fix(ai-chat): test_run_flow could test a different flow than the one asked (#11066)

* fix(ai-chat): test_run_flow could test a different flow than the one asked Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TmQDoFXVPwYyEN8PfV2oL7 * fix(ai-chat): prefer the flow editor stored at the path over one renamed to it Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TmQDoFXVPwYyEN8PfV2oL7 * test(ai-chat): default the flow helpers factory and trim duplicated setup Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TmQDoFXVPwYyEN8PfV2oL7 * refactor(ai-chat): resolve the flow editor to run by its storage path Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TmQDoFXVPwYyEN8PfV2oL7 * refactor(ai-chat): move the editor storage path context out of sessions Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TmQDoFXVPwYyEN8PfV2oL7 --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-11 16:09:39 +00:00 · 2026-09-10 18:41:21 +00:00
parent 569adb85c1
commit fa73539839
9 changed files with 167 additions and 31 deletions
@@ -110,6 +110,7 @@
 	import { isCloudHosted } from '$lib/cloud'
 	import { UserDraft } from '$lib/userDraft.svelte'
 	import { setOpenInSessionHandoff } from './sessions/openInSessionContext'
+	import { getEditorStoragePath, setEditorStoragePath } from './editorStoragePathContext'

 	let {
 		initialPath = $bindable(''),
@@ -819,6 +820,12 @@
 		return entries
 	}

+	// The storage path this editor is bound to, narrowing whatever an outer mount
+	// published (a session tab): the full-page editor holds its own, a drawer mount
+	// holds none.
+	const outerStoragePath = getEditorStoragePath()
+	setEditorStoragePath(() => liveEditorDraftStoragePath ?? outerStoragePath?.())
+
 	// "Open in AI session" target: the URL draft path the editor loads/saves by
 	// (which for a new flow differs from the live-edited friendly `$pathStore`),
 	// falling back to `$pathStore` in drawer mounts that carry no storage path.
@@ -705,7 +705,13 @@ export class AIChatManager {
 	>(undefined)
 	scriptEditorShowDiffMode = $state<(() => void) | undefined>(undefined)
 	scriptEditorGetLintErrors = $state<(() => ScriptLintResult) | undefined>(undefined)
+	/** The editor a FLOW-mode chat belongs to: the page owning the chat names itself here, and a
+	 * nested editor (a subflow drawer) takes it over while it is open. Unset in a session chat,
+	 * which keeps every open editor tab mounted and could only name an arbitrary one — a session
+	 * resolves an editor by its storage path through `flowEditorFor`. */
 	flowAiChatHelpers = $state<FlowAIChatHelpers | undefined>(undefined)
+	/** Every mounted flow editor. */
+	#flowEditors = new Set<FlowAIChatHelpers>()
 	appAiChatHelpers = $state<AppAIChatHelpers | undefined>(undefined)
 	/** Datatable creation policy: enabled flag, datatable name, and optional schema */
 	datatableCreationPolicy = $state<{
@@ -2415,7 +2421,8 @@ export class AIChatManager {
 						openArtifact: this.openArtifact
 					}
 				: {}),
-			testActiveFlow: async (args?: Record<string, any>) => this.flowAiChatHelpers?.testFlow(args),
+			testActiveFlow: async (storagePath: string, args?: Record<string, any>) =>
+				this.flowEditorFor(storagePath)?.testFlow(args),
 			getModifiedItems: () => (this.modifiedItems ? [...this.modifiedItems] : undefined),
 			attachedFiles: this.attachedFiles,
 			getUserInstructions: () => getUserCustomPrompts()[AIMode.GLOBAL] ?? '',
@@ -4664,7 +4671,11 @@ export class AIChatManager {
 	}

