llmist 5.0.0 → 5.1.0

package/dist/cli.cjs

@@ -9479,13 +9479,24 @@ ${endPrefix}`
             observers: {
               ...hooks?.observers,
               onLLMCallStart: async (context) => {
+                let inputTokens;
+                try {
+                  if (this.client) {
+                    inputTokens = await this.client.countTokens(
+                      context.options.model,
+                      context.options.messages
+                    );
+                  }
+                } catch {
+                }
                 onSubagentEvent({
                   type: "llm_call_start",
                   gadgetInvocationId: invocationId,
                   depth,
                   event: {
                     iteration: context.iteration,
-                    model: context.options.model
+                    model: context.options.model,
+                    inputTokens
                   }
                 });
                 if (existingOnLLMCallStart) {
@@ -9958,7 +9969,7 @@ var import_commander2 = require("commander");
 // package.json
 var package_default = {
   name: "llmist",
-  version: "5.0.0",
+  version: "5.1.0",
   description: "TypeScript LLM client with streaming tool execution. Tools fire mid-stream. Built-in function calling works with any model\u2014no structured outputs or native tool support required.",
   type: "module",
   main: "dist/index.cjs",
@@ -12954,7 +12965,7 @@ function getRawValue(value) {
 function truncateValue(str, maxLen) {
   if (maxLen <= 0) return "";
   if (str.length <= maxLen) return str;
-  return `${str.slice(0, maxLen)}\u2026`;
+  return `${str.slice(0, maxLen - 1)}\u2026`;
 }
 function formatParametersInline(params, maxWidth) {
   if (!params || Object.keys(params).length === 0) {
@@ -12982,6 +12993,11 @@ function formatParametersInline(params, maxWidth) {
           const proportion = v.length / totalRawLength;
           return Math.max(minPerValue, Math.floor(proportion * availableForValues));
         });
+        const totalLimits = limits.reduce((sum, l) => sum + l, 0);
+        if (totalLimits > availableForValues) {
+          const scale = availableForValues / totalLimits;
+          limits = limits.map((l) => Math.max(1, Math.floor(l * scale)));
+        }
       }
     }
   } else {
@@ -12997,8 +13013,8 @@ function formatGadgetLine(info, maxWidth) {
   const gadgetLabel = import_chalk3.default.magenta.bold(info.name);
   const timeStr = `${info.elapsedSeconds.toFixed(1)}s`;
   const timeLabel = import_chalk3.default.dim(timeStr);
-  const fixedLength = 2 + info.name.length + 2 + 1 + timeStr.length;
-  const availableForParams = Math.max(40, terminalWidth - fixedLength - 2);
+  const fixedLength = 3 + info.name.length + 2 + 1 + timeStr.length;
+  const availableForParams = Math.max(40, terminalWidth - fixedLength - 3);
   const paramsStr = formatParametersInline(info.parameters, availableForParams);
   const paramsLabel = paramsStr ? `${import_chalk3.default.dim("(")}${paramsStr}${import_chalk3.default.dim(")")}` : "";
   if (info.error) {
@@ -13008,17 +13024,21 @@ function formatGadgetLine(info, maxWidth) {
   if (!info.isComplete) {
     return `${import_chalk3.default.blue("\u23F5")} ${gadgetLabel}${paramsLabel} ${timeLabel}`;
   }
-  let outputStr;
+  let outputLabel;
   if (info.tokenCount !== void 0 && info.tokenCount > 0) {
-    outputStr = `${formatTokens(info.tokenCount)} tokens`;
+    outputLabel = import_chalk3.default.dim("\u2193") + import_chalk3.default.green(` ${formatTokens(info.tokenCount)} `);
