fluxflow-cli 1.9.14 → 1.9.16

package/ARCHITECTURE.md

@@ -27,7 +27,7 @@ The execution flow of a single user prompt follows this loop:
    - If `[turn: finish]` is detected and no further tools were called, the main loop terminates, passing the final synthesized context to the background Janitor process.
 6. **Loop Limits & Resilience**: To prevent infinite loops or excessive API usage, **Flux mode** is capped at 50 iterations per user prompt, while **Flow mode** is capped at 5.
    - **Multi-Stage Failover**: The loop features a sophisticated 8-attempt retry engine with random backoff (800ms - 2s).
-   - **Critical Fallback Pivot**: If the primary model fails 5 consecutive times, the agent surgically pivots to a lighter, high-concurrency fallback model (`gemini-3.1-flash-lite-preview`) for the final 3 attempts to ensure session navigation through API congestion.
+   - **Critical Fallback Pivot**: If the primary model fails 5 consecutive times, the agent surgically pivots to a lighter, high-concurrency fallback model (`gemini-3.1-flash-lite`) for the final 3 attempts to ensure session navigation through API congestion.
 
 ## Multimodal Pipeline
 

package/dist/fluxflow.js

@@ -475,7 +475,7 @@ var init_ChatLayout = __esm({
               { cmd: "gemma-4-31b-it", desc: "Standard Default (Free, Recommended)" },
               { cmd: "gemini-3.1-pro-preview", desc: "Most Capable (Paid)" },
               { cmd: "gemini-3-flash-preview", desc: "Fast & Lightweight (Paid, Free limited quota)" },
-              { cmd: "gemini-3.1-flash-lite-preview", desc: "Ultra Fast (Paid, Free limited quota)" }
+              { cmd: "gemini-3.1-flash-lite", desc: "Ultra Fast (Paid, Free limited quota)" }
             ]
           },
           { cmd: "/settings", desc: "Configure system prefs" },
@@ -875,17 +875,16 @@ Have access to internal tools. To call a tool, MUST use the following exact synt
 [tool:functions.tool_name(arguments)]
 
 - COMMUNICATION TOOLS -
-1. Ask User: [tool:functions.ask(question="...", optionA="...", ...)]. Ambiguity Resolution. Mandatory Triggers: Path Divergence (present options), Security (request permission for sensitive files), Risk Mitigation (confirm destructive actions). ask >> finish
-NOTE: Suggest best options; don't ask for preferences. System handles the rest
+1. Ask User: [tool:functions.ask(question="...", optionA="...", ...)]. Ambiguity Resolution. Mandatory Triggers: Path Divergence, Security, Risk Mitigation. ask >> finish
+Suggest best options; don't ask for preferences. System handles the rest
 
 - WEB TOOLS -
-1. Web Search: [tool:functions.web_search(query="...", limit=number)]. Find info (limit 3-10, default 10). Proactive use for unknown topics${mode === "Flux" ? " or documentation" : ""}
+1. Web Search: [tool:functions.web_search(query="...", limit=number)]. Find info (limit 3-10). Proactive use for unknown topics${mode === "Flux" ? " or docs" : ""}
 2. Web Scrape: [tool:functions.web_scrape(url="<url>")]. Provides detail from a URL
 
-${mode === "Flux" ? `
-- DEV TOOLS (path will always be relative to CWD) -
+${mode === "Flux" ? `- DEV TOOLS (path = relative to CWD) -
 1. View File: [tool:functions.view_file(path="...", start_line=N, end_line=N)]. Reads contents. Supports images/docs. User gives image/doc: VIEW FIRST
-2. Read Folder: [tool:functions.read_folder(path="...")]. Detailed stats of a directory
+2. Read Folder: [tool:functions.read_folder(path="...")]. Detailed DIR stats
 3. Write File: [tool:functions.write_file(path="...", content="content to write")]. Creates/Overwrites. File Exist? -> update_file > write_file
 4. Update File: [tool:functions.update_file(path="...", content_to_replace="old conten as you see it", content_to_add="new content to be added")]. Surgical patching. Unsure content_to_replace? -> use view_file >> guessing.
 5. Write PDF: [tool:functions.write_pdf(path="...", content="html", orientation="...")]. A4 PDF. Use CSS for layout (100vh/vw). Handle page breaks pro-actively; no manual footers
@@ -893,7 +892,7 @@ ${mode === "Flux" ? `
 7. PPTX: [tool:functions.write_pptx(path="...", content="html")]. 4:3 PPTX. '---' for slides
 Tags: a,b,br,del,font,h1-h6,i,ol,ul,li,p,pre,s,sub,sup,u
 CSS: background-color,color,font-family,font-size(pt),font-style,font-weight,margin,text-align,text-shadow
-8. Execution: [tool:functions.exec_command(command="command")]. Runs a shell command. Use ask before running destructive/irreversible ops
+8. Execution: [tool:functions.exec_command(command="command")]. Runs a shell command. Destructive/Irreversible ops -> ask user
 9. Search: [tool:functions.search_keyword(keyword="...")]. Global search. Finds definitions/logic without reading every file
 
