[release] v0.128.6

api/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "api"
-version = "0.128.5"
+version = "0.128.6"
 description = "Agenta API"
 requires-python = ">=3.11,<3.14"
 authors = [

clients/python/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agenta-client"
-version = "0.128.5"
+version = "0.128.6"
 description = "Fern-generated Python client for the Agenta API."
 requires-python = ">=3.11,<3.14"
 authors = [

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: 0.128.5
-appVersion: "v0.128.5"
+version: 0.128.6
+appVersion: "v0.128.6"
 keywords:
   - agenta
   - llm

sdks/python/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agenta"
-version = "0.128.5"
+version = "0.128.6"
 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 = "0.128.5"
+version = "0.128.6"
 description = "Agenta Services (Chat & Completion)"
 requires-python = ">=3.11,<3.14"
 authors = [

web/ee/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@agenta/ee",
-    "version": "0.128.5",
+    "version": "0.128.6",
     "private": true,
     "engines": {
         "node": "24.x"

web/mobile/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@agenta/mobile",
-    "version": "0.128.5",
+    "version": "0.128.6",
     "private": true,
     "engines": {
         "node": "24.x"

web/mobile/src/features/playground/usePlaygroundSelection.ts

@@ -3,13 +3,17 @@ import {useCallback, useEffect, useRef} from "react"
 import {isLocalDraftId} from "@agenta/entities/shared"
 import {workflowDetailQueryAtomFamily, workflowMolecule} from "@agenta/entities/workflow"
 import {playgroundController, playgroundInitializedAtom} from "@agenta/playground"
+import {projectIdAtom} from "@agenta/shared/state"
 import {atom, useAtomValue, useStore} from "jotai"
 import {atomFamily} from "jotai-family"
 import {queryClientAtom} from "jotai-tanstack-query"
 import {useRouter} from "next/router"
 
 import {clearPlayground} from "./clearPlayground"
-import {playgroundLatestRevisionIdAtomFamily} from "./playgroundLatestRevision"
+import {
+    playgroundLatestRevisionIdAtomFamily,
+    playgroundLatestRevisionKey,
+} from "./playgroundLatestRevision"
 import {parseRevisionIds} from "./playgroundUrl"
 
 /** How far the app and its selected revisions have loaded. */
@@ -109,10 +113,20 @@ export const usePlaygroundSelection = ({appId}: {appId: string}): PlaygroundSele
         if (kept.length !== current.length) {
             store.set(playgroundController.actions.setEntityIds, kept)
         }
-        if (kept.length === 0 && latestId) select([latestId])
-        // `keptIds` only triggers the effect: in the commit that sets the playground up it is
-        // still the value from before, so the store is read instead. `appUsable` and
-        // `router.isReady` run it again in the commit that sets the playground up: the latest
+        if (kept.length === 0) {
+            // The latest revision comes from the query cache, not from `latestId`. The query
+            // library tells this screen of a new answer on a later task. Right after a delete
+            // the reload has answered, but `latestId` can still be the deleted revision.
+            const queryClient = store.get(queryClientAtom)
+            const latest = queryClient.getQueryState<string | null>(
+                playgroundLatestRevisionKey(appId, store.get(projectIdAtom)),
+            )
+            // A failed reload keeps the previous answer, which can be a revision just deleted.
+            if (latest && latest.status !== "error" && latest.data) select([latest.data])
+        }
+        // `keptIds` and `latestId` only trigger the effect: in the commit that sets the
+        // playground up `keptIds` is still the value from before, so the store is read instead.
+        // `appUsable` and `router.isReady` run it again in the commit that sets the playground up: the latest
         // revision can answer before the app, and nothing else would run it then.
     }, [appId, appUsable, keptIds, latestId, router.isReady, select, store])
 

web/mobile/tests/unit/playgroundRevisionActions.render.test.tsx

@@ -17,12 +17,18 @@ import {playgroundController} from "@agenta/playground"
 import {queryClient as hostQueryClient} from "@agenta/shared/api"
 import {idbQueryStorage} from "@agenta/shared/api/persist"
 import {projectIdAtom, sessionAtom} from "@agenta/shared/state"
-import {QueryClientProvider, type QueryClient} from "@tanstack/react-query"
+import {
+    defaultScheduler,
+    notifyManager,
+    QueryClientProvider,
+    type QueryClient,
+} from "@tanstack/react-query"
 import {getDefaultStore} from "jotai"
 import {queryClientAtom} from "jotai-tanstack-query"
 import {createRoot, type Root} from "react-dom/client"
 import {afterEach, beforeEach, describe, expect, it, onTestFinished, vi} from "vitest"
 