   } else if (info.outputBytes !== void 0 && info.outputBytes > 0) {
-    outputStr = formatBytes(info.outputBytes);
+    outputLabel = import_chalk3.default.green(formatBytes(info.outputBytes)) + " ";
   } else {
-    outputStr = "";
+    outputLabel = "";
   }
   const icon = info.breaksLoop ? import_chalk3.default.yellow("\u23F9") : import_chalk3.default.green("\u2713");
-  const outputLabel = outputStr ? ` ${import_chalk3.default.dim("\u2192")} ${import_chalk3.default.green(outputStr)}` : "";
-  return `${icon} ${gadgetLabel}${paramsLabel}${outputLabel} ${timeLabel}`;
+  const nameRef = import_chalk3.default.magenta(info.name);
+  const line1 = `${icon} ${gadgetLabel}${paramsLabel}`;
+  const line2Prefix = `  ${import_chalk3.default.dim("\u2192")} ${nameRef} ${outputLabel}`;
+  const line2 = `${line2Prefix}${timeLabel}`;
+  return `${line1}
+${line2}`;
 }
 function formatBytes(bytes) {
   if (bytes < 1024) {
@@ -13029,6 +13049,11 @@ function formatBytes(bytes) {
   }
   return `${(bytes / (1024 * 1024)).toFixed(1)} MB`;
 }
+function truncateOutputPreview(output, maxWidth) {
+  const normalized = output.replace(/\s+/g, " ").trim();
+  if (normalized.length <= maxWidth) return normalized;
+  return normalized.slice(0, maxWidth - 1) + "\u2026";
+}
 function getMediaIcon(kind) {
   switch (kind) {
     case "image":
@@ -13056,37 +13081,99 @@ function formatGadgetSummary2(result) {
   const gadgetLabel = import_chalk3.default.magenta.bold(result.gadgetName);
   const timeStr = result.executionTimeMs >= 1e3 ? `${(result.executionTimeMs / 1e3).toFixed(1)}s` : `${Math.round(result.executionTimeMs)}ms`;
   const timeLabel = import_chalk3.default.dim(timeStr);
-  let outputStr;
-  if (result.tokenCount !== void 0 && result.tokenCount > 0) {
-    outputStr = `${formatTokens(result.tokenCount)} tokens`;
-  } else if (result.result) {
+  const fixedLength = 3 + result.gadgetName.length + 2;
+  const availableForParams = Math.max(40, terminalWidth - fixedLength - 3);
+  const paramsStr = formatParametersInline(result.parameters, availableForParams);
+  const paramsLabel = paramsStr ? `${import_chalk3.default.dim("(")}${paramsStr}${import_chalk3.default.dim(")")}` : "";
+  const icon = result.breaksLoop ? import_chalk3.default.yellow("\u23F9") : result.error ? import_chalk3.default.red("\u2717") : import_chalk3.default.green("\u2713");
+  const line1 = `${icon} ${gadgetLabel}${paramsLabel}`;
+  const nameRef = import_chalk3.default.magenta(result.gadgetName);
+  const hasSubagentMetrics = result.subagentMetrics && result.subagentMetrics.callCount > 0;
+  let outputLabel;
+  let outputStrRaw;
+  if (!hasSubagentMetrics && result.tokenCount !== void 0 && result.tokenCount > 0) {
+    const tokenStr = formatTokens(result.tokenCount);
+    outputLabel = import_chalk3.default.dim("\u2193") + import_chalk3.default.green(` ${tokenStr} `);
+    outputStrRaw = `\u2193 ${tokenStr} `;
+  } else if (!hasSubagentMetrics && result.result) {
     const outputBytes = Buffer.byteLength(result.result, "utf-8");
-    outputStr = outputBytes > 0 ? formatBytes(outputBytes) : "no output";
+    if (outputBytes > 0) {
+      const bytesStr = formatBytes(outputBytes);
+      outputLabel = import_chalk3.default.green(bytesStr) + " ";
+      outputStrRaw = bytesStr + " ";
+    } else {
+      outputLabel = "";
+      outputStrRaw = "";
+    }
   } else {
-    outputStr = "no output";
+    outputLabel = "";
+    outputStrRaw = "";
   }
-  const fixedLength = 2 + result.gadgetName.length + 2 + 3 + outputStr.length + 1 + timeStr.length;