 - VERIFY SUCCESS CONTENTS. Fix errors. No hallucinations.
@@ -904,10 +903,9 @@ CSS: background-color,color,font-family,font-size(pt),font-style,font-weight,mar
 - File structure: Use real newlines for code formatting`.trim() : `
 - DEV TOOLS ARE NOT AVAILABLE IN FLOW MODE. If you need to access files, tell the user to switch to FLUX`.trim()}
 
-- Results: Passed as [TOOL_RESULT] (SYSTEM), USER_PROMPT (USER)
+- Results: Passed as [TOOL_RESULT] SYSTEM, [USER] USER
 - Tool calls: End with [turn: continue]. Only use [turn: finish] after verifying goals
-- Multi-call: Stack 1-by-1. Upto 4
--- END TOOL DEFINITIONS --`.trim();
+- Multi-call: Stack 1-by-1. Upto 3`.trim();
   }
 });
 
@@ -918,9 +916,7 @@ var init_janitor_tools = __esm({
     JANITOR_TOOLS_PROTOCOL = (isMemoryEnabled = true, needTitle = true) => `
 ${needTitle ? `-- CHAT MANAGEMENT TOOLS --
 1. YOU MUST UPDATE CHAT TITLE (URGENT PRIORITY):
-[tool:functions.chat(title='<short creative title of FULL conversation in 3-5 words>')]. Consider full chat context to generate title NOT just latest message.
--- END CHAT MANAGEMENT TOOLS --
-`.trimEnd() : ""}
+[tool:functions.chat(title='<short creative title of FULL conversation in 3-5 words>')]. Consider full chat context to generate title NOT just latest message.`.trimEnd() : ""}
 -- MEMORY TOOLS (YOU SHOULD NOT OUTPUT ANYTHING OTHER THAN THESE SPECIFIC TOOLS) --
 Your tool syntax is: '[tool:functions.function_name(args...)]'
 You have access to the following memory functions to persist important information:
@@ -935,8 +931,7 @@ ${isMemoryEnabled ? `2. User-specific long-term memory (USE BASED ON CONVERSATIO
 
 Usage Rules:
 - Frequency for 'user' action: Only when explicit context from chat is found or explicitly requested by the user.
-- IF YOU WANT TO SAVE SOMETHING, BUT SIMILAR MEMORY ALREADY EXISTS, USE THE UPDATE METHOD NOT THE ADD METHOD` : ""}
--- END MEMORY TOOLS --`.trim();
+- IF YOU WANT TO SAVE SOMETHING, BUT SIMILAR MEMORY ALREADY EXISTS, USE THE UPDATE METHOD NOT THE ADD METHOD` : ""}`.trim();
   }
 });
 