 	setFlowHelpers = (flowHelpers: FlowAIChatHelpers) => {
-		this.flowAiChatHelpers = flowHelpers
+		this.#flowEditors.add(flowHelpers)
+		// Only a chat that can reach FLOW mode names an editor (see `flowAiChatHelpers`).
+		if (!this.isSessionChat) {
+			this.flowAiChatHelpers = flowHelpers
+		}
 		untrack(() => {
 			if (this.autoAcceptEditsActive) {
 				this.acceptPendingFlowEdits(flowHelpers)
@@ -4672,10 +4683,17 @@ export class AIChatManager {
 		})

 		return () => {
-			this.flowAiChatHelpers = undefined
+			this.#flowEditors.delete(flowHelpers)
+			if (!this.isSessionChat) {
+				this.flowAiChatHelpers = undefined
+			}
 		}
 	}

+	private flowEditorFor(storagePath: string): FlowAIChatHelpers | undefined {
+		return [...this.#flowEditors].find((helpers) => helpers.getStoragePath() === storagePath)
+	}
+
 	// Registered by the /pipeline editor while it is mounted. Rebuilds the global
 	// tool set so the pipeline tools appear (and disappear on unregister). Pipeline
 	// AI edits apply directly as drafts, so there is nothing to auto-accept.
@@ -191,15 +191,18 @@ beforeEach(() => {
 })

 function createFlowHelpers({
-	hasPendingChanges,
-	acceptAllModuleActions,
-	testFlow = vi.fn()
+	hasPendingChanges = () => false,
+	acceptAllModuleActions = vi.fn(),
+	testFlow = vi.fn(),
+	storagePath = 'u/admin/live_flow'
 }: {
-	hasPendingChanges: () => boolean
-	acceptAllModuleActions: () => void
+	hasPendingChanges?: () => boolean
+	acceptAllModuleActions?: () => void
 	testFlow?: FlowAIChatHelpers['testFlow']
-}): FlowAIChatHelpers {
+	storagePath?: string
+} = {}): FlowAIChatHelpers {
 	return {
+		getStoragePath: () => storagePath,
 		getFlowAndSelectedId: vi.fn(),
 		getRootModules: vi.fn(),
 		inlineScriptSession: { get: vi.fn(), set: vi.fn(), clear: vi.fn() },
@@ -853,19 +856,36 @@ describe('AIChatManager autonomy mode', () => {

 		manager.isSessionChat = true
 		manager.sessionId = 'htc1xouxd96dcyo6ruqo39'
-		manager.setFlowHelpers(
-			createFlowHelpers({
-				hasPendingChanges: () => false,
-				acceptAllModuleActions: vi.fn(),
-				testFlow
-			})
-		)
+		manager.setFlowHelpers(createFlowHelpers({ testFlow }))

 		manager.changeMode(AIMode.GLOBAL)
-		const jobId = await manager.helpers.testActiveFlow({ name: 'Ada' })
+		const jobId = await manager.helpers.testActiveFlow('u/admin/live_flow', { name: 'Ada' })

 		expect(jobId).toBe('job-flow-preview')
 		expect(testFlow).toHaveBeenCalledWith({ name: 'Ada' })
+		// A session chat resolves an editor by its storage path, so it never names one.
+		expect(manager.flowAiChatHelpers).toBeUndefined()
+	})
+
+	// Session tabs keep every open flow editor mounted, so the last one to register is routinely
+	// a different flow than the one being tested.
+	it('tests the flow editor mounted on the storage path, not the last one registered', async () => {
+		const manager = new AIChatManager()
+		const testTarget = vi.fn(async () => 'job-target-flow')
+		const testLast = vi.fn(async () => 'job-last-flow')
+
+		manager.setFlowHelpers(
+			createFlowHelpers({ testFlow: testTarget, storagePath: 'u/admin/live_flow' })
+		)
+		manager.setFlowHelpers(
+			createFlowHelpers({ testFlow: testLast, storagePath: 'u/admin/other_flow' })
+		)
+
+		manager.changeMode(AIMode.GLOBAL)
+		const jobId = await manager.helpers.testActiveFlow('u/admin/live_flow', { name: 'Ada' })
+
+		expect(jobId).toBe('job-target-flow')
+		expect(testLast).not.toHaveBeenCalled()
 	})
 })

