nolo-cli 0.90.0-alpha.1 → 0.91.1

package/{chunk-VHJLQHPQ.js → chunk-F64B2W3R.js}

@@ -6,13 +6,13 @@ import {
 } from "./chunk-DWLEOOGR.js";
 import {
   addContentAction
-} from "./chunk-4DRWWIOL.js";
+} from "./chunk-B7ANJQWZ.js";
 import {
   upload
-} from "./chunk-UA5S3LQQ.js";
+} from "./chunk-QALRKMHF.js";
 import {
   getRuntimeServerContext
-} from "./chunk-UKWGR5SS.js";
+} from "./chunk-AUXHZ2M5.js";
 import {
   API_ENDPOINTS
 } from "./chunk-LCBAMY2M.js";
@@ -367,6 +367,7 @@ var truncateToolOutputForContext = (text, contextWindow) => {
 // packages/ai/context/planCompression.ts
 var MIN_COMPRESS_COUNT = 5;
 var ACTIVE_SUMMARY_TAIL_KEEP_COUNT = 2;
+var COMPACTION_RETENTION_CUSHION_RATIO = 0.18;
 var MIN_TRIGGER_PENDING_COUNT = MIN_COMPRESS_COUNT + ACTIVE_SUMMARY_TAIL_KEEP_COUNT;
 var getMessageTokenCount = (msg) => {
   const content = serializeMessageContent(msg.content) || "";
@@ -409,13 +410,14 @@ var normalizeContextUsageRatio = (value) => {
   const ratio = value > 1 ? value / 100 : value;
   return ratio >= 0 && ratio <= 1 ? ratio : void 0;
 };
-var emptyPlan = (startIndex) => ({
+var emptyPlan = (startIndex, diagnostics) => ({
   shouldCompress: false,
   compressCount: 0,
   msgsToCompress: [],
   msgsToKeep: [],
   newSummarizedBeforeId: void 0,
-  startIndex
+  startIndex,
+  ...diagnostics ? { diagnostics } : {}
 });
 function findPendingMessages(allMsgs, summarizedBeforeId) {
   let startIndex = 0;
@@ -464,15 +466,42 @@ function calculateCompressCount(pendingMsgs, rawMessageBudget, opts) {
   }
   return pendingMsgs.length - keepCount;
 }
+function resolveEffectiveCompactionBudgets(args) {
+  const {
+    contextWindow,
+    triggerRatio,
+    historyBudget,
+    rawMessageBudget,
+    summaryTokens,
+    cushionRatio = COMPACTION_RETENTION_CUSHION_RATIO
+  } = args;
+  const maxRetainedRatio = Math.max(
+    0.2,
+    Number((triggerRatio - cushionRatio).toFixed(4))
+  );
+  const maxRetainedHistoryBudget = Math.floor(contextWindow * maxRetainedRatio);
+  const effectiveHistoryBudget = Math.min(historyBudget, maxRetainedHistoryBudget);
+  const effectiveRawMessageBudget = Math.min(
+    rawMessageBudget,
+    Math.max(0, effectiveHistoryBudget - summaryTokens)
+  );
+  return {
+    effectiveHistoryBudget,
+    effectiveRawMessageBudget,
+    maxRetainedRatio
+  };
+}
 function guardToolChainBoundary(pendingMsgs, compressCount) {
   let count = compressCount;
-  while (count > 0 && count < pendingMsgs.length && pendingMsgs[count].role === "tool") {
-    count--;
+  while (count < pendingMsgs.length && pendingMsgs[count].role === "tool") {
+    count++;
   }
-  if (count > 0) {
-    const lastCompressed = pendingMsgs[count - 1];
-    if (hasOpenEndedToolCall(lastCompressed)) {
+  while (count > 0 && hasOpenEndedToolCall(pendingMsgs[count - 1])) {
+    if (count < pendingMsgs.length && pendingMsgs[count].role === "tool") {
+      count++;
+    } else {
       count--;
+      break;
     }
   }
   return count;
@@ -489,7 +518,20 @@ function planCompression(input) {
     coldResume
   } = input;
   const { pendingMsgs, startIndex } = findPendingMessages(allMsgs, summarizedBeforeId);
-  if (pendingMsgs.length === 0) return emptyPlan(startIndex);
+  if (pendingMsgs.length === 0) {
+    return emptyPlan(startIndex, {
+      pendingMsgCount: 0,
+      totalUsed: estimateTokenCount(summary || ""),
+      historyBudget: 0,
+      triggerRatio: resolveCompressionTriggerRatio(contextWindow),
+      realUsageRatio: normalizeContextUsageRatio(realContextUsagePercent),
+      triggeredByRealUsage: false,
+      triggeredByEstimate: false,
+      triggeredByColdResume: false,
+      shouldRunActiveSummary: false,
+      emptyReason: "no-pending"
+    });
+  }
   const summaryTokens = estimateTokenCount(summary || "");
   const pendingTokens = pendingMsgs.reduce((sum, msg) => sum + getMessageTokenCount(msg), 0);
   const totalUsed = summaryTokens + pendingTokens;
@@ -517,14 +559,47 @@ function planCompression(input) {
     coldResume,
     lastMsg: pendingMsgs[pendingMsgs.length - 1]
   });
-  if (!trigger) return emptyPlan(startIndex);
-  let compressCount = calculateCompressCount(pendingMsgs, rawMessageBudget, {
-    keepTailCount: shouldRunActiveSummary || triggeredByRealUsage || triggeredByColdResume,
+  const diagnostics = {
+    pendingMsgCount: pendingMsgs.length,
     totalUsed,
-    historyBudget
+    historyBudget,
+    triggerRatio,
+    realUsageRatio: normalizeContextUsageRatio(realContextUsagePercent),
+    triggeredByRealUsage,
+    triggeredByEstimate: normalizeContextUsageRatio(realContextUsagePercent) === void 0 && totalUsed >= historyBudget,
+    triggeredByColdResume,
+    shouldRunActiveSummary
+  };
+  if (!trigger) {
+    return emptyPlan(startIndex, {
+      ...diagnostics,
+      emptyReason: "not-triggered"
+    });
+  }
+  const { effectiveHistoryBudget, effectiveRawMessageBudget } = resolveEffectiveCompactionBudgets({
+    contextWindow,
+    triggerRatio,
+    historyBudget,
+    rawMessageBudget,
+    summaryTokens
   });
+  let compressCount = calculateCompressCount(
+    pendingMsgs,
+    effectiveRawMessageBudget,
+    {
+      keepTailCount: shouldRunActiveSummary || triggeredByRealUsage || triggeredByColdResume,
+      totalUsed,
+      historyBudget: effectiveHistoryBudget
+    }
+  );
   compressCount = guardToolChainBoundary(pendingMsgs, compressCount);
-  if (compressCount < MIN_COMPRESS_COUNT) return emptyPlan(startIndex);
+  if (compressCount < MIN_COMPRESS_COUNT) {
+    return emptyPlan(startIndex, {
+      ...diagnostics,
+      compressCount,
+      emptyReason: "below-min-compress-count"
+    });
+  }
   const msgsToCompress = pendingMsgs.slice(0, compressCount);
   const msgsToKeep = pendingMsgs.slice(compressCount);
   const newSummarizedBeforeId = msgsToCompress[msgsToCompress.length - 1].id;
@@ -534,7 +609,8 @@ function planCompression(input) {
     msgsToCompress,
     msgsToKeep,
     newSummarizedBeforeId,
-    startIndex
+    startIndex,
+    diagnostics: { ...diagnostics, compressCount }
   };
 }
 
