@vellumai/assistant 0.8.6 → 0.8.7

package/src/__tests__/conversation-routes-disk-view.test.ts

@@ -255,8 +255,9 @@ function createFakeConversation(conversationId: string): Conversation {
       },
       _content: string,
       _userMessageId: string,
-      onEvent: (msg: Record<string, unknown>) => void,
+      options?: { onEvent?: (msg: Record<string, unknown>) => void },
     ): Promise<void> {
+      const onEvent = options?.onEvent ?? (() => {});
       const assistantText = "Synthetic assistant reply";
       const assistantMessage = createAssistantMessage(assistantText);
       const assistantMetadata = {
@@ -282,7 +283,7 @@ function createFakeConversation(conversationId: string): Conversation {
         this.conversationId,
         "assistant",
         JSON.stringify(assistantMessage.content),
-        assistantMetadata,
+        { metadata: assistantMetadata },
       );
       this.messages.push(assistantMessage);
 

package/src/__tests__/conversation-routes-guardian-reply.test.ts

@@ -81,8 +81,8 @@ mock.module("../memory/conversation-crud.js", () => ({
     conversationId: string,
     role: string,
     content: string,
-    metadata?: Record<string, unknown>,
-  ) => addMessageMock(conversationId, role, content, metadata),
+    options?: { metadata?: Record<string, unknown> },
+  ) => addMessageMock(conversationId, role, content, options),
   reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 

package/src/__tests__/conversation-routes-slash-commands.test.ts

@@ -130,8 +130,8 @@ mock.module("../memory/conversation-crud.js", () => ({
     conversationId: string,
     role: string,
     content: string,
-    metadata?: Record<string, unknown>,
-  ) => addMessageMock(conversationId, role, content, metadata),
+    options?: { metadata?: Record<string, unknown> },
+  ) => addMessageMock(conversationId, role, content, options),
   extractImageSourcePaths: () => undefined,
   getConversation: () => null,
   getConversationOverrideProfile: () => "short-context",

package/src/__tests__/conversation-runtime-assembly.test.ts

@@ -65,7 +65,7 @@ import {
   writeSlackMetadata,
 } from "../messaging/providers/slack/message-metadata.js";
 import { parentAlias } from "../messaging/providers/slack/render-transcript.js";
-import { defaultInjectorsPlugin } from "../plugins/defaults/injectors.js";
+import { defaultInjectorsPlugin } from "../plugins/defaults/injectors/register.js";
 import {
   registerPlugin,
   resetPluginRegistryForTests,
@@ -3990,11 +3990,11 @@ describe("assembleSlackActiveThreadFocusBlock", () => {
     expect(result!).not.toContain("Sibling top-level");
   });
 
-  test("preserves speaker attribution when flattening to plain text", () => {
+  test("preserves user attribution while leaving assistant text unchanged", () => {
     // The `<active_thread>` block is rendered as newline-joined plain text,
-    // discarding `Message.role`. Assistant rows and unnamed user rows must
-    // therefore carry an explicit `@assistant` / `@user` label so the model
-    // can still tell turns apart inside the flattened block.
+    // discarding `Message.role`. User rows keep explicit Slack attribution,
+    // while assistant rows remain raw to avoid teaching the model a synthetic
+    // reply prefix.
     const rows: SlackTranscriptInputRow[] = [
       buildRow(
         "user",
@@ -4021,17 +4021,16 @@ describe("assembleSlackActiveThreadFocusBlock", () => {
     const result = assembleSlackActiveThreadFocusBlock(rows, SLACK_CAPS);
     expect(result).not.toBeNull();
     expect(result!).toContain("@alice");
-    expect(result!).toContain("@assistant");
+    expect(result!).toContain("Assistant reply");
+    expect(result!).not.toContain("@assistant: Assistant reply");
     expect(result!).toContain("@user");
   });
 