@@ -14,6 +14,9 @@
 	import type { ScriptLintResult } from '../shared'
 	import { applyFlowJsonUpdate, updateRawScriptModuleContent } from './helperUtils'
 	import { findModuleInFlow } from '$lib/components/flows/flowTree'
+	import { getEditorStoragePath } from '$lib/components/editorStoragePathContext'
+
+	const editorStoragePath = getEditorStoragePath()

 	let {
 		flowModuleSchemaMap,
@@ -162,6 +165,8 @@
 			selectionManager.selectId(id, { openPanel: true })
 		},

+		getStoragePath: () => editorStoragePath?.(),
+
 		testFlow: async (args, conversationId) => {
 			// Set preview args if provided
 			if (args) {
@@ -136,6 +136,10 @@ export interface FlowAIChatHelpers {
 	/** Run a test of the current flow using the UI's preview mechanism */
 	testFlow: (args?: Record<string, any>, conversationId?: string) => Promise<string | undefined>

+	/** The path this editor's draft is stored under. Tells a caller which of several mounted
+	 * editors is the one an active-editor context names. */
+	getStoragePath: () => string | undefined
+
 	/** Get lint errors from a specific module (focuses it first, waits for Monaco to analyze) */
 	getLintErrors: (moduleId: string) => Promise<ScriptLintResult>
 }
@@ -4806,10 +4806,12 @@ describe('global AI tools', () => {
 		})
 	})

-	it('test_run_flow uses the live flow editor test hook when the active editor matches the path', async () => {
+	// The editor is driven by the key its draft is stored under, which reads and edits of the
+	// path resolve through too: a staged rename leaves that key where it was.
+	it('test_run_flow drives the live flow editor by its storage path', async () => {
 		seedBackendDraft(
 			'flow',
-			'',
+			'u/admin/live_flow_storage',
 			{
 				path: 'u/admin/live_flow',
 				summary: 'Live flow',
@@ -4825,7 +4827,7 @@ describe('global AI tools', () => {
 		UserDraft.setLiveEditorDraft({
 			workspace: WORKSPACE,
 			itemKind: 'flow',
-			storagePath: '',
+			storagePath: 'u/admin/live_flow_storage',
 			effectivePath: 'u/admin/live_flow'
 		})
 		const testActiveFlow = vi.fn(async () => 'job-live-flow')
@@ -4842,7 +4844,7 @@ describe('global AI tools', () => {
 			)
 		)

-		expect(testActiveFlow).toHaveBeenCalledWith({ name: 'Ada' })
+		expect(testActiveFlow).toHaveBeenCalledWith('u/admin/live_flow_storage', { name: 'Ada' })
 		expect(FlowService.getFlowByPath).not.toHaveBeenCalled()
 		expect(JobService.runFlowPreview).not.toHaveBeenCalled()
 		expect(result).toContain('Result (SUCCESS)')
@@ -4851,7 +4853,7 @@ describe('global AI tools', () => {
 	it('test_run_flow falls back to preview when the live flow editor test hook returns undefined', async () => {
 		seedBackendDraft(
 			'flow',
-			'',
+			'u/admin/live_flow_fallback',
 			{
 				path: 'u/admin/live_flow_fallback',
 				summary: 'Live flow fallback',
@@ -4867,7 +4869,7 @@ describe('global AI tools', () => {
 		UserDraft.setLiveEditorDraft({
 			workspace: WORKSPACE,
 			itemKind: 'flow',
-			storagePath: '',
+			storagePath: 'u/admin/live_flow_fallback',
 			effectivePath: 'u/admin/live_flow_fallback'
 		})
 		const testActiveFlow = vi.fn(async () => undefined)
@@ -4884,7 +4886,7 @@ describe('global AI tools', () => {
 			)
 		)

