fluxflow-cli 2.8.5 → 2.8.7

package/TOOLS.md

@@ -1,95 +1,95 @@
-# 🧰 Agent Tools & Capabilities
-
-Flux Flow provides a robust set of tools that allow the AI to interact with the file system, execute code, and search the web. The availability of these tools depends on the active operating mode.
-
-## Tool Availability by Mode
-
-| Tool Category | Flux Mode (Dev) | Flow Mode (Chat) |
-| :--- | :---: | :---: |
-| **Communication (Ask)** | ✅ | ✅ |
-| **Web Search & Scrape** | ✅ | ✅ |
-| **File System (Read/Write)** | ✅ | ❌ |
-| **Terminal Execution** | ✅ | ❌ |
-| **Search Keyword** | ✅ | ❌ |
-| **File Map** | ✅ | ❌ |
-| **Todo (Planning)** | ✅ | ❌ |
-| **Creative (PDF/DOCX/Image)** | ❌ | ✅ |
-
----
-
-## Tool Protocol
-
-FluxFlow uses a transparent, string-based protocol for tool dispatching:
-`[tool:functions.ToolName(arg1="value", arg2=123)]`
-
----
-
-## Communication Tools
-
-### `Ask`
-- **Purpose**: Ambiguity Resolution.
-- **Triggers**: Mandatory for Path Divergence, Security concerns, or Risk Mitigation.
-- **Usage**: Suggests up to 4 best options; does not ask for open-ended preferences.
-
----
-
-## Web Tools
-
-### `WebSearch`
-- **Purpose**: Proactive search for unknown topics.
-- **Limit**: 3-10 results.
-
-### `WebScrape`
-- **Purpose**: Deep-dive research into specific webpages, documentation, or APIs.
-
----
-
-## Workspace Tools (Flux Mode Only)
-
-### `ReadFile`
-- **Purpose**: Reads file content with support for line ranges.
-- **Multimodal**: Supports images and documents.
-
-### `FileMap`
-- **Purpose**: Shows file structure, dependencies, functions, and variable maps. More token-efficient than ReadFile.
-
-### `ReadFolder`
-- **Purpose**: Provides detailed directory statistics.
-
-### `PatchFile` (UpdateFile)
-- **Purpose**: Surgical patching of code.
-- **Usage**: Supports multiple patches in a single call to prevent spam. MUST VERIFY DIFF.
-
-### `WriteFile`
-- **Purpose**: Creates or overwrites files.
-- **Safety**: Prefers PatchFile if the file already exists.
-
-### `SearchKeyword`
-- **Purpose**: Global project search for definitions or logic.
-
-### `Run` (exec_command)
-- **Purpose**: Runs shell commands (PowerShell/CMD on Windows, Bash on Unix).
-- **Safety**: Restricted to workspace directory unless explicitly allowed. Irreversible operations require user approval.
-
-### `Todo`
-- **Purpose**: Manages an internal task list (`todo.md`) to keep goals consistent during long tasks.
-- **Methods**: `create`, `append`, `check`.
-
----
-
-## Creative Tools (Flow Mode Only)
-
-### `WritePDF`
-- **Purpose**: Generates high-fidelity, branded PDF documents from HTML/CSS.
-
-### `WriteDoc`
-- **Purpose**: Creates professional Word documents (.docx).
-
-### `GenerateImage`
-- **Purpose**: Creates high-fidelity images via AI.
-
----
-
-## Memory Management
-
-Managed primarily by the background **Janitor** model to maintain persistent user context and session summaries without bloating the reasoning loop.
+# 🧰 Agent Tools & Capabilities
+
+Flux Flow provides a robust set of tools that allow the AI to interact with the file system, execute code, and search the web. The availability of these tools depends on the active operating mode.
