shared/api` reads it. The local dev env files set it
 to `true`.
 
+## One gate in beta developer mode
+
+With beta developer mode active, every view in the registry is on: pages, nav links, feature
+gates and every cross link (Overview, Open in playground, the prompt row click). Nothing on
+the developer side checks `stage` for that user, and Preferences shows no preview switches
+(they would change nothing). A view added to the registry needs no extra step to appear.
+
+Stages, the preview switches and `AGENTA_DEVELOPER_VIEWS` stay for users without beta
+developer mode (and for deployments that do not offer it, such as production). They are the
+only place `stage` is read. Do not read `stage` in a screen or a cross link: ask
+`useDeveloperViewEnabled(id)`.
+
 ## Helpers (`features/developer/developerViewAccess.ts`)
 
 - `isDeveloperViewEnabled(view, {previewOptedIn, forcedIds})`: the pure rule.
 - `useDeveloperViewEnabled(viewId)`: `true`, `false`, or `null` until the signed-in user is
+  Under beta developer mode it is `true` for every view in the registry.
   known. The first render is always `null`, so the answer never differs from the prerendered
   HTML.
 - `useEnabledDeveloperViews()`: the views the nav shows.

hosting/kubernetes/helm/Chart.yaml

@@ -2,8 +2,8 @@ apiVersion: v2
 name: agenta
 description: A Helm chart for deploying Agenta (OSS or EE) on Kubernetes
 type: application
-version: 0.128.8
-appVersion: "v0.128.8"
+version: 0.128.9
+appVersion: "v0.128.9"
 keywords:
   - agenta
   - llm

sdks/python/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agenta"
-version = "0.128.8"
+version = "0.128.9"
 description = "Agenta is the open-source workspace for your agents. Build agents through chat, improve them with feedback, and share them with your team."
 readme = "README.md"
 requires-python = ">=3.11,<3.14"

services/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "services"
-version = "0.128.8"
+version = "0.128.9"
 description = "Agenta Services (Chat & Completion)"
 requires-python = ">=3.11,<3.14"
 authors = [

web/AGENTS.md

@@ -425,6 +425,7 @@ the source of truth for this area. Load it when you:
   antd-style call sites; never raw antd `Modal`).
 - Use entity state (molecules), the loadable/runnable bridges, or the `EntityPicker`.
 - Write package unit tests (they live in `tests/unit/`, not `src/`).
+- Decide what to test or review tests: load the **`test-design`** skill (risk list, how to write each test, mutation check).
 
 Quick placement heuristic:
 

web/ee/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@agenta/ee",
-    "version": "0.128.8",
+    "version": "0.128.9",
     "private": true,
     "engines": {
         "node": "24.x"

web/mobile/AGENTS.md

@@ -108,3 +108,13 @@ cold compile time, and the `next build` time. The budget is in `scripts/speed/bu
   a barrel that pulls a view into a shared page. If he approves, run `speed:check --write`
   after a fresh build, review the diff of `budget.json`, and say why in the PR body.
   A new route has no budget until it is added the same way.
+
+<!-- BEGIN:nextjs-agent-rules -->
+
+## This is NOT the Next.js you know
+
+This version has breaking changes — APIs, conventions, and file structure may all differ from your training data. Read the relevant guide in `node_modules/next/dist/docs/` (resolved from this file's directory; in monorepos the `next` package may not be visible from the repo root) before writing any code. Heed deprecation notices.
+
+This block is written and re-added by `next dev` — verify at `node_modules/next/dist/server/lib/generate-agent-files.js`. Removing it from a diff only re-creates the uncommitted change; committing it with your work keeps the tree clean.
+
+<!-- END:nextjs-agent-rules -->

web/mobile/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@agenta/mobile",
-    "version": "0.128.8",
+    "version": "0.128.9",
     "private": true,
     "engines": {
         "node": "24.x"
@@ -71,6 +71,8 @@
     "devDependencies": {
         "@eslint/js": "^9.39.4",
         "@tailwindcss/postcss": "^4.1.0",
+        "@testing-library/dom": "^10.4.0",
+        "@testing-library/react": "^16.3.2",
         "@types/node": "^20.19.20",
         "@types/react": "^19.0.10",
         "@types/react-dom": "^19.0.4",

web/mobile/src/features/annotations/FocusScenarioPicker.tsx

@@ -14,6 +14,7 @@ import {copyToClipboard} from "@agenta/ui/utils"
 import {ArrowSquareOut, Copy} from "@phosphor-icons/react"
 import {useAtomValue, useSetAtom} from "jotai"
 
