@vellumai/assistant 0.6.5 → 0.6.6

package/src/daemon/conversation-tool-setup.ts

@@ -121,6 +121,17 @@ export interface ToolSetupContext extends SurfaceConversationContext {
   cesClient?: CesClient;
   /** The interface ID of the connected client driving the current turn (e.g. "macos", "chrome-extension"). Propagated into ToolContext for browser backend selection. */
   readonly transportInterface?: InterfaceId;
+  /**
+   * Registry-routed sender override for the host browser proxy. When set
+   * (non-undefined), indicates the browser proxy sender was replaced by a
+   * ChromeExtensionRegistry WebSocket sender — i.e., an extension connection
+   * was actively wired at turn-start. Propagated into ToolContext as
+   * `hostBrowserRegistryRouted` (boolean) so the CDP factory can distinguish
+   * SSE-backed proxies from extension-backed proxies.
+   */
+  hostBrowserSenderOverride?: (
+    msg: import("./message-protocol.js").ServerMessage,
+  ) => void;
   /** Turn-scoped flag: true when any tool call in the current turn received explicit user approval via interactive prompt. Cleared at turn end. */
   approvedViaPromptThisTurn?: boolean;
 }
@@ -159,6 +170,7 @@ export function createToolExecutor(
   input: Record<string, unknown>,
   onOutput?: (chunk: string) => void,
   toolUseId?: string,
+  turnContext?: import("../plugins/types.js").TurnContext,
 ) => Promise<ToolExecutionResult> {
   // Register the conversation's sendToClient for browser screencast surface messages
   registerConversationSender(ctx.conversationId, (msg) =>
@@ -170,6 +182,7 @@ export function createToolExecutor(
     input: Record<string, unknown>,
     onOutput?: (chunk: string) => void,
     toolUseId?: string,
+    turnContext?: import("../plugins/types.js").TurnContext,
   ) => {
     if (isDoordashCommand(name, input)) {
       markDoordashStepInProgress(ctx, input);
@@ -229,6 +242,7 @@ export function createToolExecutor(
       isPlatformHosted: getIsPlatform(),
       cesClient: ctx.cesClient,
       transportInterface: ctx.transportInterface,
+      hostBrowserRegistryRouted: !!ctx.hostBrowserSenderOverride,
       onToolLifecycleEvent: handleToolLifecycleEvent,
       sendToClient: (msg) => {
         // Tool context's sendToClient uses a loose { type: string; [key: string]: unknown }
@@ -306,7 +320,12 @@ export function createToolExecutor(
         };
       }
 
-      const result = await executor.execute(toolName, toolInput, toolContext);
+      const result = await executor.execute(
+        toolName,
+        toolInput,
+        toolContext,
+        turnContext,
+      );
       if (toolContext.approvedViaPrompt) {
         ctx.approvedViaPromptThisTurn = true;
       }
@@ -319,7 +338,12 @@ export function createToolExecutor(
       return result;
     }
 
-    const result = await executor.execute(name, input, toolContext);
+    const result = await executor.execute(
+      name,
+      input,
+      toolContext,
+      turnContext,
+    );
     if (toolContext.approvedViaPrompt) {
       ctx.approvedViaPromptThisTurn = true;
     }
@@ -364,10 +388,23 @@ export function createProxyApprovalCallback(
 
     const input: Record<string, unknown> = {
       url,
-      proxy_session_id: request.sessionId,
+      scheme,
     };
+    if (request.method) {
+      input.method = request.method;
+    }
+    if (request.requestHeaders && Object.keys(request.requestHeaders).length) {
+      input.request_headers = request.requestHeaders;
+    }
+    input.reason =
+      decision.kind === "ask_missing_credential"
+        ? "A known credential template matches this host, but no credential is bound to this session. Approving will forward the request as-is — the assistant won't inject a credential, but any caller-supplied auth headers will still be sent."
+        : "This host isn't covered by any known credential template. Approving will forward the request as-is, including any caller-supplied auth headers.";
     if (decision.kind === "ask_missing_credential") {
-      input.matching_patterns = decision.matchingPatterns;
+      input.known_credential_patterns = decision.matchingPatterns;
+    }
+    if (!request.method) {
+      input.connection_detail_available = "no";
     }
 