-  const availableForParams = Math.max(40, terminalWidth - fixedLength - 2);
-  const paramsStr = formatParametersInline(result.parameters, availableForParams);
-  const paramsLabel = paramsStr ? `${import_chalk3.default.dim("(")}${paramsStr}${import_chalk3.default.dim(")")}` : "";
   if (result.error) {
     const errorMsg = result.error.length > 50 ? `${result.error.slice(0, 50)}\u2026` : result.error;
-    return `${import_chalk3.default.red("\u2717")} ${gadgetLabel}${paramsLabel} ${import_chalk3.default.red("error:")} ${errorMsg} ${timeLabel}`;
+    const line22 = `  ${import_chalk3.default.dim("\u2192")} ${nameRef} ${import_chalk3.default.red("error:")} ${errorMsg} ${timeLabel}`;
+    return `${line1}
+${line22}`;
+  }
+  const previewWidth = Math.floor(terminalWidth * 0.6);
+  const prefixLength = 4 + result.gadgetName.length + 1 + outputStrRaw.length + 1 + timeStr.length + 2;
+  const availablePreview = Math.max(20, previewWidth - prefixLength);
+  let customPreview;
+  if (result.gadgetName === "TodoUpsert" && result.parameters?.content) {
+    const statusEmoji = result.parameters.status === "done" ? "\u2705" : result.parameters.status === "in_progress" ? "\u{1F504}" : "\u2B1C";
+    const content = String(result.parameters.content);
+    customPreview = `${statusEmoji} ${truncateOutputPreview(content, availablePreview - 3)}`;
+  }
+  if (result.gadgetName === "GoogleSearch" && result.parameters?.query) {
+    const query = String(result.parameters.query);
+    const countMatch = result.result?.match(/\((\d+)\s+of\s+[\d,]+\s+results?\)/i) || // "(10 of 36400000 results)"
+    result.result?.match(/(\d+)\s+results?\s+found/i) || // "10 results found"
+    result.result?.match(/found\s+(\d+)\s+results?/i);
+    const count = countMatch?.[1] ?? (result.parameters.maxResults ? String(result.parameters.maxResults) : null);
+    const countStr = count ? ` \u2192 ${count} results` : "";
+    const queryPreview = truncateOutputPreview(query, availablePreview - 5 - countStr.length);
+    customPreview = `\u{1F50D} "${queryPreview}"${countStr}`;
+  }
+  let subagentMetricsStr = "";
+  if (result.subagentMetrics && result.subagentMetrics.callCount > 0) {
+    const parts = [];
+    const m = result.subagentMetrics;
+    if (m.inputTokens > 0) {
+      parts.push(import_chalk3.default.dim("\u2191") + import_chalk3.default.yellow(` ${formatTokens(m.inputTokens)}`));
+    }
+    if (m.cachedInputTokens > 0) {
+      parts.push(import_chalk3.default.dim("\u27F3") + import_chalk3.default.blue(` ${formatTokens(m.cachedInputTokens)}`));
+    }
+    if (m.outputTokens > 0) {
+      parts.push(import_chalk3.default.dim("\u2193") + import_chalk3.default.green(` ${formatTokens(m.outputTokens)}`));
+    }
+    if (m.cost > 0) {
+      parts.push(import_chalk3.default.cyan(`$${formatCost(m.cost)}`));
+    }
+    if (parts.length > 0) {
+      subagentMetricsStr = parts.join(import_chalk3.default.dim(" | ")) + import_chalk3.default.dim(" | ");
+    }
   }
-  const outputLabel = outputStr === "no output" ? import_chalk3.default.dim(outputStr) : import_chalk3.default.green(outputStr);
-  const icon = result.breaksLoop ? import_chalk3.default.yellow("\u23F9") : import_chalk3.default.green("\u2713");
-  let summaryLine = `${icon} ${gadgetLabel}${paramsLabel} ${import_chalk3.default.dim("\u2192")} ${outputLabel} ${timeLabel}`;
+  let line2;
+  const previewContent = customPreview ?? (result.result?.trim() ? truncateOutputPreview(result.result, availablePreview) : null);