-		expect(testActiveFlow).toHaveBeenCalledWith({ name: 'Ada' })
+		expect(testActiveFlow).toHaveBeenCalledWith('u/admin/live_flow_fallback', { name: 'Ada' })
 		expect(FlowService.getFlowByPath).not.toHaveBeenCalled()
 		expect(JobService.runFlowPreview).toHaveBeenCalledWith({
 			workspace: WORKSPACE,
@@ -4896,6 +4898,52 @@ describe('global AI tools', () => {
 		})
 	})

+	// The flow may be open in a session tab that isn't the one on screen: driving its editor
+	// would paint the run into a tab the user is not looking at.
+	it('test_run_flow previews rather than driving an editor the user is not looking at', async () => {
+		seedBackendDraft(
+			'flow',
+			'u/admin/background_flow',
+			{
+				path: 'u/admin/background_flow',
+				summary: 'Background flow',
+				value: { modules: [{ id: 'background_step', value: { type: 'identity' } }] },
+				schema: {},
+				edited_by: '',
+				edited_at: '',
+				archived: false,
+				extra_perms: {}
+			},
+			{ workspace: WORKSPACE }
+		)
+		UserDraft.setLiveEditorDraft({
+			workspace: WORKSPACE,
+			itemKind: 'flow',
+			storagePath: 'u/admin/flow_on_screen',
+			effectivePath: 'u/admin/flow_on_screen'
+		})
+		const testActiveFlow = vi.fn(async () => 'job-live-flow')
+
+		await withCompletedTestJob(() =>
+			callGlobalTool(
+				'test_run_flow',
+				{ path: 'u/admin/background_flow', args: { name: 'Ada' } },
+				toolCallbacks,
+				{ testActiveFlow }
+			)
+		)
+
+		expect(testActiveFlow).not.toHaveBeenCalled()
+		expect(JobService.runFlowPreview).toHaveBeenCalledWith({
+			workspace: WORKSPACE,
+			requestBody: {
+				path: 'u/admin/background_flow',
+				value: { modules: [{ id: 'background_step', value: { type: 'identity' } }] },
+				args: { name: 'Ada' }
+			}
+		})
+	})
+
 	it('test_run_step previews rawscript steps from the draft flow', async () => {
 		const content = 'export async function main(name: string) {\n\treturn name.toUpperCase()\n}'
 		await callGlobalTool('write_flow', {
@@ -288,6 +288,8 @@ const ACTIVE_GLOBAL_EDITOR_DRAFTS: readonly {
 export type GlobalActiveEditorContext = {
 	type: ActiveGlobalEditorType
 	path: string
+	/** The key the draft is stored under, which `path` leaves behind on a rename. */
+	storagePath: string
 	isLiveDraft: true
 }

@@ -4367,7 +4369,8 @@ type WriteDraftCtx = {
 export type SessionToolHelpers = { sessionId?: string }

 export type GlobalToolHelpers = SessionToolHelpers & {
-	testActiveFlow?: (args?: Record<string, any>) => Promise<string | undefined>
+	/** Runs the flow editor mounted on `storagePath`, if one is. */
+	testActiveFlow?: (storagePath: string, args?: Record<string, any>) => Promise<string | undefined>
 	attachedFiles?: AttachedFilesStore
 	// Read/write the user-level Global instructions. `setUserInstructions` persists the
 	// value and rebuilds the system message so the change applies on the next chat-loop
@@ -4396,15 +4399,21 @@ function operatingWorkspaceFromHelpers(helpers: unknown): string | undefined {
 	return (helpers as GlobalToolHelpers | undefined)?.operatingWorkspace
 }

