@mindstudio-ai/agent 0.1.137 → 0.1.138

package/dist/cli.js

@@ -5191,7 +5191,7 @@ function logTaskResult(result) {
   }
   const toolSummary = result.toolCalls.map((tc) => `${tc.name} (${tc.durationMs}ms) ${tc.success ? "\u2713" : "\u2717"}`).join(", ");
   console.log(
-    `[task] ${result.turns} turn${result.turns === 1 ? "" : "s"}` + (toolSummary ? `: ${toolSummary}` : "") + ` | ${result.parsedSuccessfully ? "output OK" : "\u26A0 output not valid JSON"} | cost: ${result.usage.totalBillingCost}`
+    `[task] ${result.turns} turn${result.turns === 1 ? "" : "s"}` + (toolSummary ? `: ${toolSummary}` : "") + ` | ${result.parsedSuccessfully ? "output OK" : "\u26A0 output not valid JSON"} | cost: ${result.usage.totalBillingCost} | trace: ${result.traceId}`
   );
 }
 async function runTaskPoll(httpConfig, body) {
@@ -5686,7 +5686,8 @@ ${structuredOutputExample}
         errors,
         turns: loopCount,
         usage: totalUsage,
-        toolCalls: toolCallLog
+        toolCalls: toolCallLog,
+        traceId
       }
     );
   };
@@ -5700,7 +5701,12 @@ ${structuredOutputExample}
           `[task] ${err.message}`,
           "task_execution_error",
           500,
-          { turns: loopCount, usage: totalUsage, toolCalls: toolCallLog }
+          {
+            turns: loopCount,
+            usage: totalUsage,
+            toolCalls: toolCallLog,
+            traceId
+          }
         );
       }
       return finish(buildResult(null, "", false, loopCount));
@@ -5709,7 +5715,8 @@ ${structuredOutputExample}
       throw new MindStudioError(
         `[task] ${err.message}`,
         err.errorCode ?? "task_turn_error",
-        err.status ?? 500
+        err.status ?? 500,
+        { traceId }
       );
     }
     throw err;
@@ -5773,9 +5780,13 @@ ${structuredOutputExample}
                 isError: true
               };
             }
+          } else if (kind === "method") {
+            result = await deps.executeMethodTool(toolCall.name, mergedInput, {
+              traceId,
+              toolCallId: toolCall.id
+            });
           } else {
-            const execute = kind === "method" ? deps.executeMethodTool : deps.executeStepTool;
-            result = await execute(toolCall.name, mergedInput);
+            result = await deps.executeStepTool(toolCall.name, mergedInput);
           }
           toolCallLog.push({
             name: toolCall.name,
@@ -7086,7 +7097,7 @@ var init_client = __esm({
                   };
                 }
               },
-              executeMethodTool: async (methodId, input) => {
+              executeMethodTool: async (methodId, input, call) => {
                 try {
                   const { data } = await request(
                     { ...httpConfig, maxRetries: 0 },
@@ -7094,7 +7105,9 @@ var init_client = __esm({
                     "/task/invoke-method",
                     {
                       methodId,
-                      input
+                      input,
+                      traceId: call.traceId,
+                      toolCallId: call.toolCallId
                     }
                   );
                   if (data.error) {
@@ -8627,7 +8640,7 @@ function buildReferenceDocs(data) {
 
   User-typed columns: The \`User\` type is just \`string\` at runtime. Write plain UUIDs to user-typed columns \u2014 do NOT add a \`@@user@@\` prefix manually and do NOT use \`as any\`. The SDK adds and strips the prefix automatically on read/write. If TypeScript complains about assigning a plain string to a \`User\` field, use \`db.userRef(id)\` to type-cast cleanly. Reads always return bare UUIDs; use \`resolveUser(id)\` when you need display info (name, email, avatar). Never do your own prefix handling \u2014 it will fight the SDK and cause double-prefixing or orphan refs. The \`@@user@@\` prefix is only visible in storage if you bypass the SDK with raw SQL; app code should never see or produce it.
 