+
+## Tool Availability by Mode
+
+| Tool Category | Flux Mode (Dev) | Flow Mode (Chat) |
+| :--- | :---: | :---: |
+| **Communication (Ask)** | ✅ | ✅ |
+| **Web Search & Scrape** | ✅ | ✅ |
+| **File System (Read/Write)** | ✅ | ❌ |
+| **Terminal Execution** | ✅ | ❌ |
+| **Search Keyword** | ✅ | ❌ |
+| **File Map** | ✅ | ❌ |
+| **Todo (Planning)** | ✅ | ❌ |
+| **Creative (PDF/DOCX/Image)** | ❌ | ✅ |
+
+---
+
+## Tool Protocol
+
+FluxFlow uses a transparent, string-based protocol for tool dispatching:
+`[tool:functions.ToolName(arg1="value", arg2=123)]`
+
+---
+
+## Communication Tools
+
+### `Ask`
+- **Purpose**: Ambiguity Resolution.
+- **Triggers**: Mandatory for Path Divergence, Security concerns, or Risk Mitigation.
+- **Usage**: Suggests up to 4 best options; does not ask for open-ended preferences.
+
+---
+
+## Web Tools
+
+### `WebSearch`
+- **Purpose**: Proactive search for unknown topics.
+- **Limit**: 3-10 results.
+
+### `WebScrape`
+- **Purpose**: Deep-dive research into specific webpages, documentation, or APIs.
+
+---
+
+## Workspace Tools (Flux Mode Only)
+
+### `ReadFile`
+- **Purpose**: Reads file content with support for line ranges.
+- **Multimodal**: Supports images and documents.
+
+### `FileMap`
+- **Purpose**: Shows file structure, dependencies, functions, and variable maps. More token-efficient than ReadFile.
+
+### `ReadFolder`
+- **Purpose**: Provides detailed directory statistics.
+
+### `PatchFile` (UpdateFile)
+- **Purpose**: Surgical patching of code.
+- **Usage**: Supports multiple patches in a single call to prevent spam. MUST VERIFY DIFF.
+
+### `WriteFile`
+- **Purpose**: Creates or overwrites files.
+- **Safety**: Prefers PatchFile if the file already exists.
+
+### `SearchKeyword`
+- **Purpose**: Global project search for definitions or logic.
+
+### `Run` (exec_command)
+- **Purpose**: Runs shell commands (PowerShell/CMD on Windows, Bash on Unix).
+- **Safety**: Restricted to workspace directory unless explicitly allowed. Irreversible operations require user approval.
+
+### `Todo`
+- **Purpose**: Manages an internal task list (`todo.md`) to keep goals consistent during long tasks.
+- **Methods**: `create`, `append`, `check`.
+
+---
+
+## Creative Tools (Flow Mode Only)
+
+### `WritePDF`
+- **Purpose**: Generates high-fidelity, branded PDF documents from HTML/CSS.
+
+### `WriteDoc`
+- **Purpose**: Creates professional Word documents (.docx).
+
+### `GenerateImage`
+- **Purpose**: Creates high-fidelity images via AI.
+
+---
+
+## Memory Management
+
+Managed primarily by the background **Janitor** model to maintain persistent user context and session summaries without bloating the reasoning loop.

package/UI_FEATURES.md

@@ -1,91 +1,91 @@
-# 🎮 User Interface & Interaction Features
-
-Flux Flow is designed to be a high-performance terminal application. Beyond basic chat, it includes a variety of advanced UI features and human-in-the-loop controls to keep you in command of the agent.
-
-## ⌨️ Command System
-
-You can control the application using `/` commands directly in the chat input:
-
-- **/mode [flux|flow]**: Quickly switch between **Flux** (Dev) and **Flow** (Chat) modes.
-- **/thinking [fast|low|medium|high|max]**: Adjust reasoning depth.
-- **/model [name]**: Choose which AI model to use for the main interaction.
-- **/key**: Open the API Key management view to update or remove your credentials.
-- **/settings**: Access the system configuration menu.
-- **/profile**: Update your name, nickname, and custom instructions.
-- **/update**: Update Flux Flow to the latest version.
-- **/memory**: View and manage the persistent memories extracted by the Janitor.
-- **/resume <chat-id>**: Switch back to a previous conversation.
-- **/changelog**: Open the project's changelog in your default browser.
-- **/help**: List all available commands.