   test("assistant reactions are not double-attributed (`@assistant: [... @assistant reacted ...]`)", () => {
     // `renderReaction` bakes `@assistant` into the reaction tag line
     // (`[11/14/23 14:28 @assistant reacted 👍 to Mxxxxxx]`). The
-    // post-render step that prepends `@assistant: ` to assistant content
-    // lines must skip reaction lines, otherwise the flattened block
-    // produces `@assistant: [... @assistant reacted ...]` — two
-    // attributions for one event.
+    // flattened block should preserve that renderer-owned attribution without
+    // adding any extra assistant-message prefix.
     const rows: SlackTranscriptInputRow[] = [
       buildRow(
         "user",
@@ -4039,7 +4038,7 @@ describe("assembleSlackActiveThreadFocusBlock", () => {
         1_000,
         buildMeta({ channelTs: PARENT_TS, displayName: "@alice" }),
       ),
-      // Assistant reply in the thread — gets an `@assistant:` prefix.
+      // Assistant reply in the thread — stays unchanged.
       buildRow(
         "assistant",
         "Assistant reply",
@@ -4081,12 +4080,13 @@ describe("assembleSlackActiveThreadFocusBlock", () => {
     expect(result).not.toBeNull();
     // Double-attribution anti-pattern must NOT appear anywhere.
     expect(result!).not.toContain("@assistant: [");
-    // Both the reaction attribution and the reply prefix are still present.
+    // Reaction attribution remains, and the assistant reply stays raw.
     expect(result!).toContain("@assistant reacted 👍");
-    expect(result!).toContain("@assistant: Assistant reply");
+    expect(result!).toContain("Assistant reply");
+    expect(result!).not.toContain("@assistant: Assistant reply");
   });
 
-  test("timezone-aware assistant rows keep renderer attribution in active-thread focus block", () => {
+  test("timezone-aware assistant rows stay unchanged in active-thread focus block", () => {
     const rows: SlackTranscriptInputRow[] = [
       buildRow(
         "user",
@@ -4127,15 +4127,12 @@ describe("assembleSlackActiveThreadFocusBlock", () => {
 
     const result = assembleSlackActiveThreadFocusBlock(rows, SLACK_CAPS);
     expect(result).not.toBeNull();
-    expect(result!).toContain(
-      "[nov 14 2023 3:13 PM MT assistant] Assistant reply",
-    );
-    expect(result!).not.toContain(
-      "@assistant: [nov 14 2023 3:13 PM MT assistant]",
-    );
+    expect(result!).toContain("\nAssistant reply\n");
+    expect(result!).not.toContain("[nov 14 2023 3:13 PM MT assistant]");
+    expect(result!).not.toContain("@assistant: Assistant reply");
   });
 
-  test("assistant content that only looks like a compact tag still gets active-thread attribution", () => {
+  test("assistant content that only looks like a compact tag stays unchanged", () => {
     const compactLookingContent =
       "[nov 14 2023 3:13 PM MT assistant] Assistant reply";
     const rows: SlackTranscriptInputRow[] = [
@@ -4168,7 +4165,8 @@ describe("assembleSlackActiveThreadFocusBlock", () => {
 
     const result = assembleSlackActiveThreadFocusBlock(rows, SLACK_CAPS);
     expect(result).not.toBeNull();
-    expect(result!).toContain(`@assistant: ${compactLookingContent}`);
+    expect(result!).toContain(`\n${compactLookingContent}\n`);
+    expect(result!).not.toContain(`@assistant: ${compactLookingContent}`);
   });
 
   test("assistant reaction overflow trailer is not double-attributed", () => {

package/src/__tests__/conversation-runtime-workspace.test.ts

@@ -1,7 +1,7 @@
 import { beforeEach, describe, expect, test } from "bun:test";
 