     const riskLevel: string = "medium";

package/src/daemon/conversation.ts

@@ -5,7 +5,7 @@
  * - conversation-lifecycle.ts    — loadFromDb, abort, dispose
  * - conversation-messaging.ts    — enqueueMessage, persistUserMessage, redirectToSecurePrompt
  * - conversation-agent-loop.ts   — runAgentLoop, generateTitle
- * - conversation-notifiers.ts    — watch/call notifier registration
+ * - conversation-notifiers.ts    — call notifier registration
  * - conversation-tool-setup.ts   — tool definitions, executor, resolveTools callback
  * - conversation-media-retry.ts  — media trimming + raceWithTimeout
  * - conversation-process.ts      — drainQueue, processMessage
@@ -43,12 +43,9 @@ import {
   ToolProfiler,
 } from "../events/tool-profiling-listener.js";
 import { registerToolTraceListener } from "../events/tool-trace-listener.js";
-import { getHookManager } from "../hooks/manager.js";
-import { enqueueAutoAnalysisOnCompaction } from "../memory/auto-analysis-enqueue.js";
 import { resolveCanonicalGuardianRequest } from "../memory/canonical-guardian-store.js";
 import {
   getConversationOriginChannel,
-  updateConversationContextWindow,
   updateConversationHostAccess,
 } from "../memory/conversation-crud.js";
 import { ConversationGraphMemory } from "../memory/graph/conversation-graph-memory.js";
@@ -76,6 +73,7 @@ import type { AbortReason } from "../util/abort-reasons.js";
 import { getLogger } from "../util/logger.js";
 import type { AssistantAttachmentDraft } from "./assistant-attachments.js";
 import {
+  applyCompactionResult,
   runAgentLoopImpl,
   trackCompactionOutcome,
 } from "./conversation-agent-loop.js";
@@ -365,6 +363,7 @@ export class Conversation {
     toolName?: string,
     toolUseId?: string,
   ) => void;
+  private cacheWarmAbort?: AbortController;
 
   constructor(
     conversationId: string,
@@ -430,7 +429,7 @@ export class Conversation {
       broadcastToAllClients,
     );
 
-    // Register watch/call notifiers (reads ctx properties lazily)
+    // Register call notifiers (reads ctx properties lazily)
     registerConversationNotifiers(conversationId, this);
 
     // Tool infrastructure
@@ -543,11 +542,6 @@ export class Conversation {
       config: llmDefault.contextWindow,
       toolTokenBudget: this.agentLoop.getToolTokenBudget(),
     });
-
-    void getHookManager().trigger("conversation-start", {
-      conversationId: this.conversationId,
-      workingDir: this.workingDir,
-    });
   }
 
   // ── Onboarding context ───────────────────────────────────────────
@@ -560,6 +554,61 @@ export class Conversation {
     return this.onboardingContext;
   }
 
+  // ── Prompt Cache Warming ─────────────────────────────────────────
+
+  /**
+   * Fire-and-forget LLM call with max_tokens=1 to populate the provider's
+   * prompt cache (system prompt + tools). Called after the canned first
+   * greeting so the user's next real message gets a cache hit.
+   */
+  warmPromptCache(): void {
+    this.cacheWarmAbort?.abort();
+    const abort = new AbortController();
+    this.cacheWarmAbort = abort;
+
+    const systemPrompt = this.hasSystemPromptOverride
+      ? this.systemPrompt
+      : (() => {
+          const persona = resolvePersonaContext(
+            this.currentTurnTrustContext,
+            this.currentTurnChannelCapabilities,
+          );
+          return buildSystemPrompt({
+            hasNoClient: this.hasNoClient,
+            userPersona: persona.userPersona,
+            channelPersona: persona.channelPersona,
+            userSlug: persona.userSlug,
+            onboardingContext: this.getOnboardingContext(),
+          });
+        })();
+    const tools = buildToolDefinitions();
+    const provider = this.provider;
+
+    const warmMessage: Message = {
+      role: "user",
+      content: [{ type: "text", text: "hi" }],
+    };
+
+    provider
+      .sendMessage([warmMessage], tools, systemPrompt, {
+        config: { max_tokens: 1, callSite: "mainAgent" },
+        signal: abort.signal,
+      })
+      .then(() => {
+        log.info("Prompt cache warmed successfully");
+      })
+      .catch((err) => {
+        if (!abort.signal.aborted) {
+          log.warn({ err }, "Prompt cache warming failed (non-fatal)");
+        }
+      })
+      .finally(() => {
+        if (this.cacheWarmAbort === abort) {
+          this.cacheWarmAbort = undefined;
+        }
+      });
+  }
+
   // ── Lifecycle ────────────────────────────────────────────────────
 
   async loadFromDb(): Promise<void> {
@@ -671,13 +720,16 @@ export class Conversation {
   }
 