+import {playgroundLatestRevisionKey} from "@/features/playground/playgroundLatestRevision"
 import {PlaygroundScreen} from "@/features/playground/PlaygroundScreen"
 
 const {http, router} = vi.hoisted(() => ({
@@ -319,15 +325,23 @@ const find = (name: string) => {
         ),
     ].find((item) => label(item) === name || item.getAttribute("aria-label") === name)
 }
+/**
+ * Waits for a control to show. The compare view loads its code the first time two revisions
+ * show, which can take longer than one settle.
+ */
+const findSoon = async (name: string) => {
+    for (let i = 0; i < 40 && !find(name); i++) await settle()
+    return find(name)
+}
 const click = async (name: string) => {
-    const target = find(name)
+    const target = await findSoon(name)
     expect(target, `a control named ${name}`).toBeDefined()
     await act(async () => target!.click())
     await settle()
 }
 /** Radix opens a menu on pointer down, not on click. */
 const openMenu = async (name: string) => {
-    const trigger = find(name)
+    const trigger = await findSoon(name)
     expect(trigger, `a menu named ${name}`).toBeDefined()
     await act(async () => {
         trigger!.dispatchEvent(
@@ -742,6 +756,29 @@ describe("Delete", () => {
         expect(latestOnDisk?.id ?? null).not.toBe("rev-2")
     })
 
+    it("After a delete, the selection takes the newest revision the reload answered, before the screen hears of it", async () => {
+        await mount()
+        expect(shown()).toEqual(["rev-2"])
+        // The server deletes rev-2. The latest-revision reload answers rev-1, but the screen
+        // hears of it later: the query library tells its observers on a later task. After a
+        // delete, the page takes the revision out in that gap.
+        world.revisions = world.revisions.filter((revision) => revision.id !== "rev-2")
+        const held: (() => void)[] = []
+        notifyManager.setScheduler((callback) => held.push(callback))
+        onTestFinished(() => notifyManager.setScheduler(defaultScheduler))
+        const latestKey = playgroundLatestRevisionKey("app-1", projectId)
+        await client.refetchQueries({queryKey: latestKey})
+        expect(client.getQueryData(latestKey)).toBe("rev-1")
+
+        act(() => store.set(playgroundController.actions.setEntityIds, []))
+
+        // The deleted revision must not come back.
+        expect(shown()).toEqual(["rev-1"])
+        notifyManager.setScheduler(defaultScheduler)
+        act(() => held.splice(0).forEach((callback) => callback()))
+        await expectShown(["rev-1"])
+    })
+
     it("The dialog closes once the delete succeeds; a revision picked meanwhile stays with its edits", async () => {
         await mount()
         let release: () => void = () => undefined

web/oss/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@agenta/oss",
-    "version": "0.128.5",
+    "version": "0.128.6",
     "private": true,
     "engines": {
         "node": "24.x"

web/oss/src/components/pages/evaluations/NewEvaluation/Components/NewEvaluationModalContent.tsx

@@ -16,6 +16,8 @@ import {openEvaluatorDrawerAtom} from "@/oss/state/evaluator/evaluatorDrawerStor
 import TabLabel from "../assets/TabLabel"
 import {NewEvaluationModalContentProps} from "../types"
 
+import {newEvaluationEvaluatorDrawerParams} from "./evaluatorDrawerParams"
+
 const tabsContainerClass =
     "h-full flex [&_.ant-tabs-content-holder]:pl-4 [&_.ant-tabs-content-holder]:flex-1 " +
     "[&_.ant-tabs-content-holder]:overflow-auto [&_.ant-tabs-tab]:text-colorTextSecondary " +
@@ -133,13 +135,7 @@ const NewEvaluationModalContent: FC<NewEvaluationModalContentProps> = ({
                 return
             }
 