 import { applyRuntimeInjections } from "../daemon/conversation-runtime-assembly.js";
-import { defaultInjectorsPlugin } from "../plugins/defaults/injectors.js";
+import { defaultInjectorsPlugin } from "../plugins/defaults/injectors/register.js";
 import {
   registerPlugin,
   resetPluginRegistryForTests,
@@ -77,6 +77,24 @@ describe("applyRuntimeInjections — workspace top-level context", () => {
     );
     expect((result[0].content[2] as { text: string }).text).toBe("Hello");
   });
+
+  test("app-backed active surface tells the model to load app-builder with the right argument", async () => {
+    const messages: Message[] = [userMsg("Edit this app")];
+    const { messages: result } = await applyRuntimeInjections(messages, {
+      activeSurface: {
+        surfaceId: "sf_1",
+        html: "<div>test</div>",
+        appId: "app-1",
+        appName: "Example App",
+        appFiles: [],
+      },
+      workspaceTopLevelContext: null,
+    });
+
+    const activeWorkspaceText = (result[0].content[0] as { text: string }).text;
+    expect(activeWorkspaceText).toContain('skill: "app-builder"');
+    expect(activeWorkspaceText).not.toContain('id: "app-builder"');
+  });
 });
 
 describe("applyRuntimeInjections — minimal mode skips workspace blocks", () => {

package/src/__tests__/conversation-slash-queue.test.ts

@@ -1,14 +1,11 @@
 import { beforeEach, describe, expect, mock, test } from "bun:test";
 
-import type {
-  AgentEvent,
-  CheckpointDecision,
-  CheckpointInfo,
-} from "../agent/loop.js";
+import { CompactionCircuit } from "../agent/compaction-circuit.js";
+import type { AgentEvent, AgentLoopRunResult } from "../agent/loop.js";
 import type { ServerMessage } from "../daemon/message-protocol.js";
 import {
   conversationMessagesSyncTag,
-  type SyncChangedMessage,
+  type SyncChangedEvent,
 } from "../daemon/message-types/sync.js";
 import type { Message, ProviderResponse } from "../providers/types.js";
 
@@ -213,6 +210,7 @@ let pendingRuns: PendingRun[] = [];
 
 mock.module("../agent/loop.js", () => ({
   AgentLoop: class {
+    compactionCircuit = new CompactionCircuit("test-conv");
     constructor() {}
     getToolTokenBudget() {
       return 0;
@@ -226,15 +224,11 @@ mock.module("../agent/loop.js", () => ({
     async run(
       messages: Message[],
       onEvent: (event: AgentEvent) => void | Promise<void>,
-      _signal?: AbortSignal,
-      _requestId?: string,
-      _onCheckpoint?: (
-        checkpoint: CheckpointInfo,
-      ) => CheckpointDecision | Promise<CheckpointDecision>,
-    ): Promise<Message[]> {
-      return new Promise<Message[]>((resolve) => {
+    ): Promise<AgentLoopRunResult> {
+      const history = await new Promise<Message[]>((resolve) => {
         pendingRuns.push({ resolve, messages, onEvent });
       });
+      return { history, exitReason: null };
     }
   },
 }));
@@ -306,9 +300,9 @@ function makeConversation(): Conversation {
     "conv-1",
     provider,
     "system prompt",
-    4096,
     () => {},
     "/tmp",
+    { maxTokens: 4096 },
   );
   // Bypass real workspace git init: with "/tmp" as the workspace dir, a real
   // ensureInitialized() walks all of /tmp and can exceed the 2s waitForPendingRun
@@ -357,10 +351,10 @@ async function resolveRun(index: number) {
 }
 
 function syncChangedMessages(): {
-  messages: SyncChangedMessage[];
+  messages: SyncChangedEvent[];
   dispose: () => void;
 } {
-  const messages: SyncChangedMessage[] = [];
+  const messages: SyncChangedEvent[] = [];
   const subscription = assistantEventHub.subscribe({
     type: "process",
     callback: (event) => {
@@ -393,12 +387,12 @@ describe("Conversation queue — slash-like messages pass through to agent loop"
     const events3: ServerMessage[] = [];
 
     // Start first message — blocks on agent loop
-    const p1 = conversation.processMessage(
-      "msg-1",
-      [],
-      (e) => events1.push(e),
-      "req-1",
-    );
+    const p1 = conversation.processMessage({
+      content: "msg-1",
+      attachments: [],
+      onEvent: (e) => events1.push(e),
+      requestId: "req-1",
+    });
     await waitForPendingRun(1);
 
