@sanity-labs/nuum 0.5.3 → 0.5.4

package/dist/index.js

@@ -35674,6 +35674,20 @@ function estimateSummaryTokens(input) {
 }
 // src/agent/loop.ts
 var log5 = Log.create({ service: "agent-loop" });
+var MODEL_MAX_OUTPUT_TOKENS = {
+  "claude-opus-4-6": 128000,
+  "claude-opus-4-6-20250918": 128000,
+  "claude-sonnet-4-5-20250929": 64000,
+  "claude-sonnet-4-5": 64000,
+  "claude-haiku-4-5-20251001": 64000,
+  "claude-haiku-4-5": 64000,
+  "claude-3-5-sonnet-20241022": 8192,
+  "claude-3-5-haiku-20241022": 8192
+};
+var DEFAULT_MAX_OUTPUT_TOKENS = 16384;
+function getMaxOutputTokens(modelId) {
+  return MODEL_MAX_OUTPUT_TOKENS[modelId] ?? DEFAULT_MAX_OUTPUT_TOKENS;
+}
 function addCacheMarkers(messages) {
   if (messages.length === 0)
     return messages;
@@ -35702,7 +35716,7 @@ async function runAgentLoop(options) {
     systemPrompt,
     initialMessages,
     tools,
-    maxTokens = 4096,
+    maxTokens: maxTokensOverride,
     temperature,
     maxTurns,
     abortSignal,
@@ -35713,6 +35727,12 @@ async function runAgentLoop(options) {
     onBeforeTurn,
     onThinking
   } = options;
+  const maxTokens = maxTokensOverride ?? getMaxOutputTokens(model.modelId);
+  log5.info("agent loop starting", {
+    model: model.modelId,
+    maxTokens,
+    maxTurns
+  });
   if (abortSignal?.aborted) {
     throw new AgentLoopCancelledError;
   }
@@ -35768,6 +35788,13 @@ async function runAgentLoop(options) {
         cacheHitRate: total > 0 ? `${Math.round(cacheRead / total * 100)}%` : "0%"
       });
     }
+    if (response.finishReason === "length") {
+      log5.warn("output truncated - model hit maxTokens limit", {
+        maxTokens,
+        outputTokens: response.usage.completionTokens,
+        hasToolCalls: (response.toolCalls?.length ?? 0) > 0
+      });
+    }
     if (response.text) {
       finalText = response.text;
       await onText?.(response.text);
@@ -35812,6 +35839,15 @@ async function runAgentLoop(options) {
         content: toolResultParts
       };
       messages.push(toolMsg);
+      if (response.finishReason === "length") {
+        const hadInvalidCalls = toolCallInfos.some((tc) => tc.toolName === "__invalid_tool_call__");
+        if (hadInvalidCalls) {
+          messages.push({
+            role: "user",
+            content: "[SYSTEM: Your previous output was truncated because it exceeded the output token limit. " + "Your tool call was incomplete \u2014 parameters were cut off mid-generation. " + "To fix this: break large content into smaller chunks, or use bash with echo/cat to write files incrementally. " + "Do NOT retry the same large tool call \u2014 it will truncate again.]"
+          });
+        }
+      }
       if (isDone(toolCallInfos)) {
         stopReason = "done";
         break;
@@ -36138,7 +36174,6 @@ async function runCompaction(storage, config) {
         systemPrompt: ctx.systemPrompt,
         initialMessages,
         tools,
-        maxTokens: 4096,
         temperature: 0,
         maxTurns: 5,
         isDone: stopOnTool("finish_distillation"),
@@ -36244,7 +36279,7 @@ async function runSubAgent(storage, config) {
     extractResult,
     tier = "workhorse",
     maxTurns = 20,
-    maxTokens = 4096,
+    maxTokens,
     temperature = 0,
     onToolResult
   } = config;
@@ -36431,8 +36466,7 @@ async function runReflection(storage, question) {
     finishToolName: "finish_reflection",
     extractResult: getAnswer,
     tier: "workhorse",
-    maxTurns: 20,
-    maxTokens: 4096
+    maxTurns: 20
   });
   const answer = result.result ?? "Unable to find relevant information.";
   activity.reflection.complete(`${result.turnsUsed} turns, ${answer.length} chars`);
@@ -36836,7 +36870,6 @@ async function runResearch(storage, topic) {
     },
     tier: "workhorse",
     maxTurns: MAX_RESEARCH_TURNS,
-    maxTokens: 8192,
     onToolResult: (toolCallId) => {
       const toolResult = getLastResult(toolCallId);
       if (!toolResult)
@@ -45371,8 +45404,8 @@ var Mcp;
 })(Mcp ||= {});
 
 // src/version.ts
-var VERSION = "0.5.3";
-var GIT_HASH = "a2a5ddc";
+var VERSION = "0.5.4";
+var GIT_HASH = "e1f89a6";
 var VERSION_STRING = `nuum v${VERSION} (${GIT_HASH})`;
 
 // src/tool/mcp-status.ts
@@ -46890,7 +46923,6 @@ async function runConsolidation(storage, messages) {
     },
     tier: "workhorse",
     maxTurns: MAX_CONSOLIDATION_TURNS,
-    maxTokens: 2048,
     onToolResult: (toolCallId) => {
       const toolResult = getLastResult(toolCallId);
       if (!toolResult)
@@ -47432,7 +47464,6 @@ async function runAgent(prompt, options) {
     systemPrompt: ctx.systemPrompt,
     initialMessages,
     tools,
-    maxTokens: 8192,
     maxTurns: MAX_TURNS,
     abortSignal,
     onText: async (text3) => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sanity-labs/nuum",
-  "version": "0.5.3",
+  "version": "0.5.4",
   "description": "AI coding agent with continuous memory - infinite context across sessions",
   "type": "module",
   "bin": {