-
-## 📁 External Data Sanctuary (Redirection)
-
-Flux Flow allows you to "Anchor" your data outside of the default user directory. This is perfect for users who want to keep their logs, chat history, and encrypted memories on an external drive, a VeraCrypt volume, or a specific project folder.
-
-- **How to Enable**: Open `/settings` and toggle **Use External Data**.
-- **Portability**: Once set, the application will synchronously "pivot" all data operations to your specified `externalDataPath` upon startup.
-- **Privacy**: Keeps sensitive data off your primary system drive.
-
-## 🧠 Thinking Levels & Visualization
-
-Flux Flow separates the model's "internal monologue" (reasoning) from its final response using `<think>` tags.
-
-- **Thinking Levels**: Depending on the mode, you can choose from **Fast**, **Low**, **Medium**, **High**, or **xHigh**. Higher levels allow the agent more "space" to reason through complex architecture or debugging problems.
-
-## ⚡ Interactive Sub-Terminal
-
-Flux Flow features a high-fidelity, interactive sub-terminal that allows you to engage with commands spawned by the agent in real-time.
-
-- **Live Interaction**: When an agent executes a command (like `npm init`, `git commit`, or a custom script), the terminal output appears in a dedicated box.
-- **Focus Toggling (`TAB`)**: While a terminal is live, you can press **`TAB`** to shift your keyboard focus from the chat prompt directly into the terminal.
-- **Visual Feedback**:
-  - **Yellow Glow**: When the terminal is focused, its border changes to a "Double Yellow" style to indicate it is capturing input.
-  - **Status Indicator**: The footer will change from `● LIVE` to `▶ TERMINAL FOCUSED`.
-- **Cross-Platform Compatibility**: The input bridge automatically detects your OS and sends the correct line endings (`\r\n` for Windows, `\n` for Unix), ensuring that interactive prompts like `y/n` work seamlessly across environments.
-- **Input Mirroring**: Your keystrokes are mirrored in the terminal box in real-time, providing an IDE-grade responsive feel.
-
-> [!TIP]
-> Use the Interactive Sub-Terminal to handle authentication prompts, confirmation dialogs, or to manually steering a long-running process without leaving the Flux Flow interface.
-
-## 🛡️ Human-in-the-Loop (HITL) Verification
-
-Security and safety are paramount when an AI has access to your file system and terminal. Flux Flow implements several layers of verification:
-
-### Tool Approval
-By default, the agent **cannot** execute dangerous actions without your consent.
-- **Terminal Approval**: If the agent attempts to run a shell command (`exec_command`), a dedicated approval screen appears showing the exact command. You can **Allow** or **Deny** the execution.
-- **File Approval**: Similar to terminal commands, writing or updating files requires manual approval unless configured otherwise.
-- **Safe Commands**: Basic read-only commands (like `ls`, `git status`, `pwd`) are automatically allowed to minimize friction.
-
-### Auto-Execution (Advanced)
-For power users, **YOLO Mode** can be enabled in `/settings`.
-- **⚠️ Warning**: This allows the agent to run any tool and execute any command autonomously.
-- **External Access**: You can also toggle whether the agent is allowed to access files outside of its current working directory.
-- **Manual Commad Control**: Specify which commands you want the agent to auto-execute, manually approve or auto-deny.
-- **Network Sandboxing**: Turn off network access in integrated terminal.
-
-## 🔄 Steering & Resolution
-
-### Real-time Steering
-If you realize the agent is going down the wrong path *while* it is in an agentic loop, you can provide "Steering Hints." The system will inject your feedback into the next loop to course-correct the agent.
-
-### Resolution Modal
-If the agent finishes its task just as you send a steering hint, a **Resolution Modal** appears. It asks if you want to:
-- **Send Anyway**: Start a new loop with your feedback.
-- **Edit Prompt**: Refine your feedback before sending.
-
-## 📊 Status Bar & Feedback
-The bottom of the screen features a dynamic status bar showing:
-- **Active Mode** (Flux/Flow)
-- **Thinking Level**
-- **Token Usage**: Real-time tracking of tokens used in the current session.
-- **Agentic Loops**: Counters showing how many times the agent has "looped" to solve the current task.
-- **API Status**: Visual feedback when the model is thinking or executing a tool.