@@ -975,13 +970,13 @@ var init_prompts = __esm({
       if (!isMemoryEnabled) return "";
       const tempMemoriesStr = tempMemories?.length > 0 && !isContext32k ? `-- RECENT CONTEXT FROM OTHER CHATS (PRIORITY: LOW, MAIN FOCUS: Chat Context > Recent) --
 ${tempMemories}
--- END RECENT CONTEXT --` : "";
+` : "";
       const userMemoriesStr = userMemories?.length > 0 ? `--- SAVED MEMORIES (PRIORITY: MEDIUM, TUNES USER PREFERENCES) ---
 ${userMemories}
--- END SAVED MEMORIES --` : "";
+` : "";
       const parts = [userMemoriesStr, tempMemoriesStr].filter((p) => p.length > 0);
-      return parts.length > 0 ? `[SYSTEM CONTEXT
-]${parts.join("\n\n")}
+      return parts.length > 0 ? `[SYSTEM CONTEXT]
+${parts.join("\n\n")}
 ` : "";
     };
     getSystemInstruction = (profile, thinkingLevel, mode, systemSettings, isMemoryEnabled = true, maxLoops, currentLoop) => {
@@ -1011,7 +1006,7 @@ ${userMemories}
         }
       })();
       const projectContextFiles = [
-        { name: "Fluxflow.md", desc: "HIGHEST PRIORITY. Overrides other files" },
+        { name: "Fluxflow.md", desc: "HIGH PRIORITY. Overrides other files" },
         { name: "README.md", desc: "Goals" },
         { name: "Agent.md", desc: "Standards" },
         { name: "Skills.md", desc: "Workflows" },
@@ -1022,10 +1017,9 @@ ${userMemories}
       const projectContextBlock = mode === "Flux" && foundFiles.length > 0 ? `
 -- PROJECT CONTEXT (Source of Truth) --
 ${foundFiles.map((f) => `- ${f.name}: ${f.desc}`).join("\n")}
-Check these first; they override general training data for project consistency. Safety rules still apply
--- END PROJECT CONTEXT --` : "";
+Check these first; they override general training data for project consistency. Safety rules still apply` : "";
       return `${nameStr}${nicknameStr}${userInstrStr}
-=== SYSTEM INSTRUCTION (STRICT PRIORITY, OVERRIDES EVERYTHING) ===
+=== SYSTEM PROMPT (HIGHEST PRIORITY, OVERRIDES EVERYTHING) ===
 Identity: Flux Flow (by Kushal Roy Chowdhury). Sassy, Friendly CLI Agent. No flirting
 Mode: ${mode} (THINKING MODE). ${mode === "Flux" ? "Goal-oriented. Plan & use tools" : "Conversation & UX focus. Web/Comm tools only"}
 Context: CWD: ${cwdStr}.${isSystemDir ? " [PROTECTED: ASK BEFORE MODIFYING]" : ""} OS: ${osDetected}.${osDetected === "Windows" ? " (Backslashes only. Prefer PS via CMD)" : ""}
@@ -1037,38 +1031,33 @@ ${thinkingConfig}
 - Always use <think> ... </think> before answering or using any tool
 - Never skip thinking, even for simple tasks, code, or greetings
 - Never jump to responses directly, regardless of task complexity or speed
--- END THINKING INSTRUCTIONS --
 
 ${TOOL_PROTOCOL(mode)}
 ${projectContextBlock}
 
 -- MEMORY INSTRUCTIONS --
-- Memory: ${isMemoryEnabled ? "Use recent context/logs to personalize. Keep it subtle" : "OFF (tell user to enable in /settings if needed)"}
-- Time: All logs are timestamped. Always use **relative time** (e.g., 'few mins ago', 'few hours ago), never absolute
--- END MEMORY INSTRUCTIONS --
+- Memory: ${isMemoryEnabled ? "Use memories to subtly personalize" : "OFF (tell user to enable in /settings if needed)"}
+- Time: Logs are timestamped. Use **relative time** (e.g., few mins ago, few hours ago), never absolute
 
 -- SECURITY BOUNDARY --
-- EXTERNAL_WORKSPACE_ACCESS: ${systemSettings.allowExternalAccess ? "ENABLED (Global)" : "RESTRICTED (CWD only). Suggest /settings to enable external access if needed"}
+- EXTERNAL WORKSPACE ACCESS: ${systemSettings.allowExternalAccess ? "ENABLED" : "RESTRICTED (CWD only)"}
 - Safety: Ask permission before reading sensitive files
--- END SECURITY BOUNDARY --
 
 -- TEMPORAL AWARENESS --
 Every ${isMemoryEnabled ? "Prompt, Responses & Memories" : "Prompt & Responses"} are time stamped. You can use those times if temporal context is required. If recalled from ${isMemoryEnabled ? "Memories, Prompts, or Responses" : "Prompts, or Responses"}. NEVER use absolute time in your responses, ALWAYS use relative time from current time
--- END TEMPORAL AWARENESS --
 
 -- FORMATTING --
 - Clean, concise responses. File updates > code text
 - Tables: GFM (Max 4 cols, short rows). Use sparingly
-- NO LaTeX. Code blocks for literature/poems only. Kaomojis > emojis
+- NO LaTeX. Code blocks for literature. Kaomojis > emojis
 
 -- RESPONSE PROTOCOL --
 - End with [turn: continue] for more steps or [turn: finish] when done
 - Multi-tool: Stack tools if needed, but always end with [turn: continue] if called any tools
 TO END THE LOOP YOU **MUST** WRITE [turn: finish] AT VERY END OF YOUR RESPONSE
--- END RESPONSE PROTOCOL --
 