+  if (previewContent) {
+    line2 = `  ${import_chalk3.default.dim("\u2192")} ${nameRef} ${outputLabel}${subagentMetricsStr}${timeLabel}${import_chalk3.default.dim(":")} ${import_chalk3.default.dim(previewContent)}`;
+  } else {
+    line2 = `  ${import_chalk3.default.dim("\u2192")} ${nameRef} ${outputLabel}${subagentMetricsStr}${timeLabel}`;
+  }
+  let output = `${line1}
+${line2}`;
   if (result.media && result.media.length > 0) {
     const mediaLines = result.media.map(formatMediaLine);
-    summaryLine += "\n" + mediaLines.join("\n");
+    output += "\n" + mediaLines.join("\n");
   }
   if (result.gadgetName === "TellUser" && result.parameters?.message) {
     const message = String(result.parameters.message);
     const rendered = renderMarkdownWithSeparators(message);
-    return `${summaryLine}
+    return `${output}
 ${rendered}`;
   }
-  return summaryLine;
+  return output;
 }
 
 // src/cli/utils.ts
@@ -13283,14 +13370,15 @@ var StreamProgress = class {
    * Add a nested agent LLM call (called when nested llm_call_start event received).
    * Used to display hierarchical progress for subagent gadgets.
    */
-  addNestedAgent(id, parentInvocationId, depth, model, iteration, inputTokens) {
+  addNestedAgent(id, parentInvocationId, depth, model, iteration, info) {
     this.nestedAgents.set(id, {
       parentInvocationId,
       depth,
       model,
       iteration,
       startTime: Date.now(),
-      inputTokens
+      inputTokens: info?.inputTokens,
+      cachedInputTokens: info?.cachedInputTokens
     });
     if (this.isRunning && this.isTTY) {
       this.render();
@@ -13304,22 +13392,23 @@ var StreamProgress = class {
   updateNestedAgent(id, info) {
     const agent = this.nestedAgents.get(id);
     if (agent) {
-      agent.inputTokens = info.inputTokens;
-      agent.outputTokens = info.outputTokens;
-      agent.cachedInputTokens = info.cachedInputTokens;
-      agent.cacheCreationInputTokens = info.cacheCreationInputTokens;
-      agent.finishReason = info.finishReason;
+      if (info.inputTokens !== void 0) agent.inputTokens = info.inputTokens;
+      if (info.outputTokens !== void 0) agent.outputTokens = info.outputTokens;
+      if (info.cachedInputTokens !== void 0) agent.cachedInputTokens = info.cachedInputTokens;
+      if (info.cacheCreationInputTokens !== void 0)
+        agent.cacheCreationInputTokens = info.cacheCreationInputTokens;
+      if (info.finishReason !== void 0) agent.finishReason = info.finishReason;
       if (info.cost !== void 0) {
         agent.cost = info.cost;
-      } else if (this.modelRegistry && agent.model && info.outputTokens) {
+      } else if (this.modelRegistry && agent.model && agent.outputTokens) {
         try {
           const modelName = agent.model.includes(":") ? agent.model.split(":")[1] : agent.model;
           const costResult = this.modelRegistry.estimateCost(
             modelName,
-            info.inputTokens ?? 0,
-            info.outputTokens,
-            info.cachedInputTokens,
-            info.cacheCreationInputTokens
+            agent.inputTokens ?? 0,
+            agent.outputTokens,
+            agent.cachedInputTokens,
+            agent.cacheCreationInputTokens
           );
           agent.cost = costResult?.totalCost;
         } catch {
@@ -13341,6 +13430,27 @@ var StreamProgress = class {
       this.render();
     }
   }
+  /**
+   * Get aggregated metrics from all nested agents for a parent gadget.
+   * Used to show total token counts and cost for subagent gadgets like BrowseWeb.