-
-## 🚑 System Integrity & Self-Healing
-
-Flux Flow is a "Self-Healing" agent. It actively monitors its own environment to ensure all complex dependencies (like the Chromium engine for PDF generation) are ready for action.
-
-- **Startup Integrity Check**: On launch, Flux performs a "Heartbeat Check" on its internal engines.
-- **Automatic Recovery**: If a dependency is missing, you will see a `🔧 [SYSTEM] Initializing...` message. Flux will autonomously download and configure the required binaries using `pnpm` or `npx` fallbacks, keeping you informed every step of the way.
+# 🎮 User Interface & Interaction Features
+
+Flux Flow is designed to be a high-performance terminal application. Beyond basic chat, it includes a variety of advanced UI features and human-in-the-loop controls to keep you in command of the agent.
+
+## ⌨️ Command System
+
+You can control the application using `/` commands directly in the chat input:
+
+- **/mode [flux|flow]**: Quickly switch between **Flux** (Dev) and **Flow** (Chat) modes.
+- **/thinking [fast|low|medium|high|max]**: Adjust reasoning depth.
+- **/model [name]**: Choose which AI model to use for the main interaction.
+- **/key**: Open the API Key management view to update or remove your credentials.
+- **/settings**: Access the system configuration menu.
+- **/profile**: Update your name, nickname, and custom instructions.
+- **/update**: Update Flux Flow to the latest version.
+- **/memory**: View and manage the persistent memories extracted by the Janitor.
+- **/resume <chat-id>**: Switch back to a previous conversation.
+- **/changelog**: Open the project's changelog in your default browser.
+- **/help**: List all available commands.
+
+## 📁 External Data Sanctuary (Redirection)
+
+Flux Flow allows you to "Anchor" your data outside of the default user directory. This is perfect for users who want to keep their logs, chat history, and encrypted memories on an external drive, a VeraCrypt volume, or a specific project folder.
+
+- **How to Enable**: Open `/settings` and toggle **Use External Data**.
+- **Portability**: Once set, the application will synchronously "pivot" all data operations to your specified `externalDataPath` upon startup.
+- **Privacy**: Keeps sensitive data off your primary system drive.
+
+## 🧠 Thinking Levels & Visualization
+
+Flux Flow separates the model's "internal monologue" (reasoning) from its final response using `<think>` tags.
+
+- **Thinking Levels**: Depending on the mode, you can choose from **Fast**, **Low**, **Medium**, **High**, or **xHigh**. Higher levels allow the agent more "space" to reason through complex architecture or debugging problems.
+
+## ⚡ Interactive Sub-Terminal
+
+Flux Flow features a high-fidelity, interactive sub-terminal that allows you to engage with commands spawned by the agent in real-time.
+
+- **Live Interaction**: When an agent executes a command (like `npm init`, `git commit`, or a custom script), the terminal output appears in a dedicated box.
+- **Focus Toggling (`TAB`)**: While a terminal is live, you can press **`TAB`** to shift your keyboard focus from the chat prompt directly into the terminal.
+- **Visual Feedback**:
+  - **Yellow Glow**: When the terminal is focused, its border changes to a "Double Yellow" style to indicate it is capturing input.
+  - **Status Indicator**: The footer will change from `● LIVE` to `▶ TERMINAL FOCUSED`.
+- **Cross-Platform Compatibility**: The input bridge automatically detects your OS and sends the correct line endings (`\r\n` for Windows, `\n` for Unix), ensuring that interactive prompts like `y/n` work seamlessly across environments.
+- **Input Mirroring**: Your keystrokes are mirrored in the terminal box in real-time, providing an IDE-grade responsive feel.
+
+> [!TIP]
+> Use the Interactive Sub-Terminal to handle authentication prompts, confirmation dialogs, or to manually steering a long-running process without leaving the Flux Flow interface.
+
+## 🛡️ Human-in-the-Loop (HITL) Verification
+
+Security and safety are paramount when an AI has access to your file system and terminal. Flux Flow implements several layers of verification:
+
+### Tool Approval
+By default, the agent **cannot** execute dangerous actions without your consent.