-  Task agents: For multi-step tasks requiring autonomous tool use, use \`runTask()\`. Provide a prompt, input, a set of tools, an output contract, and a model. The SDK runs a tool-use loop and returns structured output. Prefer \`outputSchema\` for the output contract: plain JSON Schema (\`type\`/\`properties\`/\`required\`/\`enum\`/\`items\`; nullable via type arrays like \`["string","null"]\` \u2014 never \`nullable: true\`, never \`oneOf\`/\`$ref\`). Output is validated every turn with automatic repair and \`result.output\` is typed by inference from the schema \u2014 no generic argument and no \`parsedSuccessfully\` check needed; the call throws \`MindStudioError\` code \`task_output_schema_mismatch\` instead of ever returning nonconforming output. The legacy \`structuredOutputExample\` option (example object the model imitates) still works but is unvalidated \u2014 always check \`result.parsedSuccessfully\` before using its output. \`result.traceId\` is a stable id for the task's model transcript \u2014 inside a jewel, attach it to the propose/grade return (\`trace: task.traceId\`) to preserve the transcript with the pair. Tools can include any SDK action \u2014 e.g. \`['searchGoogle', 'fetchUrl', { method: 'generateImage', defaults: { imageModelOverride: { model: 'seedream-4.5' } } }]\` \u2014 plus the app's own methods (\`{ appMethod: 'save-vendor', description: 'when to use it' }\`, run as the user who invoked the method that started the task) and inline functions (\`{ name, description, inputSchema, execute }\` \u2014 the function runs in the calling process; thrown errors go back to the model as tool output; no \`defaults\`). Supports SSE streaming via \`onEvent\` callback.
+  Task agents: For multi-step tasks requiring autonomous tool use, use \`runTask()\`. Provide a prompt, input, a set of tools, an output contract, and a model. The SDK runs a tool-use loop and returns structured output. Prefer \`outputSchema\` for the output contract: plain JSON Schema (\`type\`/\`properties\`/\`required\`/\`enum\`/\`items\`; nullable via type arrays like \`["string","null"]\` \u2014 never \`nullable: true\`, never \`oneOf\`/\`$ref\`). Output is validated every turn with automatic repair and \`result.output\` is typed by inference from the schema \u2014 no generic argument and no \`parsedSuccessfully\` check needed; the call throws \`MindStudioError\` code \`task_output_schema_mismatch\` instead of ever returning nonconforming output. The legacy \`structuredOutputExample\` option (example object the model imitates) still works but is unvalidated \u2014 always check \`result.parsedSuccessfully\` before using its output. \`result.traceId\` is a stable id for the task's model transcript, which the platform records for every task (\`remy-admin tasks get <traceId>\` reads it back) \u2014 inside a jewel, also attach it to the propose/grade return (\`trace: task.traceId\`) to preserve the transcript with the pair. Tools can include any SDK action \u2014 e.g. \`['searchGoogle', 'fetchUrl', { method: 'generateImage', defaults: { imageModelOverride: { model: 'seedream-4.5' } } }]\` \u2014 plus the app's own methods (\`{ appMethod: 'save-vendor', description: 'when to use it' }\`, run as the user who invoked the method that started the task) and inline functions (\`{ name, description, inputSchema, execute }\` \u2014 the function runs in the calling process; thrown errors go back to the model as tool output; no \`defaults\`). \`onEvent\` receives each turn's events (text, thinking, tool calls and results) as the turn completes.
 
   Uploading files / getting a hosted URL: For app file storage (user uploads, generated documents, images, marketing assets), use the \`files\` store \u2014 define a store at module scope and call \`await SomeStore.put(bytes, { contentType?, filename? })\` \u2192 returns \`{ url }\` served on the app's own domain (private by default; use a public store for world-readable assets). \`bytes\` is a \`Buffer\`/\`Uint8Array\`, so read/fetch the content first. The older \`mindstudio.uploadFile(...)\` (and the \`mindstudio upload <filepath>\` CLI) upload to the shared account media CDN and are DEPRECATED for app storage \u2014 they remain only for account-level one-off assets (e.g. an agent avatar for \`changeProfilePicture\`). Actions that generate a file (generateImage, generateVideo, generateSpeech, generatePdf, upscaleImage, \u2026) can write their output directly into a store: pass the store handle as \`store\` in the options object, e.g. \`await mindstudio.generateImage({ prompt }, { store: Assets })\`. It's optional \u2014 omit it and the asset goes to the shared CDN as before. More generally: some SDK capabilities are exposed as \`mindstudio\` CLI commands (list-actions, run-agent, exec, batch, \u2026), not only as code \u2014 reach for the CLI for one-off operations instead of writing and running code.
 
