kimiflare 0.17.0 → 0.18.0

package/dist/index.js

@@ -653,7 +653,8 @@ async function logTurnDebug(ctx) {
     toolTotalReducedBytes: toolTotalReduced,
     toolSavingsPct: toolTotalRaw > 0 ? Math.round((toolTotalRaw - toolTotalReduced) / toolTotalRaw * 100) : 0,
     cacheDiagnostics,
-    compaction: ctx.compaction
+    compaction: ctx.compaction,
+    shadowStrip: ctx.shadowStrip
   });
 }
 var LOG_VERSION;
@@ -665,6 +666,53 @@ var init_cost_debug = __esm({
   }
 });
 
+// src/agent/strip-reasoning.ts
+function stripHistoricalReasoning(messages, opts2 = {}) {
+  const keepLast = opts2.keepLast ?? DEFAULT_KEEP_LAST;
+  const assistantIndices = [];
+  for (let i = 0; i < messages.length; i++) {
+    if (messages[i].role === "assistant") {
+      assistantIndices.push(i);
+    }
+  }
+  const preservedSet = keepLast === 0 ? /* @__PURE__ */ new Set() : new Set(assistantIndices.slice(-keepLast));
+  return messages.map((m, idx) => {
+    if (m.role !== "assistant") return m;
+    if (preservedSet.has(idx)) return m;
+    const next = { ...m };
+    delete next.reasoning_content;
+    if (next.tool_calls && next.tool_calls.length > 0) {
+      if (typeof next.content === "string") {
+        next.content = "";
+      } else if (Array.isArray(next.content)) {
+        next.content = next.content.map(
+          (part) => part.type === "text" ? { ...part, text: "" } : part
+        );
+      }
+      return next;
+    }
+    const textLen = typeof next.content === "string" ? next.content.length : Array.isArray(next.content) ? next.content.filter((p) => p.type === "text").reduce((sum, p) => sum + p.text.length, 0) : 0;
+    if (textLen <= SUBSTANTIVE_TEXT_THRESHOLD) {
+      if (typeof next.content === "string") {
+        next.content = "";
+      } else if (Array.isArray(next.content)) {
+        next.content = next.content.map(
+          (part) => part.type === "text" ? { ...part, text: "" } : part
+        );
+      }
+    }
+    return next;
+  });
+}
+var DEFAULT_KEEP_LAST, SUBSTANTIVE_TEXT_THRESHOLD;
+var init_strip_reasoning = __esm({
+  "src/agent/strip-reasoning.ts"() {
+    "use strict";
+    DEFAULT_KEEP_LAST = 1;
+    SUBSTANTIVE_TEXT_THRESHOLD = 200;
+  }
+});
+
 // src/agent/loop.ts
 async function runAgentTurn(opts2) {
   const max = opts2.maxToolIterations ?? 50;
@@ -679,11 +727,44 @@ async function runAgentTurn(opts2) {
     let content = "";
     let reasoning = "";
     opts2.callbacks.onAssistantStart?.();
+    const stripReasoning = process.env.KIMIFLARE_STRIP_REASONING === "1";
+    const shadowStrip = process.env.KIMIFLARE_SHADOW_STRIP === "1";
+    const keepLastRaw = process.env.KIMIFLARE_REASONING_KEEP_LAST;
+    const keepLast = keepLastRaw ? parseInt(keepLastRaw, 10) : 1;
+    let apiMessages = opts2.messages;
+    let shadowStripMetrics;
+    if (stripReasoning || shadowStrip) {
+      const stripped = stripHistoricalReasoning(opts2.messages, {
+        keepLast: Number.isNaN(keepLast) ? 1 : keepLast
+      });
+      if (shadowStrip) {
+        const originalSections = analyzePrompt(opts2.messages);
+        const strippedSections = analyzePrompt(stripped);
+        const originalApproxTokens = originalSections.reduce(
+          (sum, s) => sum + s.approxTokens,
+          0
+        );
+        const strippedApproxTokens = strippedSections.reduce(
+          (sum, s) => sum + s.approxTokens,
+          0
+        );
+        shadowStripMetrics = {
+          originalApproxTokens,
+          strippedApproxTokens,
+          savingsPct: originalApproxTokens > 0 ? Math.round(
+            (originalApproxTokens - strippedApproxTokens) / originalApproxTokens * 100
+          ) : 0
+        };
+      }
+      if (stripReasoning) {
+        apiMessages = stripped;
+      }
+    }
     const events = runKimi({
       accountId: opts2.accountId,
       apiToken: opts2.apiToken,
       model: opts2.model,
-      messages: opts2.messages,
+      messages: apiMessages,
       tools: toolDefs,
       signal: opts2.signal,
       temperature: opts2.temperature,
@@ -750,7 +831,8 @@ async function runAgentTurn(opts2) {
           messages: opts2.messages,
           previousMessages,
           toolResults,
-          usage: lastUsage
+          usage: lastUsage,
+          shadowStrip: shadowStripMetrics
         });
       }
       return;
@@ -778,7 +860,8 @@ async function runAgentTurn(opts2) {
         messages: opts2.messages,
         previousMessages,
         toolResults,
-        usage: lastUsage
+        usage: lastUsage,
+        shadowStrip: shadowStripMetrics
       });
     }
   }
@@ -800,6 +883,7 @@ var init_loop = __esm({
     init_registry();
     init_messages();
     init_cost_debug();
+    init_strip_reasoning();
   }
 });