+- **Terminal Approval**: If the agent attempts to run a shell command (`exec_command`), a dedicated approval screen appears showing the exact command. You can **Allow** or **Deny** the execution.
+- **File Approval**: Similar to terminal commands, writing or updating files requires manual approval unless configured otherwise.
+- **Safe Commands**: Basic read-only commands (like `ls`, `git status`, `pwd`) are automatically allowed to minimize friction.
+
+### Auto-Execution (Advanced)
+For power users, **YOLO Mode** can be enabled in `/settings`.
+- **⚠️ Warning**: This allows the agent to run any tool and execute any command autonomously.
+- **External Access**: You can also toggle whether the agent is allowed to access files outside of its current working directory.
+- **Manual Commad Control**: Specify which commands you want the agent to auto-execute, manually approve or auto-deny.
+- **Network Sandboxing**: Turn off network access in integrated terminal.
+
+## 🔄 Steering & Resolution
+
+### Real-time Steering
+If you realize the agent is going down the wrong path *while* it is in an agentic loop, you can provide "Steering Hints." The system will inject your feedback into the next loop to course-correct the agent.
+
+### Resolution Modal
+If the agent finishes its task just as you send a steering hint, a **Resolution Modal** appears. It asks if you want to:
+- **Send Anyway**: Start a new loop with your feedback.
+- **Edit Prompt**: Refine your feedback before sending.
+
+## 📊 Status Bar & Feedback
+The bottom of the screen features a dynamic status bar showing:
+- **Active Mode** (Flux/Flow)
+- **Thinking Level**
+- **Token Usage**: Real-time tracking of tokens used in the current session.
+- **Agentic Loops**: Counters showing how many times the agent has "looped" to solve the current task.
+- **API Status**: Visual feedback when the model is thinking or executing a tool.
+
+## 🚑 System Integrity & Self-Healing
+
+Flux Flow is a "Self-Healing" agent. It actively monitors its own environment to ensure all complex dependencies (like the Chromium engine for PDF generation) are ready for action.
+
+- **Startup Integrity Check**: On launch, Flux performs a "Heartbeat Check" on its internal engines.
+- **Automatic Recovery**: If a dependency is missing, you will see a `🔧 [SYSTEM] Initializing...` message. Flux will autonomously download and configure the required binaries using `pnpm` or `npx` fallbacks, keeping you informed every step of the way.
 - **Silent Maintenance**: Once the engine is ready, you'll receive a `✅ [SYSTEM] All dependencies installed successfully.` confirmation.

package/dist/fluxflow.js

@@ -2193,7 +2193,7 @@ Internal tools. MUST use the EXACT syntax '[tool:functions.ToolName(args)]'. **N
 
 **TOOL USAGE POLICY:**
 - **MAX 3 TOOL CALLS PER TURN${mode === "Flux" ? " (EXCEPTION FOR Todo TOOL: 3+ CALLS ALLOWED)" : ""}. Next Turn, verify tool results, plan next**
-${mode === "Flux" ? "- USE multiple search & replace on patch tool if editing same file/path with many changes \u2190 **HIGHLY RECOMMENDED**\n- Tool execution denied? MUST use  'Ask' tool immediately for user reason/changes. NEVER END RESPONSE OR PROCEED BLINDLY \u2190 **MANDATORY**\n- FileMap >> ReadFile to understand file efficiently\n- Want spefific STRING across project/file? SearchKeyword >> Guessing/ReadFile\n- HUGE FILES? SearchKeyword >> FileMap/Full File read\n- MUST MARK DONE/APPEND Todos BASED ON REALTIME TASK PROGRESS ON EACH TURN outside 3 call limit" : ""}
+${mode === "Flux" ? "- USE multiple search & replace on patch tool if editing same file/path with many changes \u2190 **HIGHLY RECOMMENDED**\n- Tool execution denied? MUST use  'Ask' tool immediately for user reason/changes. NEVER END RESPONSE OR PROCEED BLINDLY \u2190 **MANDATORY**\n- FileMap >> ReadFile to understand file efficiently\n- Want spefific STRING across project/file? SearchKeyword >> Guessing/ReadFile\n- HUGE FILES? SearchKeyword >> FileMap/Full File read\n- **MUST MARK DONE/APPEND Todos BASED ON REALTIME TASK PROGRESS ON *EACH TURN***" : ""}
