@tyvm/knowhow 0.0.100 → 0.0.101

package/jest.manual.config.js

@@ -0,0 +1,19 @@
+module.exports = {
+  extensionsToTreatAsEsm: ['.ts'],
+  moduleNameMapper: {
+    '^(\\.{1,2}/.*)\\.js$': '$1',
+  },
+  transform: {
+    '^.+\\.ts?$': [
+      'ts-jest',
+      {
+        useESM: true,
+      },
+    ],
+  },
+  testEnvironment: 'node',
+  testRegex: '/tests/manual/.*\.(test|spec)\.(ts|tsx|js)$',
+  moduleFileExtensions: ['ts', 'tsx', 'js', 'jsx', 'json', 'node'],
+  modulePathIgnorePatterns: ["ts_build", "benchmarks"],
+  testPathIgnorePatterns: [],
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tyvm/knowhow",
-  "version": "0.0.100",
+  "version": "0.0.101",
   "description": "ai cli with plugins and agents",
   "main": "ts_build/src/index.js",
   "bin": {
@@ -9,6 +9,7 @@
   "scripts": {
     "test": "jest --testTimeout 300000",
     "test:debug": "node --inspect-brk ../../node_modules/jest/bin/jest.js --detectOpenHandles --forceExit --testTimeout 300000",
+    "build": "npm run compile",
     "compile": "npm run clean && tsc",
     "clean": "rm -rf ts_build",
     "node:build": "bash scripts/build-for-node.sh",

package/src/agents/base/base.ts

@@ -619,6 +619,17 @@ export abstract class BaseAgent implements IAgent {
         tool_choice: "auto",
       });
 
+      // If the agent was paused while the completion was in-flight, wait here
+      // before processing tool calls. This allows the user to send messages
+      // (via addPendingUserMessage) and prevents the agent from proceeding to
+      // tool calls (e.g. finalAnswer) without seeing those interactions.
+      if (this.status === this.eventTypes.pause) {
+        this.log(
+          "Agent was paused after completion, waiting before processing tool calls"
+        );
+        await this.unpaused();
+      }
+
       if (response?.usd_cost === undefined) {
         this.log(
           `Response cost is undefined: ${JSON.stringify(response, null, 2)}`,
@@ -662,6 +673,13 @@ export abstract class BaseAgent implements IAgent {
           this.updateCurrentThread(messages);
 
           for (const toolCall of toolCalls) {
+            if(this.status === this.eventTypes.pause) {
+              this.log(
+                "Agent was paused before tool call, waiting before processing tool calls"
+              );
+              await this.unpaused();
+            }
+
             const toolMessages = await this.processToolMessages(toolCall);
             // Add the tool responses to the thread
             messages.push(...(toolMessages as Message[]));
@@ -676,6 +694,18 @@ export abstract class BaseAgent implements IAgent {
             );
 
             if (finalMessage) {
+              // If user added pending messages after finalAnswer was called,
+              // continue running to respond to that feedback instead of returning
+              if (this.pendingUserMessages.length > 0) {
+                this.log(
+                  "finalAnswer called but pending user messages exist, continuing to respond to feedback"
+                );
+                messages.push(...this.pendingUserMessages);
+                this.pendingUserMessages = [];
+                this.updateCurrentThread(messages);
+                return this.call(userInput, messages);
+              }
+
               // Emit task completion event for plugins (like GitPlugin)
               this.events.emit(this.eventTypes.agentTaskComplete, {
                 taskId:
@@ -685,6 +715,8 @@ export abstract class BaseAgent implements IAgent {
                 result: finalMessage.content || "Done",
               });
               const doneMsg = finalMessage.content || "Done";
+
+
               this.agentEvents.emit(this.eventTypes.done, doneMsg);
               this.status = this.eventTypes.done;
               return doneMsg;

package/src/cli.ts

@@ -38,6 +38,20 @@ import { readPromptFile } from "./ai";
 import { SetupModule } from "./chat/modules/SetupModule";
 import { CliChatService } from "./chat/CliChatService";
 
+// Handle unhandled promise rejections gracefully — particularly from MCP SDK
+// which fires errors via event emitters that can bypass Promise.allSettled.
+// Without this, a single failing MCP server (e.g. expired Notion token) will
+// crash the entire CLI with an unhandled rejection.
+process.on("unhandledRejection", (reason: unknown) => {
+  const message =
+    reason instanceof Error ? reason.message : String(reason);
+  // Only warn — don't exit. The MCP connect errors are recoverable;
+  // the server will simply be unavailable but others continue working.
+  console.warn(
+    `⚠ Unhandled MCP/async error (non-fatal): ${message}`
+  );
+});
+
 async function setupServices() {
   const { Agents, Mcp, Clients, Tools: OldTools } = services();
   const Tools = new LazyToolsService(); // eslint-disable-line no-shadow
@@ -74,7 +88,14 @@ async function setupServices() {
   Agents.setAgentContext(agentContext);
 
