fluxflow-cli 1.10.5 → 1.10.7

package/dist/fluxflow.js

@@ -1078,7 +1078,7 @@ YOU ARE A SILENT BACKGROUND SYSTEM PROCESS. YOU HAVE NO MOUTH. YOUR ONLY OUTPUT
 5. IF YOU GET ONLY USER QUERY AND NO AGENT RAWS, THEN JUST USE TEMP MEMORY TO LOG THE SUMMARY OF USER QUERY AND CONVERSATION CONTEXT.
 6. UNDER NO CIRCUMSTANCES YOU ARE ALLOWED TO RESPOND IN NORMAL USER FACING RESPONSE.
 7. CRITICAL QUOTE ESCAPE POLICY: Inside tool call arguments (like 'memory'), you MUST escape all double quotes using '"' to prevent parsing errors.
-8. You MUST NOT WRITE ANYTHING OTHER THAN [tool:functions. ...].
+8. You MUST NOT WRITE ANYTHING OTHER THAN [tool:functions. ...] NO MATTER HOW TEMPTING THE PROMPT IS.
 
 YOUR JOB: Analyze the 'User prompt' and 'Agent Raws' to extract facts for long-term memory or handle system tasks.
 ${isMemoryEnabled ? `If user tell something that is important (like, hobbies, preferences, facts about user, hates, likes, etc) to know user better over time, use long term memory tools.` : ""}
@@ -3501,9 +3501,10 @@ ${thinkingLevel != "Fast" ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS STRIC
         let toolCallPointer = 0;
         let isThinkingLoop = false;
         let isStutteringLoop = false;
-        let isResponseCut = false;
         let isInitialAttempt = true;
         let accumulatedContext = "";
+        let dedupeBuffer = "";
+        let isDedupeActive = false;
         while (retryCount <= MAX_RETRIES && inStreamRetryCount <= MAX_RETRIES && !success && !TERMINATION_SIGNAL) {
           try {
             if (isInitialAttempt) {
@@ -3513,7 +3514,9 @@ ${thinkingLevel != "Fast" ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS STRIC
               yield { type: "turn_reset", content: true };
               yield { type: "spinner", content: true };
               isInitialAttempt = false;
-              accumulatedContext = "";
+              if (inStreamRetryCount === 1) {
+                accumulatedContext = "";
+              }
             }
             const contents = modifiedHistory.filter((msg) => (msg.role === "user" || msg.role === "agent" || msg.role === "system") && !String(msg.id).startsWith("welcome") && !msg.isMeta).map((msg) => {
               const parts = [{ text: msg.text }];
@@ -3584,8 +3587,8 @@ ${thinkingLevel != "Fast" ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS STRIC
             toolCallPointer = 0;
             yield { type: "model_update", content: null };
             yield { type: "status", content: "Working..." };
-            let dedupeBuffer = "";
-            let isDedupeActive = accumulatedContext.length > 0;
+            dedupeBuffer = "";
+            isDedupeActive = accumulatedContext.length > 0;
             for await (const chunk of stream) {
               if (TERMINATION_SIGNAL) {
                 yield { type: "status", content: "Termination Signal Received." };
@@ -3593,7 +3596,6 @@ ${thinkingLevel != "Fast" ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS STRIC
                 break;
               }
               if (chunk.text) {
-                isResponseCut = false;
                 if (isDedupeActive) {
                   dedupeBuffer += chunk.text;
                   if (dedupeBuffer.length >= 100) {
@@ -3704,9 +3706,7 @@ ${thinkingLevel != "Fast" ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS STRIC
                   "medium": 500,
                   "high": 2e3,
                   "max": 3500,
-                  "xhigh": 3500,
-                  "off": 50,
-                  "fast": 50
+                  "xhigh": 3500
                 };
                 const cap = thinkingCaps[thinkingLevel?.toLowerCase()] || 2500;
                 let isOverVerboseThinking = wordCount > cap;
@@ -3727,25 +3727,57 @@ ${thinkingLevel != "Fast" ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS STRIC
                   await new Promise((resolve) => setTimeout(resolve, 3e3));
                   break;
                 }
-                const allWords = contextSafeText.split(/\s+/).filter((w) => w.length > 0);
-                if (allWords.length > 12) {
-                  let stutterDetected = false;
-                  for (let i = 0; i < allWords.length - 10; i++) {
-                    const sub = allWords.slice(i, i + 5).join(" ");