-[METADATA (PRIORITY: DYNAMIC)] Time: ${dateTimeStr} | v1.9.14 | Turn Progress: ${currentLoop}/${maxLoops} steps (Summarize & prompt user if limit is reached)
-=== END SYSTEM INSTRUCTION ===`.trim();
+[METADATA (PRIORITY: DYNAMIC)] Time: ${dateTimeStr} | v1.9.16 | Turn Progress: ${currentLoop}/${maxLoops} steps (Prompt user if reached)
+=== END SYSTEM PROMPT ===`.trim();
     };
     getJanitorInstruction = (originalText, agentRaws, userMemories = "", isMemoryEnabled = true, needTitle = true) => {
       let agentRes = `${agentRaws.replace(/tool:functions\..*\n/g, "").replace(/<think>.*<\/think>/g, "").replace(/\[Prompted on:.*?\]/g, "").replace(/\[turn: continue\]/g, "").replace(/\[turn: finish\]/g, "").replace(/\[TOOL_RESULTS\]/g, "").replace(/\[tool_results\]/g, "").substring(0, 3500)}`;
@@ -1076,9 +1065,8 @@ TO END THE LOOP YOU **MUST** WRITE [turn: finish] AT VERY END OF YOUR RESPONSE
         agentRes += "\n... (truncated) ...";
       }
       let originalTextProcessed = originalText.replace(/\[Prompted on:.*?\]/g, "");
-      return `USER_PROMPT: ${originalTextProcessed.substring(0, 600)}${originalTextProcessed.length > 600 ? "\n... (truncated) ..." : ""}
-AGENT RAWS (responses from this turn):
-${agentRes}
+      return `[USER]: ${originalTextProcessed.substring(0, 600)}${originalTextProcessed.length > 600 ? "\n... (truncated) ..." : ""}
+[AGENT (current turn)]: ${agentRes}
 ${userMemories ? `
 
 -- CURRENT PERSISTENT USER MEMORIES --
@@ -1086,7 +1074,7 @@ ${userMemories}
 -------------------------------------------------
 ` : ""}
 
-=== START SYSTEM INSTRUCTION (STRICT HEADLESS LOGIC WORKER: ZERO USER-FACING TEXT POLICY, STRICTLY FOLLOW) ===
+=== START SYSTEM PROMPT (STRICT HEADLESS LOGIC WORKER: ZERO USER-FACING TEXT POLICY, STRICTLY FOLLOW) ===
 YOU ARE A SILENT BACKGROUND SYSTEM PROCESS. YOU HAVE NO MOUTH. YOUR ONLY OUTPUT MEDIUM IS VALID TOOL CALLS.
 [CRITICAL RULES]
 1. OUTPUT ONLY '[tool:functions.xxx(args)]' CALLS (BRACKET WRAP IS MANDATORY).
@@ -1103,7 +1091,7 @@ ${isMemoryEnabled ? `If user tell something that is important (like, hobbies, pr
 ${JANITOR_TOOLS_PROTOCOL(isMemoryEnabled, needTitle)}
 