   console.log("🔌 Connecting to MCP...");
-  await Mcp.connectToConfigured(Tools);
+  try {
+    await Mcp.connectToConfigured(Tools);
+  } catch (mcpError) {
+    const msg = mcpError instanceof Error ? mcpError.message : String(mcpError);
+    console.warn(
+      `⚠ Some MCP servers failed to connect (continuing without them): ${msg}`
+    );
+  }
   console.log("Connecting to clients...");
   await Clients.registerConfiguredModels();
   console.log("✓ Services are set up and ready to go!");

package/src/clients/contextLimits.ts

@@ -52,7 +52,6 @@ export const ContextLimits: Record<string, number> = {
   [Models.anthropic.Haiku4_5]: 200_000,
   [Models.anthropic.Sonnet3_7]: 200_000,
   [Models.anthropic.Sonnet3_5]: 200_000,
-  [Models.anthropic.Haiku3_5]: 200_000,
   [Models.anthropic.Opus3]: 200_000,
   [Models.anthropic.Haiku3]: 200_000,
 
@@ -74,7 +73,6 @@ export const ContextLimits: Record<string, number> = {
   [Models.google.Gemini_25_Pro_TTS]: 1_000_000,
   [Models.google.Gemini_20_Flash]: 1_000_000,
   [Models.google.Gemini_20_Flash_Preview_Image_Generation]: 1_000_000,
-  [Models.google.Gemini_20_Flash_Lite]: 1_000_000,
   [Models.google.Gemini_20_Flash_Live]: 1_000_000,
   [Models.google.Gemini_20_Flash_TTS]: 1_000_000,
   [Models.google.Gemini_15_Flash]: 1_000_000,

package/src/clients/pricing/anthropic.ts

@@ -65,12 +65,6 @@ export const AnthropicTextPricing = {
     cache_hit: 0.3,
     output: 15.0,
   },
-  [Models.anthropic.Haiku3_5]: {
-    input: 0.8,
-    cache_write: 1.0,
-    cache_hit: 0.08,
-    output: 4.0,
-  },
   [Models.anthropic.Opus3]: {
     input: 15.0,
     cache_write: 18.75,

package/src/clients/pricing/google.ts

@@ -174,10 +174,6 @@ export const GeminiPricing: Record<string, GeminiModelPricing> = {
     output: 0.4,
     image_generation: 0.039,
   },
-  [Models.google.Gemini_20_Flash_Lite]: {
-    input: 0.075,
-    output: 0.3,
-  },
 
   // ── Gemini 1.5 (legacy) ───────────────────────────────────────────────────
   [Models.google.Gemini_15_Flash]: {

package/src/clients/pricing/xai.ts

@@ -1,6 +1,17 @@
 import { Models } from "../../types";
 
 export const XaiTextPricing = {
+
+  [Models.xai.Grok_4_20_Reasoning]: {
+    input: 2.0,
+    cache_hit: 0.20,
+    output: 6.0,
+  },
+  [Models.xai.Grok_4_20_NonReasoning]: {
+    input: 2.0,
+    cache_hit: 0.20,
+    output: 6.0,
+  },
   [Models.xai.Grok4_1_Fast_NonReasoning]: {
     input: 0.2,
     cache_hit: 0.05,

package/src/login.ts

@@ -49,20 +49,17 @@ export async function login(jwtFlag?: boolean): Promise<void> {
     );
 
     const config = await getConfig();
-    const proxyUrl = KNOWHOW_API_URL + "/api/proxy";
 
     if (!config.modelProviders) {
       config.modelProviders = [];
     }
 
     const hasProvider = config.modelProviders.find(
-      (provider) => provider.provider === "knowhow" && provider.url === proxyUrl
+      (provider) => provider.provider === "knowhow"
     );
     if (!hasProvider) {
       config.modelProviders.push({
         provider: "knowhow",
-        url: proxyUrl,
-        jwtFile: ".knowhow/.jwt",
       });
 
       await updateConfig(config);

package/src/services/Mcp.ts

@@ -136,13 +136,24 @@ export class McpService {
         if (shouldAutoConnect && !this.connected[index]) {
           console.log(`Connecting to MCP server: ${config.name}`);
           try {
-            await client.connect(this.transports[index]);
+            // Wrap connect in a race with a timeout to prevent hanging,
+            // and use a wrapping Promise to catch errors that may come
+            // from event-emitter callbacks (unhandled rejection pattern in MCP SDK)
+            await new Promise<void>(async (resolve, reject) => {
+              try {
+                await client.connect(this.transports[index]);
+                resolve();
+              } catch (err) {
+                reject(err);
+              }
+            });
           } catch (error) {
             console.error(
               `Failed to connect to MCP server '${config.name}':`,
               error.message || error
             );
-            throw error; // Re-throw to mark as rejected in Promise.allSettled
+            // Don't re-throw — just log and continue so other servers can still connect
+            return;
           }
           this.connected[index] = true;
         } else if (!shouldAutoConnect) {
@@ -155,7 +166,10 @@ export class McpService {
 