+   */
+  getAggregatedSubagentMetrics(parentInvocationId) {
+    let inputTokens = 0;
+    let outputTokens = 0;
+    let cachedInputTokens = 0;
+    let cost = 0;
+    let callCount = 0;
+    for (const [, nested] of this.nestedAgents) {
+      if (nested.parentInvocationId === parentInvocationId) {
+        inputTokens += nested.inputTokens ?? 0;
+        outputTokens += nested.outputTokens ?? 0;
+        cachedInputTokens += nested.cachedInputTokens ?? 0;
+        cost += nested.cost ?? 0;
+        callCount++;
+      }
+    }
+    return { inputTokens, outputTokens, cachedInputTokens, cost, callCount };
+  }
   /**
    * Add a nested gadget call (called when nested gadget_call event received).
    */
@@ -13508,20 +13618,22 @@ var StreamProgress = class {
     this.clearRenderedLines();
     const spinner = SPINNER_FRAMES[this.frameIndex++ % SPINNER_FRAMES.length];
     const lines = [];
-    if (this.mode === "streaming") {
-      lines.push(this.formatStreamingLine(spinner));
-    } else {
-      lines.push(this.formatCumulativeLine(spinner));
-    }
+    const activeNestedStreams = [];
     if (this.isTTY) {
       for (const [gadgetId, gadget] of this.inFlightGadgets) {
         const elapsedSeconds = (Date.now() - gadget.startTime) / 1e3;
-        const gadgetLine = `  ${formatGadgetLine({
-          name: gadget.name,
-          parameters: gadget.params,
-          elapsedSeconds,
-          isComplete: false
-        })}`;
+        const termWidth = process.stdout.columns ?? 80;
+        const gadgetIndent = "  ";
+        const line = formatGadgetLine(
+          {
+            name: gadget.name,
+            parameters: gadget.params,
+            elapsedSeconds,
+            isComplete: false
+          },
+          termWidth - gadgetIndent.length
+        );
+        const gadgetLine = line.split("\n").map((l) => gadgetIndent + l).join("\n");
         lines.push(gadgetLine);
         const nestedOps = [];
         for (const [_agentId, nested] of this.nestedAgents) {
@@ -13540,6 +13652,18 @@ var StreamProgress = class {
               completed: nested.completed,
               completedTime: nested.completedTime
             });
+            if (!nested.completed) {
+              activeNestedStreams.push({
+                depth: nested.depth,
+                iteration: nested.iteration,
+                model: nested.model,
+                inputTokens: nested.inputTokens,
+                cachedInputTokens: nested.cachedInputTokens,
+                outputTokens: nested.outputTokens,
+                cost: nested.cost,
+                startTime: nested.startTime
+              });
+            }
           }
         }
         for (const [_nestedId, nestedGadget] of this.nestedGadgets) {
@@ -13557,11 +13681,14 @@ var StreamProgress = class {
         }
         nestedOps.sort((a, b) => a.startTime - b.startTime);
         for (const op of nestedOps) {
-          const indent = "  ".repeat(op.depth + 1);
+          if (op.type === "agent" && !op.completed) {
+            continue;
+          }
+          const indent = "  ".repeat(op.depth + 2);
           const endTime = op.completedTime ?? Date.now();
           const elapsedSeconds2 = (endTime - op.startTime) / 1e3;
           if (op.type === "agent") {
-            const line = formatLLMCallLine({
+            const line2 = formatLLMCallLine({
               iteration: op.iteration ?? 0,
               model: op.model ?? "",
               inputTokens: op.inputTokens,
@@ -13573,21 +13700,48 @@ var StreamProgress = class {
               isStreaming: !op.completed,
               spinner
             });
-            lines.push(`${indent}${line}`);
+            lines.push(`${indent}${line2}`);
           } else {
-            const line = formatGadgetLine({
-              name: op.name ?? "",
-              parameters: op.parameters,
-              elapsedSeconds: elapsedSeconds2,
-              isComplete: op.completed ?? false
-            });
-            lines.push(`${indent}${line}`);
+            const termWidth2 = process.stdout.columns ?? 80;