   /**
-   * Restore host browser proxy availability only. Used for interfaces that
-   * support host_browser but not the full desktop proxy set, so calling
-   * restoreProxyAvailability() would incorrectly re-enable bash/file/CU
-   * proxies that should stay disabled. Applicable to chrome-extension turns
-   * (which only support host_browser) and macOS turns with an active
-   * extension connection (which route browser tools through the extension
-   * registry instead of cdp-inspect/local).
+   * Restore host browser proxy availability only. Used in two scenarios:
+   *
+   * 1. **Chrome-extension turns** — only support host_browser (not the full
+   *    desktop proxy set), so calling restoreProxyAvailability() would
+   *    incorrectly re-enable bash/file/CU proxies.
+   * 2. **macOS turns** — when called from queue-drain, the browser proxy
+   *    sender needs to be either the registry-routed sender (when an
+   *    extension connection is present) or the SSE hub sender (when no
+   *    extension is connected). This helper resolves the correct sender
+   *    via `hostBrowserSenderOverride ?? sendToClient`.
    *
    * Unlike `restoreProxyAvailability()`, this helper does NOT gate on
    * `hasNoClient`. The chrome-extension interface is non-interactive (so
@@ -693,7 +745,9 @@ export class Conversation {
    * WebSocket rather than the SSE hub: if the queue-drain path called this
    * helper with `sendToClient`, the registry-routed sender established at
    * turn-start would be clobbered by the SSE hub emitter and
-   * host_browser_request frames would stop reaching the extension.
+   * host_browser_request frames would stop reaching the extension. When
+   * no override is set (macOS without extension), `sendToClient` is used
+   * so frames reach the desktop client via SSE.
    *
    * Callers must only invoke this when they know the current interface
    * supports host_browser (see `supportsHostProxy(id, "host_browser")`)
@@ -1249,24 +1303,14 @@ export class Conversation {
     // is `undefined` on early-return paths (no eligible messages, disabled,
     // etc.) — skip those so they don't silently reset the counter.
     if (result.summaryFailed !== undefined) {
-      trackCompactionOutcome(this, result.summaryFailed, this.sendToClient);
+      await trackCompactionOutcome(
+        this,
+        result.summaryFailed,
+        this.sendToClient,
+      );
     }
     if (result.compacted) {
-      this.messages = result.messages;
-      this.contextCompactedMessageCount += result.compactedPersistedMessages;
-      this.contextCompactedAt = Date.now();
-      updateConversationContextWindow(
-        this.conversationId,
-        result.summaryText,
-        this.contextCompactedMessageCount,
-      );
-      // Fire auto-analysis on compaction so the reflective agent can
-      // crystallize anything worth remembering before the context window
-      // narrows further.
-      enqueueAutoAnalysisOnCompaction(
-        this.conversationId,
-        this.trustContext?.trustClass,
-      );
+      applyCompactionResult(this, result, this.sendToClient, null);
     }
     return result;
   }
@@ -1416,7 +1460,6 @@ export class Conversation {
     userMessageId: string,
     onEvent: (msg: ServerMessage) => void,
     options?: {
-      skipPreMessageRollback?: boolean;
       isInteractive?: boolean;
       isUserMessage?: boolean;
       titleText?: string;
@@ -1440,6 +1483,8 @@ export class Conversation {
     options?: { isInteractive?: boolean; callSite?: LLMCallSite },
     displayContent?: string,
   ): Promise<string> {
+    this.cacheWarmAbort?.abort();
+    this.cacheWarmAbort = undefined;
     return processMessageImpl(
       this as ProcessConversationContext,
       content,