     // Enqueue a slash-like passthrough and a normal passthrough after it.
@@ -438,7 +432,11 @@ describe("Conversation queue — slash-like messages pass through to agent loop"
     const sharedEvents: ServerMessage[] = [];
     const sharedOnEvent = (event: ServerMessage) => sharedEvents.push(event);
 
-    const p1 = conversation.processMessage("msg-1", [], () => {}, "req-1");
+    const p1 = conversation.processMessage({
+      content: "msg-1",
+      attachments: [],
+      requestId: "req-1",
+    });
     await waitForPendingRun(1);
 
     conversation.enqueueMessage({
@@ -477,12 +475,12 @@ describe("Conversation queue — slash-like messages pass through to agent loop"
     const eventsSlash: ServerMessage[] = [];
 
     // Start first message — blocks on agent loop
-    const p1 = conversation.processMessage(
-      "msg-1",
-      [],
-      (e) => events1.push(e),
-      "req-1",
-    );
+    const p1 = conversation.processMessage({
+      content: "msg-1",
+      attachments: [],
+      onEvent: (e) => events1.push(e),
+      requestId: "req-1",
+    });
     await waitForPendingRun(1);
 
     // Enqueue a slash command that matches a skill name — still passes through
@@ -521,7 +519,11 @@ describe("Conversation queue — slash-like messages pass through to agent loop"
     const eventsBye: ServerMessage[] = [];
 
     // Start in-flight message
-    const p1 = conversation.processMessage("msg-1", [], () => {}, "req-1");
+    const p1 = conversation.processMessage({
+      content: "msg-1",
+      attachments: [],
+      requestId: "req-1",
+    });
     await waitForPendingRun(1);
 
     // Enqueue ["hi", "/compact", "bye"]. /compact is non-passthrough, so the
@@ -576,7 +578,11 @@ describe("Conversation queue — slash-like messages pass through to agent loop"
     const sync = syncChangedMessages();
     try {
       await expect(
-        conversation.processMessage("/compact", [], () => {}, "req-compact"),
+        conversation.processMessage({
+          content: "/compact",
+          attachments: [],
+          requestId: "req-compact",
+        }),
       ).rejects.toThrow("compaction failed");
 
       await waitFor(

package/src/__tests__/conversation-slash-unknown.test.ts

@@ -1,10 +1,7 @@
 import { beforeEach, describe, expect, mock, test } from "bun:test";
 
-import type {
-  AgentEvent,
-  CheckpointDecision,
-  CheckpointInfo,
-} from "../agent/loop.js";
+import { CompactionCircuit } from "../agent/compaction-circuit.js";
+import type { AgentEvent } from "../agent/loop.js";
 import type { ServerMessage } from "../daemon/message-protocol.js";
 import type { Message, ProviderResponse } from "../providers/types.js";
 
@@ -237,6 +234,7 @@ let agentLoopRunCalled = false;
 
 mock.module("../agent/loop.js", () => ({
   AgentLoop: class {
+    compactionCircuit = new CompactionCircuit("test-conv");
     constructor() {}
     getToolTokenBudget() {
       return 0;
@@ -250,11 +248,6 @@ mock.module("../agent/loop.js", () => ({
     async run(
       messages: Message[],
       onEvent: (event: AgentEvent) => void,
-      _signal?: AbortSignal,
-      _requestId?: string,
-      _onCheckpoint?: (
-        checkpoint: CheckpointInfo,
-      ) => CheckpointDecision | Promise<CheckpointDecision>,
     ): Promise<Message[]> {
       // Prime the assistant row anchor — production code emits this from
       // `AgentLoop.run` just before `provider.sendMessage`.
@@ -318,9 +311,9 @@ function makeConversation(): Conversation {
     "conv-1",
     provider,
     "system prompt",
-    4096,
     () => {},
     "/tmp",
+    { maxTokens: 4096 },
   );
 }
 