-function activeFlowTestFromCtx(
+// Drive a live editor only for the flow on screen: several can be open at once (session tabs),
+// and a run painted into a background tab is a side effect the user never sees. Undefined
+// sends the caller to a preview run, which reports into the chat alone. The hook is bound to
+// that editor's storage path — the same key reads and edits of `path` resolve through, so a
+// staged rename cannot send the run to a different editor than the one being edited.
+function liveFlowTestHookFromCtx(
 	ctx: { workspace: string; helpers?: unknown },
 	path: string
-): GlobalToolHelpers['testActiveFlow'] | undefined {
+): ((args?: Record<string, any>) => Promise<string | undefined>) | undefined {
 	const activeEditor = getActiveGlobalEditorContext(ctx.workspace)
 	if (activeEditor?.type !== 'flow' || activeEditor.path !== path) {
 		return undefined
 	}
-	return (ctx.helpers as GlobalToolHelpers | undefined)?.testActiveFlow
+	const testActiveFlow = (ctx.helpers as GlobalToolHelpers | undefined)?.testActiveFlow
+	return testActiveFlow && ((args) => testActiveFlow(activeEditor.storagePath, args))
 }

 export type OpenPreviewHandler = (req: {
@@ -5775,7 +5784,7 @@ async function testRunFlowByPath(
 ): Promise<string> {
 	const { workspace, toolId, toolCallbacks } = ctx
 	const testArgs = normalizeTestRunArgs(args.args)
-	const testActiveFlow = activeFlowTestFromCtx(ctx, args.path)
+	const testActiveFlow = liveFlowTestHookFromCtx(ctx, args.path)

 	if (testActiveFlow) {
 		return executeTestRun({
@@ -8171,9 +8180,10 @@ export function getActiveGlobalEditorContext(
 ): GlobalActiveEditorContext | undefined {
 	for (const { itemKind, type } of ACTIVE_GLOBAL_EDITOR_DRAFTS) {
 		const liveDraft = UserDraft.getLiveEditorDraft(itemKind, { workspace })
-		const path = liveDraft?.effectivePath || liveDraft?.storagePath
+		if (!liveDraft) continue
+		const path = liveDraft.effectivePath || liveDraft.storagePath
 		if (!path) continue
-		return { type, path, isLiveDraft: true }
+		return { type, path, storagePath: liveDraft.storagePath, isLiveDraft: true }
 	}
 }

@@ -0,0 +1,19 @@
+import { getContext, setContext } from 'svelte'
+
+// The path an editor's draft is stored under, the same key the live-editor draft
+// registers. Published by whichever ancestor owns it (the session tab, or the
+// full-page editor) because the parts that must say which item an editor is open
+// on sit below both, and the path they can see themselves is the renamed one.
+
+const KEY = 'EditorStoragePath'
+
+/** `undefined` where the editor has no stored draft to speak of (a drawer mount). */
+export type EditorStoragePath = () => string | undefined
+
+export function setEditorStoragePath(storagePath: EditorStoragePath): void {
+	setContext(KEY, storagePath)
+}
+
+export function getEditorStoragePath(): EditorStoragePath | undefined {
+	return getContext<EditorStoragePath | undefined>(KEY)
+}
@@ -8,6 +8,7 @@
 	import { makeFlowCodec, makeScriptCodec, makeRawAppCodec } from './sessionDraftCodecs'
 	import { itemDisplayName } from './previewRouter'
 	import SessionItemNotFound from './SessionItemNotFound.svelte'
+	import { setEditorStoragePath } from '../editorStoragePathContext'

 	let {
 		runtime,
@@ -51,6 +52,10 @@
 	// rely on its presence, not its identity.
 	setContext('aiChatManager', runtime.manager)

+	// This tab's storage path, for the editor below: several tabs are mounted at
+	// once and only this one knows which item each is open on.
+	setEditorStoragePath(() => path)
+
 	// This tab's own editor cell (per (kind, path)); several tabs can be live at once.
 	const cell = $derived(
 		kind === 'flow'