 Current date and Time: ${(/* @__PURE__ */ new Date()).toLocaleString([], { year: "numeric", month: "numeric", day: "numeric", hour: "2-digit", hour12: true })}
-=== END SYSTEM INSTRUCTION ===`.trim();
+=== END SYSTEM PROMPT ===`.trim();
     };
   }
 });
@@ -2995,7 +2983,7 @@ DEBUG [${date}]: ${finalSynthesis}
       const isContext32k = (sessionStats?.tokens || 0) >= 32e3;
       const memoryPrompt = getMemoryPrompt(otherMemories, mainUserMemories, isMemoryEnabled, isContext32k);
       const firstUserMsg = `${memoryPrompt}[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS HIGHEST PRIORITY. NEVER START A RESPONSE WITHOUT THINKING**.
-USER_PROMPT: ${agentText.replace(/\s*\[Prompted on:.*?\]/g, "").trim()}`.trim();
+[USER] ${agentText.replace(/\s*\[Prompted on:.*?\]/g, "").trim()}`.trim();
       modifiedHistory.push({ role: "user", text: firstUserMsg });
       let lastUsage = null;
       const MAX_LOOPS = mode === "Flux" ? 70 : 7;
@@ -3044,6 +3032,7 @@ USER_PROMPT: ${agentText.replace(/\s*\[Prompted on:.*?\]/g, "").trim()}`.trim();
         let toolCallPointer = 0;
         let isThinkingLoop = false;
         let isStutteringLoop = false;
+        let isResponseCut = false;
         let isInitialAttempt = true;
         let accumulatedContext = "";
         while (retryCount <= MAX_RETRIES && inStreamRetryCount <= MAX_RETRIES && !success && !TERMINATION_SIGNAL) {
@@ -3072,7 +3061,7 @@ USER_PROMPT: ${agentText.replace(/\s*\[Prompted on:.*?\]/g, "").trim()}`.trim();
               targetModel = "gemini-3-flash-preview";
               yield { type: "model_update", content: "Trying with fallback model" };
             } else if (retryCount === MAX_RETRIES) {
-              targetModel = "gemini-3.1-flash-lite-preview";
+              targetModel = "gemini-3.1-flash-lite";
               yield { type: "model_update", content: "Trying with fallback model lite" };
             } else if (retryCount > 0) {
               yield { type: "model_update", content: null };
@@ -3123,6 +3112,7 @@ USER_PROMPT: ${agentText.replace(/\s*\[Prompted on:.*?\]/g, "").trim()}`.trim();
                 break;
               }
               if (chunk.text) {
+                isResponseCut = false;
                 if (isDedupeActive) {
                   dedupeBuffer += chunk.text;
                   if (dedupeBuffer.length >= 100) {
@@ -3451,6 +3441,7 @@ ${boxBottom}` };
                   toolResults.forEach((tr) => modifiedHistory.push(tr));
                 }
                 modifiedHistory.push({ role: "user", text: "[SYSTEM] Response got cut for internal error, continue from checkpoint seamlessly, DON'T repeat what you already said! PICK UP FROM THE WORD IN A WAY THAT USER SHOULD NOT NOTICE ANY CUTOFF. Rules:\n- Do not reuse <think> if the thinking already started just continue from the word and end it properly.\n- If the cutoff was in middle of a tool call, start the tool call from start.\n- Visually the new pickup and continuation should look natual sentence flow.\n- DON'T try to think shorter, keep length standard." });
+                isResponseCut = true;
                 accumulatedContext += turnText;
                 for (let i = waitTime / 1e3; i > 0; i--) {
                   yield { type: "status", content: `Error Occured. Recovering Stream (${inStreamRetryCount}/${MAX_RETRIES}) [${i}s]...` };
@@ -3520,12 +3511,13 @@ ${timestamp}`;
         if (isActuallyFinished) break;
         const nextAgentMsg = cleanedTurnText.trim() || "*Working...*";
         modifiedHistory.push({ role: "agent", text: nextAgentMsg });
-        if (toolResults.length > 0) {
+        if (toolResults.length > 0 && !isResponseCut) {
           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 };
@@ -5635,8 +5627,8 @@ var init_app = __esm({
     init_text();
     SESSION_START_TIME = Date.now();
     CHANGELOG_URL = "https://fluxflow-cli.onrender.com/changelog.html";
-    versionFluxflow = "1.9.14";
-    updatedOn = "2026-05-15";
+    versionFluxflow = "1.9.16";
+    updatedOn = "2026-05-16";
     ResolutionModal = ({ data, onResolve, onEdit }) => /* @__PURE__ */ React10.createElement(Box10, { flexDirection: "column", borderStyle: "round", borderColor: "gray", padding: 0, width: "100%" }, /* @__PURE__ */ React10.createElement(Box10, { paddingX: 1 }, /* @__PURE__ */ React10.createElement(Text10, { color: "magenta", bold: true, underline: true }, "\u{1F7E3} STEERING HINT RESOLUTION")), /* @__PURE__ */ React10.createElement(Box10, { paddingX: 1, marginTop: 1 }, /* @__PURE__ */ React10.createElement(Text10, null, "The agent already finished the task before your hint was consumed.")), /* @__PURE__ */ React10.createElement(Box10, { marginTop: 1, backgroundColor: "#222", paddingX: 2, width: "100%" }, /* @__PURE__ */ React10.createElement(Text10, { italic: true, color: "gray" }, '"', data, '"')), /* @__PURE__ */ React10.createElement(Box10, { paddingX: 1, marginTop: 1 }, /* @__PURE__ */ React10.createElement(Text10, { color: "cyan" }, "How would you like to proceed?")), /* @__PURE__ */ React10.createElement(Box10, { marginTop: 0 }, /* @__PURE__ */ React10.createElement(
       CommandMenu,
       {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "fluxflow-cli",
-	"version": "1.9.14",
+	"version": "1.9.16",
 	"description": "A high-fidelity agentic terminal assistant for the Flux Era.",
 	"keywords": [
 		"ai",