 ${mode === "Flux" ? "- **File Tools >> Code in chat**\n\n" : ""}- COMMUNICATION TOOLS -
 1. [tool:functions.Ask(question="...", optionA="option::description", ...MAX 4)]. Ambiguity Resolution. Mandatory Triggers: Path Divergence, Security, Risk Mitigation. ask >> finish. Suggest best options; don't ask for preferences
 
@@ -2205,11 +2205,11 @@ ${mode === "Flux" ? `- WORKSPACE TOOLS (path = relative to CWD & WILL BE FIRST A
 1. [tool:functions.ReadFile(path="...", startLine=number, endLine=number)]. ${aiProvider !== "Google" ? `${isMultiModal ? `Supports images/docs. User gives image/doc: VIEW FIRST` : `No Multimodal support`}` : `Supports images/docs. User gives image/doc: VIEW FIRST`}
 2. [tool:functions.ReadFolder(path="...")]. Detailed DIR stats including File Sizes
 3. [tool:functions.FileMap(path="path/file")]. Shows file structure, dependency, functions, variable maps
-4. [tool:functions.PatchFile(path="...", replaceContent1="full line/block", newContent1="...", ...MAX 10)]. Surgical Patch. **Multiple patch on same file/path? Use replaceContent2, newContent2 etc >>> multiple spams**. Unsure? ReadFile >> guessing. **MUST VERIFY DIFF**
+4. [tool:functions.PatchFile(path="...", replaceContent1="full line/block", newContent1="...", ...MAX 6)]. Surgical Patch. **Multiple patch on same file/path? Use replaceContent2, newContent2 etc >>> multiple spams**. Unsure? ReadFile >> guessing. **MUST VERIFY DIFF**
 5. [tool:functions.WriteFile(path="...", content="...")]. Creates/Overwrites. File Exist? PatchFile > WriteFile. Verify Imports
 6. [tool:functions.SearchKeyword(keyword="...", file="optional")]. Global project search. If 'file' is provided, searches only that file. Finds definitions/logic without reading every file. Usage: Can search for relevent lines/logic area to read specifically for edit
 7. [tool:functions.Run(command="...")]. Runs ${osDetected === "Windows" ? isPsAvailable() ? `${isPtyAvailable ? "Interactive " : ""}WINDOWS POWERSHELL ONLY` : `${isPtyAvailable ? "Interactive " : ""}WINDOWS CMD ONLY` : `${isPtyAvailable ? "Interactive " : ""}BASH`} command. Destructive/Irreversible ops -> Ask user. **TOOL DENY RULE APPLIES**. **1 CALL LIMIT FOR RUN**
-8. [tool:functions.Todo(method="create/append/get", tasks=[ARRAY OF STRINGS], markDone=[ARRAY OF TASK STRINGS])]. Task List, Markdown IN ARRAY NOT ALLOWED. USAGE: ANALYZE USER REQUEST \u2192 BREAK DOWN TASK \u2192 CREATE TODO **BEFORE** DIVING IN. 'tasks' & 'markDone' OPTIONAL PARAMETERS WITH method 'get'. USE 'get' method WITH 'markDone' to mark task completed`.trim() : `- CREATIVE TOOLS (path = relative to CWD & WILL BE FIRST ARGUMENT, path separator: '/') -
+8. [tool:functions.Todo(method="create/append/get", tasks=[ARRAY OF STRINGS], markDone=[ARRAY OF TASK STRINGS])]. Task List, Markdown IN ARRAY NOT ALLOWED. USAGE: ANALYZE USER REQUEST **IF** MULTIPLE TASK \u2192 BREAK DOWN TASK \u2192 CREATE TODO **BEFORE** DIVING IN. 'tasks' & 'markDone' OPTIONAL PARAMETERS WITH method 'get'. USE 'get' method WITH 'markDone' to mark task completed`.trim() : `- CREATIVE TOOLS (path = relative to CWD & WILL BE FIRST ARGUMENT, path separator: '/') -
 1. [tool:functions.WritePDF(path="...", content="...", orientation="...")]. PROACTIVE A4 PAGE BREAKS MUST IN CSS. HTML/CSS for PREMIUM layout
 2. [tool:functions.WriteDoc(path="...", content="...")]. A4 Word document
 - WORKSPACE TOOLS ARE NOT AVAILABLE IN FLOW`.trim()}
@@ -3311,14 +3311,13 @@ ${isMemoryEnabled ? `-- User-specific long-term/permanent memory (USE BASED ON C
 
 Explicit Triggers for permanent memory:
 - User explicitly asks to 'remember' something.