@@ -9234,7 +9247,7 @@ async function startMcpServer(options) {
           capabilities: { tools: {} },
           serverInfo: {
             name: "mindstudio-agent",
-            version: "0.1.137"
+            version: "0.1.138"
           },
           instructions: 'Welcome to MindStudio \u2014 a platform with 200+ AI models, 850+ third-party integrations, and pre-built agents.\n\nGetting started:\n1. Call `ask` with any question about the SDK \u2014 it knows every action, model, and connector and returns working code with real model IDs and config options. Examples: ask("generate an image with FLUX"), ask("what models support vision?"), ask("how do I send a Slack message?").\n2. Call `changeName` to set your display name \u2014 use your name or whatever your user calls you. This is how you\'ll appear in MindStudio request logs.\n3. If you have a profile picture, pass its public image URL to `changeProfilePicture`.\n4. For manual browsing, call `listActions` to discover all available actions.\n\nThen use the tools to generate text, images, video, audio, search the web, work with data sources, run agents, and more.\n\nImportant:\n- AI-powered actions (text generation, image generation, video, audio, etc.) cost money. Before running these, call `estimateActionCost` and confirm with the user before proceeding \u2014 unless they\'ve explicitly told you to go ahead.\n- Not all agents from `listAgents` are configured for API use. Do not try to run an agent just because it appears in the list \u2014 it will likely fail. Only run agents the user specifically asks you to run.'
         });
@@ -10246,7 +10259,7 @@ function isNewerVersion(current, latest) {
   return false;
 }
 async function checkForUpdate() {
-  const currentVersion = "0.1.137";
+  const currentVersion = "0.1.138";
   if (!currentVersion || process.env.MINDSTUDIO_NO_UPDATE_CHECK) {
     return null;
   }
@@ -10281,7 +10294,7 @@ async function checkForUpdate() {
   }
 }
 function printUpdateNotice(latestVersion) {
-  const currentVersion = "0.1.137";
+  const currentVersion = "0.1.138";
   process.stderr.write(
     `
   ${ansi2.cyanBright("Update available")} ${ansi2.gray(currentVersion + " \u2192")} ${ansi2.cyanBold(latestVersion)}
@@ -10294,7 +10307,7 @@ function isStandaloneBinary() {
   return !argv1.includes("node_modules");
 }
 async function cmdUpdate() {
-  const currentVersion = "0.1.137";
+  const currentVersion = "0.1.138";
   process.stderr.write(
     `  ${ansi2.gray("Current version:")} ${currentVersion}
 `
@@ -10435,7 +10448,7 @@ async function cmdLogin(options) {
   process.stderr.write("\n");
   printLogo();
   process.stderr.write("\n");
-  const ver = "0.1.137";
+  const ver = "0.1.138";
   process.stderr.write(
     `  ${ansi2.bold("MindStudio Agent")} ${ver ? " " + ansi2.gray("v" + ver) : ""}
 `
@@ -10763,7 +10776,7 @@ async function main() {
   const updatePromise = command !== "mcp" && command !== "login" && command !== "update" ? checkForUpdate() : Promise.resolve(null);
   try {
     if (command === "version" || command === "-v") {
-      process.stdout.write("0.1.137\n");
+      process.stdout.write("0.1.138\n");
       return;
     }
     if (command === "login") {