@@ -336,7 +329,10 @@ describe("Conversation slash command — passthrough for unknown tokens", () =>
 
   test("unknown slash-like input passes through to agent loop", async () => {
     const conversation = makeConversation();
-    await conversation.processMessage("/not-a-skill", [], () => {});
+    await conversation.processMessage({
+      content: "/not-a-skill",
+      attachments: [],
+    });
 
     // Should go through the normal agent loop path
     expect(agentLoopRunCalled).toBe(true);
@@ -345,9 +341,11 @@ describe("Conversation slash command — passthrough for unknown tokens", () =>
   test("normal messages still go through standard path", async () => {
     const conversation = makeConversation();
     const events: ServerMessage[] = [];
-    await conversation.processMessage("hello world", [], (msg) =>
-      events.push(msg),
-    );
+    await conversation.processMessage({
+      content: "hello world",
+      attachments: [],
+      onEvent: (msg) => events.push(msg),
+    });
     expect(agentLoopRunCalled).toBe(true);
   });
 });

package/src/__tests__/conversation-speed-override.test.ts

@@ -7,12 +7,8 @@
  */
 import { describe, expect, mock, test } from "bun:test";
 
-import type {
-  AgentEvent,
-  AgentLoopConfig,
-  CheckpointDecision,
-  CheckpointInfo,
-} from "../agent/loop.js";
+import { CompactionCircuit } from "../agent/compaction-circuit.js";
+import type { AgentEvent, AgentLoopConfig } from "../agent/loop.js";
 import type { ServerMessage } from "../daemon/message-protocol.js";
 import type { Message, ProviderResponse } from "../providers/types.js";
 
@@ -201,12 +197,13 @@ let lastAgentLoopConfig: Partial<AgentLoopConfig> | undefined;
 
 mock.module("../agent/loop.js", () => ({
   AgentLoop: class {
+    compactionCircuit = new CompactionCircuit("test-conv");
     constructor(
       _provider: unknown,
       _systemPrompt: string,
-      config?: Partial<AgentLoopConfig>,
+      options?: { config?: Partial<AgentLoopConfig> },
     ) {
-      lastAgentLoopConfig = config;
+      lastAgentLoopConfig = options?.config;
     }
     getToolTokenBudget() {
       return 0;
@@ -220,11 +217,6 @@ mock.module("../agent/loop.js", () => ({
     async run(
       _messages: Message[],
       _onEvent: (event: AgentEvent) => void,
-      _signal?: AbortSignal,
-      _requestId?: string,
-      _onCheckpoint?: (
-        checkpoint: CheckpointInfo,
-      ) => CheckpointDecision | Promise<CheckpointDecision>,
     ): Promise<Message[]> {
       return [];
     }
@@ -292,11 +284,9 @@ describe("per-conversation speed override", () => {
       "conv-speed-override-1",
       makeProvider(),
       "system prompt",
-      4096,
       makeSendToClient(),
       "/tmp",
-      undefined, // sharedCesClient
-      "standard", // speedOverride
+      { maxTokens: 4096, speedOverride: "standard" },
     );
 
     expect(lastAgentLoopConfig).toBeDefined();
@@ -312,11 +302,9 @@ describe("per-conversation speed override", () => {
       "conv-speed-global-1",
       makeProvider(),
       "system prompt",
-      4096,
       makeSendToClient(),
       "/tmp",
-      undefined, // sharedCesClient
-      // no speedOverride — should fall back to global config "fast"
+      { maxTokens: 4096 },
     );
 
     expect(lastAgentLoopConfig).toBeDefined();
@@ -331,11 +319,9 @@ describe("per-conversation speed override", () => {
       "conv-speed-override-fast-1",
       makeProvider(),
       "system prompt",
-      4096,
       makeSendToClient(),
       "/tmp",
-      undefined, // sharedCesClient
-      "fast", // speedOverride
+      { maxTokens: 4096, speedOverride: "fast" },
     );
 
     expect(lastAgentLoopConfig).toBeDefined();