     // Log summary of auto-connection results
     const successful = results.filter((r) => r.status === "fulfilled").length;
-    const failed = results.filter((r) => r.status === "rejected").length;
+    // Since we no longer re-throw, count servers that are NOT connected after the attempt
+    const failed = this.clients.filter(
+      (_, i) => this.config[i]?.autoConnect !== false && !this.connected[i]
+    ).length;
     if (failed > 0) {
       console.warn(
         `Auto-connected ${successful}/${this.clients.length} MCP servers (${failed} failed)`

package/src/types.ts

@@ -89,9 +89,9 @@ export type Config = {
     auth?: {
       required?: boolean;
       passkey?: {
-        publicKey?: string;       // base64-encoded public key
-        credentialId?: string;    // base64-encoded credential ID
-        algorithm?: string;       // e.g. "ES256"
+        publicKey?: string; // base64-encoded public key
+        credentialId?: string; // base64-encoded credential ID
+        algorithm?: string; // e.g. "ES256"
       };
       sessionDurationHours?: number;
     };
@@ -187,19 +187,20 @@ export const Models = {
   anthropic: {
     Opus4_6: "claude-opus-4-6",
     Sonnet4_6: "claude-sonnet-4-6",
-    Opus4_5: "claude-opus-4-5-20251101",
-    Opus4: "claude-opus-4-20250514",
-    Opus4_1: "claude-opus-4-1-20250805",
-    Sonnet4_5: "claude-sonnet-4-5-20250929",
-    Haiku4_5: "claude-haiku-4-5-20251001",
-    Sonnet4: "claude-sonnet-4-20250514",
-    Sonnet3_7: "claude-3-7-sonnet-20250219",
-    Sonnet3_5: "claude-3-5-sonnet-20241022",
-    Haiku3_5: "claude-3-5-haiku-20241022",
-    Opus3: "claude-3-opus-20240229",
-    Haiku3: "claude-3-haiku-20240307",
+    Opus4_5: "claude-opus-4-5",
+    Opus4: "claude-opus-4",
+    Opus4_1: "claude-opus-4-1",
+    Sonnet4_5: "claude-sonnet-4-5",
+    Haiku4_5: "claude-haiku-4-5",
+    Sonnet4: "claude-sonnet-4",
+    Sonnet3_7: "claude-3-7-sonnet",
+    Sonnet3_5: "claude-3-5-sonnet",
+    Opus3: "claude-3-opus",
+    Haiku3: "claude-3-haiku",
   },
   xai: {
+    Grok_4_20_Reasoning: "grok-4.20-0309-reasoning",
+    Grok_4_20_NonReasoning: "grok-4.20-0309-non-reasoning",
     Grok4_1_Fast_Reasoning: "grok-4-1-fast-reasoning",
     Grok4_1_Fast_NonReasoning: "grok-4-1-fast-non-reasoning",
     GrokCodeFast: "grok-code-fast-1",
@@ -277,13 +278,13 @@ export const Models = {
     Gemini_25_Pro_Preview: "gemini-2.5-pro-preview-05-06",
     Gemini_25_Flash_Image: "gemini-2.5-flash-image",
     Gemini_25_Flash_Live: "gemini-2.5-flash-live-preview",
-    Gemini_25_Flash_Native_Audio: "gemini-2.5-flash-native-audio-preview-12-2025",
+    Gemini_25_Flash_Native_Audio:
+      "gemini-2.5-flash-native-audio-preview-12-2025",
     Gemini_25_Pro_TTS: "gemini-2.5-pro-preview-tts",
     // Gemini 2.0 (deprecated)
     Gemini_20_Flash: "gemini-2.0-flash",
     Gemini_20_Flash_Preview_Image_Generation:
       "gemini-2.0-flash-exp-image-generation",
-    Gemini_20_Flash_Lite: "gemini-2.0-flash-lite",
     // Gemini 1.5 (legacy)
     Gemini_15_Flash: "gemini-1.5-flash",
     Gemini_15_Flash_8B: "gemini-1.5-flash-8b",
@@ -380,7 +381,6 @@ export const GoogleReasoningModels = [
   Models.google.Gemini_25_Flash_Preview,
   Models.google.Gemini_25_Pro_Preview,
   Models.google.Gemini_20_Flash,
-  Models.google.Gemini_20_Flash_Lite,
   Models.google.Gemini_15_Flash,
   Models.google.Gemini_15_Flash_8B,
   Models.google.Gemini_15_Pro,