-            openEvaluatorDrawer({
-                entityId: localId,
-                mode: "create",
-                onEvaluatorCreated,
-                // Created inline: stay in this modal instead of opening the new evaluator's page.
-                postCreateNavigation: "stay",
-            })
+            openEvaluatorDrawer(newEvaluationEvaluatorDrawerParams(localId, onEvaluatorCreated))
             onSelectTemplate?.(evaluator)
         },
         [openEvaluatorDrawer, onSelectTemplate, onEvaluatorCreated],

web/oss/src/components/pages/evaluations/NewEvaluation/Components/evaluatorDrawerParams.test.ts

@@ -0,0 +1,39 @@
+import {
+    workflowRevisionDrawerCallbackAtom,
+    workflowRevisionDrawerIsolatedPlaygroundAtom,
+    workflowRevisionDrawerOpenAtom,
+} from "@agenta/playground-ui/workflow-revision-drawer"
+import {createStore} from "jotai"
+import {describe, expect, it, vi} from "vitest"
+
+import {openEvaluatorDrawerAtom} from "@/oss/state/evaluator/evaluatorDrawerStore"
+
+import {newEvaluationEvaluatorDrawerParams} from "./evaluatorDrawerParams"
+
+describe("New Evaluation evaluator drawer", () => {
+    it("opens in an isolated playground so closing it cannot delete the app's test cases", () => {
+        const store = createStore()
+        const onEvaluatorCreated = vi.fn()
+
+        store.set(
+            openEvaluatorDrawerAtom,
+            newEvaluationEvaluatorDrawerParams("local-evaluator", onEvaluatorCreated),
+        )
+
+        expect(store.get(workflowRevisionDrawerOpenAtom)).toBe(true)
+        expect(store.get(workflowRevisionDrawerIsolatedPlaygroundAtom)).toBe(true)
+    })
+
+    it("still hands the created evaluator to the modal", () => {
+        const store = createStore()
+        const onEvaluatorCreated = vi.fn()
+
+        store.set(
+            openEvaluatorDrawerAtom,
+            newEvaluationEvaluatorDrawerParams("local-evaluator", onEvaluatorCreated),
+        )
+        store.get(workflowRevisionDrawerCallbackAtom)?.({configId: "rev-1"} as never)
+
+        expect(onEvaluatorCreated).toHaveBeenCalledWith("rev-1")
+    })
+})

web/oss/src/components/pages/evaluations/NewEvaluation/Components/evaluatorDrawerParams.ts

@@ -0,0 +1,20 @@
+import type {OpenDrawerParams} from "@/oss/state/evaluator/evaluatorDrawerStore"
+
+/**
+ * Arguments for the evaluator creation drawer opened from the New Evaluation modal.
+ *
+ * `isolatedPlayground: true` runs the drawer in its own store. The modal can open on
+ * top of a playground that has test cases connected, and a shared store would let the
+ * drawer's close cleanup delete those rows.
+ */
+export const newEvaluationEvaluatorDrawerParams = (
+    entityId: string,
+    onEvaluatorCreated: ((configId?: string) => void) | undefined,
+): OpenDrawerParams => ({
+    entityId,
+    mode: "create",
+    onEvaluatorCreated,
+    isolatedPlayground: true,
+    // Created inline: stay in this modal instead of opening the new evaluator's page.
+    postCreateNavigation: "stay",
+})

web/oss/src/state/evaluator/evaluatorDrawerStore.ts

@@ -25,7 +25,7 @@ import {atom} from "jotai"
 
 type EvaluatorDrawerMode = "create" | "view"
 
-interface OpenDrawerParams {
+export interface OpenDrawerParams {
     entityId: string
     mode: EvaluatorDrawerMode
     /** List of entity IDs for prev/next navigation */

web/package.json

@@ -1,6 +1,6 @@
 {
     "name": "agenta-web",
-    "version": "0.128.5",
+    "version": "0.128.6",
     "workspaces": [
         "ee",
         "mobile",

web/packages/agenta-api-client/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentaai/api-client",
-  "version": "0.128.5",
+  "version": "0.128.6",
   "private": true,
   "type": "module",
   "main": "./dist/index.js",