package/dist/index.d.ts

@@ -3880,9 +3880,10 @@ interface RunTaskOptionsBase {
     /** Thread ID for state persistence. */
     threadId?: string;
     /**
-     * SSE event callback. When provided, uses streaming mode instead of polling.
-     * Events include text chunks, tool call starts/results, thinking blocks,
-     * errors, and the final done event with the structured output.
+     * Event callback: text chunks, thinking, tool call starts and results,
+     * errors, and the final `done` event with the structured output. A turn's
+     * events arrive together once the turn succeeds, so a retried attempt never
+     * reaches it. It doesn't change how the task runs.
      */
     onEvent?: (event: TaskEvent) => void;
 }
@@ -3927,6 +3928,9 @@ interface TaskToolCall {
 interface TaskUsage {
     inputTokens: number;
     outputTokens: number;
+    /** Prompt tokens written to and read from the model's prompt cache. */
+    cacheCreationTokens?: number;
+    cacheReadTokens?: number;
     /** Total cost in nanodollars (1/1,000,000,000th of a US dollar). */
     totalBillingCost: number;
 }
@@ -3945,11 +3949,12 @@ interface RunTaskResult<T = unknown> {
     /** Summary of every tool call made during execution. */
     toolCalls: TaskToolCall[];
     /**
-     * Stable id for this task's model transcript. Inside a jewel, attach it to
-     * the propose/grade return (`trace: task.traceId`) to preserve the full
-     * transcript with the pair — the training row and the audit trail. The
-     * platform records transcripts server-side per turn; only the id rides the
-     * result.
+     * Stable id for this task's model transcript. The platform records every
+     * task's transcript server-side per turn under it, for 90 days, and only
+     * the id rides the result: `remy-admin tasks get <traceId>` reads it back.
+     * Inside a jewel, also attach it to the propose/grade return
+     * (`trace: task.traceId`) to preserve the full transcript with the pair —
+     * the training row and the audit trail.
      */
     traceId: string;
 }

package/dist/index.js

@@ -4989,7 +4989,7 @@ function logTaskResult(result) {
   }
   const toolSummary = result.toolCalls.map((tc) => `${tc.name} (${tc.durationMs}ms) ${tc.success ? "\u2713" : "\u2717"}`).join(", ");
   console.log(
-    `[task] ${result.turns} turn${result.turns === 1 ? "" : "s"}` + (toolSummary ? `: ${toolSummary}` : "") + ` | ${result.parsedSuccessfully ? "output OK" : "\u26A0 output not valid JSON"} | cost: ${result.usage.totalBillingCost}`
+    `[task] ${result.turns} turn${result.turns === 1 ? "" : "s"}` + (toolSummary ? `: ${toolSummary}` : "") + ` | ${result.parsedSuccessfully ? "output OK" : "\u26A0 output not valid JSON"} | cost: ${result.usage.totalBillingCost} | trace: ${result.traceId}`
   );
 }
 async function runTaskPoll(httpConfig, body) {
@@ -5496,7 +5496,8 @@ ${structuredOutputExample}
         errors,
         turns: loopCount,
         usage: totalUsage,
-        toolCalls: toolCallLog
+        toolCalls: toolCallLog,
+        traceId
       }
     );
   };
@@ -5510,7 +5511,12 @@ ${structuredOutputExample}
           `[task] ${err.message}`,
           "task_execution_error",
           500,
-          { turns: loopCount, usage: totalUsage, toolCalls: toolCallLog }
+          {
+            turns: loopCount,
+            usage: totalUsage,
+            toolCalls: toolCallLog,
+            traceId
+          }
         );
       }
       return finish(buildResult(null, "", false, loopCount));
@@ -5519,7 +5525,8 @@ ${structuredOutputExample}
       throw new MindStudioError(
         `[task] ${err.message}`,
         err.errorCode ?? "task_turn_error",
-        err.status ?? 500
+        err.status ?? 500,
+        { traceId }
       );
     }
     throw err;
@@ -5583,9 +5590,13 @@ ${structuredOutputExample}
                 isError: true
               };
             }
+          } else if (kind === "method") {
+            result = await deps.executeMethodTool(toolCall.name, mergedInput, {
+              traceId,
+              toolCallId: toolCall.id
+            });
           } else {
-            const execute = kind === "method" ? deps.executeMethodTool : deps.executeStepTool;
-            result = await execute(toolCall.name, mergedInput);
+            result = await deps.executeStepTool(toolCall.name, mergedInput);
           }
           toolCallLog.push({
             name: toolCall.name,
@@ -6795,7 +6806,7 @@ var MindStudioAgent = class {
               };
             }
           },
-          executeMethodTool: async (methodId, input) => {
+          executeMethodTool: async (methodId, input, call) => {
             try {
               const { data } = await request(
                 { ...httpConfig, maxRetries: 0 },
@@ -6803,7 +6814,9 @@ var MindStudioAgent = class {
                 "/task/invoke-method",
                 {
                   methodId,
-                  input
+                  input,
+                  traceId: call.traceId,
+                  toolCallId: call.toolCallId
                 }
               );
               if (data.error) {