@@ -691,14 +767,6 @@ function formatCompactionMetricsLog(metrics) {
   return `[Compaction] reason=${metrics.reason} compressed=${metrics.compressedCount}->${metrics.retainedCount} msgs, tokens=${totalBefore}->${totalAfter} (ratio=${ratio})${llmInfo}`;
 }
 
-// packages/core/clipCompactText.ts
-function clipCompactText(value, max, ellipsis = "...") {
-  const compact = compactWhitespace(value);
-  if (compact.length <= max) return compact;
-  if (max <= ellipsis.length) return compact.slice(0, max);
-  return `${compact.slice(0, max - ellipsis.length)}${ellipsis}`;
-}
-
 // packages/ai/token/providerCallTiming.ts
 var defaultNow = () => {
   const perf = globalThis.performance;
@@ -769,7 +837,23 @@ var isMeaningfulResponsesOutput = (event) => {
   return false;
 };
 
+// packages/core/clipCompactText.ts
+function clipCompactText(value, max, ellipsis = "...") {
+  const compact = compactWhitespace(value);
+  if (compact.length <= max) return compact;
+  if (max <= ellipsis.length) return compact.slice(0, max);
+  return `${compact.slice(0, max - ellipsis.length)}${ellipsis}`;
+}
+
 export {
+  normalizeTimingMs,
+  createProviderCallTimingTracker,
+  observeMeaningfulProviderResponseAt,
+  observeMeaningfulProviderResponse,
+  finalizeProviderCallTiming,
+  withProviderCallTimingFields,
+  isMeaningfulChatCompletionOutput,
+  isMeaningfulResponsesOutput,
   clipCompactText,
   planContextUsage,
   buildDatabaseFileContentUrl,
@@ -784,13 +868,5 @@ export {
   formatFileOperationsFromMessages,
   formatMessagesForSummaryWithTruncation,
   buildCompactionMetricsFromPlan,
-  formatCompactionMetricsLog,
-  normalizeTimingMs,
-  createProviderCallTimingTracker,
-  observeMeaningfulProviderResponseAt,
-  observeMeaningfulProviderResponse,
-  finalizeProviderCallTiming,
-  withProviderCallTimingFields,
-  isMeaningfulChatCompletionOutput,
-  isMeaningfulResponsesOutput
+  formatCompactionMetricsLog
 };