+import {useIsTraceQueue} from "./useIsTraceQueue"
 import {useScenarioTraceOpener} from "./useScenarioTraceOpener"
 
 /**
@@ -30,7 +31,7 @@ export const FocusScenarioPicker = ({
 }) => {
     const focusIds = useAtomValue(annotationSessionController.selectors.focusScenarioIds())
     const total = useAtomValue(annotationSessionController.selectors.scenarioIds()).length
-    const isTraces = useAtomValue(annotationSessionController.selectors.queueKind()) === "traces"
+    const isTraces = useIsTraceQueue()
     const traceRef = useAtomValue(
         annotationSessionController.selectors.scenarioTraceRef(scenarioId),
     )

web/mobile/src/features/annotations/QueueListScreen.tsx

@@ -3,7 +3,6 @@ import {
     createQueueDrawerOpenAtom,
     createQueueDrawerSelectionAtom,
 } from "@agenta/annotation"
-import {pageContentWidthClass} from "@agenta/ui/components/page-width"
 import {Button} from "@agenta/ui/ui"
 import {useSetAtom} from "jotai"
 import {Plus} from "lucide-react"
@@ -14,6 +13,7 @@ import {useBindWorkspaceMembers} from "@/features/members/useBindWorkspaceMember
 
 import {useBindProjectContext} from "../context/useBindProjectContext"
 import {useProjectPermission} from "../context/useProjectPermission"
+import {useDeveloperPageFrame} from "../developer/developerPageFrame"
 import {AppShell} from "../nav/AppShell"
 import {NavDrawer} from "../nav/NavDrawer"
 
@@ -22,8 +22,6 @@ import {QueueListContent} from "./QueueListContent"
 import {useQueueDelete} from "./useQueueDelete"
 import {useQueueList} from "./useQueueList"
 
-const PAGE_FRAME = `${pageContentWidthClass} lg:px-16`
-
 /** The annotation queues of the project: search, type filter, the list and the delete confirm. */
 export const QueueListScreen = ({
     workspaceId,
@@ -32,6 +30,7 @@ export const QueueListScreen = ({
     workspaceId: string
     projectId: string
 }) => {
+    const PAGE_FRAME = useDeveloperPageFrame()
     useBindWorkspaceMembers()
     useBindProjectContext(projectId)
     const base = `/w/${workspaceId}/p/${projectId}`

web/mobile/src/features/annotations/states/AllCaughtUp.tsx

@@ -4,12 +4,14 @@ import {Check, Plus} from "@phosphor-icons/react"
 import {useAtomValue, useSetAtom} from "jotai"
 import Link from "next/link"
 
+import {useIsTraceQueue} from "../useIsTraceQueue"
+
 /**
  * Every item of the queue is completed. Add to Testset exports them all; trace queues also point
  * back to observability for more traces.
  */
 export const AllCaughtUp = ({observabilityHref}: {observabilityHref: string}) => {
-    const isTraces = useAtomValue(annotationSessionController.selectors.queueKind()) === "traces"
+    const isTraces = useIsTraceQueue()
     const isExporting = useAtomValue(
         annotationSessionController.selectors.isAddToTestsetExporting(),
     )

web/mobile/src/features/annotations/useIsTraceQueue.ts

@@ -1,6 +1,8 @@
 import {annotationSessionController} from "@agenta/annotation"
 import {useAtomValue} from "jotai"
 
-/** True when the open queue holds traces. Only a test case queue is not a trace queue. */
-export const useIsTraceQueue = (): boolean =>
-    useAtomValue(annotationSessionController.selectors.queueKind()) !== "testcases"
+/** True when the queue's items are traces: a trace queue, or a query queue that collects them. */
+export const useIsTraceQueue = (): boolean => {
+    const kind = useAtomValue(annotationSessionController.selectors.queueKind())
+    return kind === "traces" || kind === "queries"
+}