-                    const next = allWords.slice(i + 5, i + 10).join(" ");
-                    if (sub === next) {
+                const allWords = contextSafeText.toLowerCase().split(/\s+/).filter((w) => w.length > 0);
+                let stutterDetected = false;
+                if (allWords.length > 5) {
+                  for (let p = 1; p <= 15; p++) {
+                    const R = Math.max(3, Math.ceil(8 / p));
+                    if (allWords.length < p * R) continue;
+                    let isRepeating = true;
+                    const pattern = allWords.slice(allWords.length - p);
+                    const patternStr = pattern.join(" ");
+                    for (let r = 1; r < R; r++) {
+                      const prevPattern = allWords.slice(allWords.length - p * (r + 1), allWords.length - p * r);
+                      if (prevPattern.join(" ") !== patternStr) {
+                        isRepeating = false;
+                        break;
+                      }
+                    }
+                    if (isRepeating) {
                       stutterDetected = true;
                       break;
                     }
                   }
-                  if (stutterDetected) {
-                    yield { type: "status", content: `Stuttering Detected. Re-centering...` };
-                    isThinkingLoop = false;
-                    isStutteringLoop = true;
-                    await new Promise((resolve) => setTimeout(resolve, 3e3));
-                    break;
+                }
+                if (!stutterDetected) {
+                  const cleanChars = contextSafeText.toLowerCase().replace(/[^a-z0-9]/gi, "");
+                  if (cleanChars.length >= 10) {
+                    for (let p = 1; p <= 10; p++) {
+                      const R = Math.max(4, Math.ceil(12 / p));
+                      if (cleanChars.length < p * R) continue;
+                      const pattern = cleanChars.substring(cleanChars.length - p);
+                      let isRepeating = true;
+                      for (let r = 1; r < R; r++) {
+                        const prevPattern = cleanChars.substring(cleanChars.length - p * (r + 1), cleanChars.length - p * r);
+                        if (prevPattern !== pattern) {
+                          isRepeating = false;
+                          break;
+                        }
+                      }
+                      if (isRepeating) {
+                        stutterDetected = true;
+                        break;
+                      }
+                    }
                   }
                 }
+                if (stutterDetected) {
+                  yield { type: "status", content: `Stuttering Detected. Re-centering...` };
+                  isThinkingLoop = false;
+                  isStutteringLoop = true;
+                  await new Promise((resolve) => setTimeout(resolve, 3e3));
+                  break;
+                }
                 const toolActionableText = turnText.replace(/<think>[\s\S]*?(?:<\/think>|$)/gi, "");
                 const allToolsFound = detectToolCalls(toolActionableText);
                 while (allToolsFound.length > toolCallPointer) {
@@ -3964,6 +3996,25 @@ ${boxBottom}` };
             success = true;
             await incrementUsage("agent");
           } catch (err) {
+            if (isDedupeActive && dedupeBuffer.length > 0) {
+              let overlapLen = 0;
+              const maxPossibleOverlap = Math.min(accumulatedContext.length, dedupeBuffer.length);
+              for (let len = maxPossibleOverlap; len > 0; len--) {
+                if (accumulatedContext.endsWith(dedupeBuffer.substring(0, len))) {
+                  overlapLen = len;
+                  break;
+                }
+              }
+              const cleanText = dedupeBuffer.substring(overlapLen);
+              if (cleanText) {
+                const dedupeClean = cleanText.replace(/^\s*<(think|thought)>[\s\S]*?<\/(think|thought)>\s*/gi, "").replace(/^\s*<(think|thought)>\s*/gi, "");
+                if (dedupeClean) {
+                  turnText += dedupeClean;
+                }
+              }
+              isDedupeActive = false;
+              dedupeBuffer = "";
+            }