+            const line2 = formatGadgetLine(
+              {
+                name: op.name ?? "",
+                parameters: op.parameters,
+                elapsedSeconds: elapsedSeconds2,
+                isComplete: op.completed ?? false
+              },
+              termWidth2 - indent.length
+            );
+            const indentedLine = line2.split("\n").map((l) => indent + l).join("\n");
+            lines.push(indentedLine);
           }
         }
       }
     }
-    this.lastRenderLineCount = lines.length;
-    this.target.write("\r" + lines.join("\n"));
+    for (const stream2 of activeNestedStreams) {
+      const indent = "  ".repeat(stream2.depth + 2);
+      const elapsedSeconds = (Date.now() - stream2.startTime) / 1e3;
+      const line = formatLLMCallLine({
+        iteration: stream2.iteration,
+        model: stream2.model,
+        inputTokens: stream2.inputTokens,
+        cachedInputTokens: stream2.cachedInputTokens,
+        outputTokens: stream2.outputTokens,
+        elapsedSeconds,
+        cost: stream2.cost,
+        isStreaming: true,
+        spinner
+      });
+      lines.push(`${indent}${line}`);
+    }
+    if (this.mode === "streaming") {
+      lines.push(this.formatStreamingLine(spinner));
+    } else {
+      lines.push(this.formatCumulativeLine(spinner));
+    }
+    const output = lines.join("\n");
+    this.lastRenderLineCount = (output.match(/\n/g) || []).length + 1;
+    this.target.write("\r" + output);
     this.hasRendered = true;
   }
   /**
@@ -14199,6 +14353,7 @@ async function executeAgent(promptArg, options, env) {
             env.stderr.write(`${summary}
 `);
           }
+          env.stderr.write("\n");
         }
         if (llmSessionDir) {
           const filename = `${formatCallNumber(llmCallCounter)}.response`;
@@ -14305,8 +14460,7 @@ Denied: ${result.reason ?? "by user"}`
   builder.withTrailingMessage(
     (ctx) => [
       `[Iteration ${ctx.iteration + 1}/${ctx.maxIterations}]`,
-      "Think carefully: what gadget invocations can you make in parallel right now?",
-      "Maximize efficiency by batching independent operations in a single response."
+      "Think carefully in two steps: 1. what gadget invocations we should be making next? 2. how do they depend on one another so we can run all of them in the right order? Then respond with all the gadget invocations you are able to do now."
     ].join(" ")
   );
   if (!options.quiet) {
@@ -14319,8 +14473,12 @@ Denied: ${result.reason ?? "by user"}`
           subagentEvent.gadgetInvocationId,
           subagentEvent.depth,
           info.model,
-          info.iteration,
-          info.inputTokens
+          info.iteration + 1,
+          // Make 1-indexed like main agent
+          {
+            inputTokens: info.usage?.inputTokens ?? info.inputTokens,
+            cachedInputTokens: info.usage?.cachedInputTokens
+          }
         );
       } else if (subagentEvent.type === "llm_call_end") {
         const info = subagentEvent.event;
@@ -14397,10 +14555,23 @@ Denied: ${result.reason ?? "by user"}`
           }
         } else {
           const tokenCount = await countGadgetOutputTokens(event.result.result);
-          env.stderr.write(
-            `${formatGadgetSummary2({ ...event.result, tokenCount, media: event.result.storedMedia })}
-`
+          const subagentMetrics = progress.getAggregatedSubagentMetrics(
+            event.result.invocationId
           );
+          const summary = formatGadgetSummary2({
+            ...event.result,
+            tokenCount,
+            media: event.result.storedMedia,
+            subagentMetrics: subagentMetrics.callCount > 0 ? subagentMetrics : void 0
+          });
+          if (event.result.gadgetName === "TellUser") {
+            env.stderr.write(`${summary}
+`);
+          } else {
+            const indentedSummary = summary.split("\n").map((line) => "  " + line).join("\n");
+            env.stderr.write(`${indentedSummary}
+`);
+          }
         }
         if (progress.hasInFlightGadgets()) {
           progress.start();