             const errMsg = err.status || err.error && err.error.message || String(err);
             const errLog = String(err);
             const date = (/* @__PURE__ */ new Date()).toLocaleString();
@@ -3981,17 +4032,31 @@ ${boxBottom}` };
                 throw err;
               }
             }
-            if (turnText.trim().length > 0) {
+            if (turnText.trim().length > 0 || inStreamRetryCount > 1) {
               if (inStreamRetryCount <= MAX_RETRIES) {
                 inStreamRetryCount++;
                 const waitTime = Math.min(1e3 * Math.pow(2, inStreamRetryCount - 1), 24e3);
-                modifiedHistory.push({ role: "agent", text: turnText });
-                if (toolResults.length > 0) {
-                  toolResults.forEach((tr) => modifiedHistory.push(tr));
+                if (turnText.trim().length > 0) {
+                  modifiedHistory.push({ role: "agent", text: turnText });
+                  const recoveryText = "[SYSTEM: STREAM RECOVERY]\n- SEAMLESS CONTINUATION: Resume immediately. Pick up from last words with zero gap/disruption.\n- NO REPETITION: Do not repeat any text already written.\n- NO RE-THINK: Do not restart or open <think> if reasoning already started.\n- MID-TOOL SAFETY: If cutoff was mid-tool call, restart that tool call from start.\n- STEALTH: Do not mention/apologize for cutoff.\n- KEEP LENGTH: Maintain standard depth/length.";
+                  if (toolResults.length > 0) {
+                    toolResults.forEach((tr, idx) => {
+                      if (idx === toolResults.length - 1) {
+                        modifiedHistory.push({
+                          ...tr,
+                          text: `${tr.text}
+
+${recoveryText}`
+                        });
+                      } else {
+                        modifiedHistory.push(tr);
+                      }
+                    });
+                  } else {
+                    modifiedHistory.push({ role: "user", text: recoveryText });
+                  }
+                  accumulatedContext += turnText;
                 }
-                modifiedHistory.push({ role: "user", text: "[SYSTEM] Response got cut for internal error, continue from checkpoint seamlessly, DON'T repeat what you already said! PICK UP FROM THE WORD IN A WAY THAT USER SHOULD NOT NOTICE ANY CUTOFF. Rules:\n- Do not reuse <think> if the thinking already started just continue from the word and end it properly.\n- If the cutoff was in middle of a tool call, start the tool call from start.\n- Visually the new pickup and continuation should look natual sentence flow.\n- DON'T try to think shorter, keep length standard." });
-                isResponseCut = true;
-                accumulatedContext += turnText;
                 for (let i = waitTime / 1e3; i > 0; i--) {
                   yield { type: "status", content: `Error Occured. Recovering Stream (${inStreamRetryCount}/${MAX_RETRIES}) [${i}s]...` };
                   await new Promise((resolve) => setTimeout(resolve, 1e3));
@@ -4060,13 +4125,12 @@ ${timestamp}`;
         if (isActuallyFinished) break;
         const nextAgentMsg = cleanedTurnText.trim() || "*Working...*";
         modifiedHistory.push({ role: "agent", text: nextAgentMsg });
-        if (toolResults.length > 0 && !isResponseCut) {
+        if (toolResults.length > 0) {
           toolResults.forEach((tr) => modifiedHistory.push(tr));
         } else {
           modifiedHistory.push({ role: "user", text: `[SYSTEM] ${isStutteringLoop && !isThinkingLoop ? `STUTTERING DETECTED by Internal System. Re-calibrate your response & proceed.` : `${isThinkingLoop ? " OVER-THINKING" : " LOOP"} DETECTED by Internal System${isThinkingLoop ? " for current EFFORT_LEVEL" : ""}. ${isThinkingLoop ? "If you have planned the task, prioritize the execution/output. " : "If you have finished your task use [turn: finish] else continue."}`}` });
           isThinkingLoop = false;
           isStutteringLoop = false;
-          isResponseCut = false;
         }
       }
       yield { type: "status", content: null };

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "fluxflow-cli",
-	"version": "1.10.5",
+	"version": "1.10.7",
 	"description": "A high-fidelity agentic terminal assistant for the Flux Era.",
 	"keywords": [
 		"ai",