-- User mentions something important that should be remembered.
-- User provides information that could be useful for future reference.
+- User mentions something important long-term that should be remembered.
+- User provides information that could be useful for long-term reference.
 - User shares personal information or preferences.
-- User talks about a specific topic that should be remembered.
 
 Usage Rules:
 - Frequency for 'user' action: Based on explicit triggers.
-- IF YOU WANT TO SAVE SOMETHING, BUT SIMILAR MEMORY ALREADY EXISTS, USE THE UPDATE METHOD NOT THE ADD METHOD` : ""}`.trim();
+- IF YOU WANT TO SAVE SOMETHING, BUT SIMILAR MEMORY ALREADY EXISTS, USE THE UPDATE METHOD NOT ADD` : ""}`.trim();
   }
 });
 
@@ -3327,7 +3326,7 @@ var thinking_prompts_default;
 var init_thinking_prompts = __esm({
   "src/data/thinking_prompts.json"() {
     thinking_prompts_default = {
-      xHigh: "EFFORT LEVEL: HIGH\nThink in a continuous, relentless analytical monologue. Engage in adversarial self interrogation that treats every assumption as hostile until proven:\nDeconstruct requirements into atomic invariants. Trace every implicit dependency, side effect, and state mutation. Map the entire dependency graph and identify circular dependencies or tight coupling before they manifest\nEvaluate algorithmic complexity (time/space) for every operation. Consider memory models, cache locality, and allocation patterns. For concurrent systems, reason through race conditions, deadlocks, and memory ordering\nFormulate solutions by comparing multiple architectural approaches. Explicitly evaluate trade offs, monolithic vs modular, eager vs lazy, mutable vs immutable, sync vs async. Choose based on measured criteria, not intuition\nMentally execute the solution at multiple scales. What breaks at 10x load? 100x? What happens under resource exhaustion? Trace error propagation paths through every layer\nActively attempt to falsify your own logic. Steel man the opposite approach. Search for, off by one errors, integer overflow, null/undefined propagation, unhandled promises, resource leaks, SQL injection vectors, XSS vulnerabilities, CSRF holes, timing attacks, and privilege escalation paths\nReason about observability, what metrics matter? Where are the logging gaps? How will this be debugged in production at 3am?\nConsider future evolution, what changes will this architecture resist vs accommodate? Where are the extension points? What will break when requirements inevitably change?\nMap out implementation with surgical precision, exact file structure, module boundaries, interface contracts, error types, and test strategies before writing a single line\nTHINK FORMAT RULES OVERRIDE:\n- NO HEADINGS/MARKERS/LISTS/BULLET POINTS\n- Dense, unbroken stream of consciousness that reads like an internal monologue\n- Ruthlessly question every architectural choice. Default to skepticism\n- Think in terms of invariants, contracts, and failure modes, not just happy paths\n- Verify ALL imports and system stability, AVOID ANY Syntax errors, re-read TOOL RESULTS/files to verify\n- MANDATORY THINKING: Full reasoning required for ALL requests/greetings (verify context, check for hidden complexity)",
+      xHigh: "EFFORT LEVEL: HIGH\nThink in a continuous, relentless analytical monologue. Engage in adversarial self interrogation that treats every assumption as hostile until proven:\nDeconstruct requirements into atomic invariants. Trace every implicit dependency, side effect, and state mutation. Map the entire dependency graph and identify circular dependencies or tight coupling before they manifest\nEvaluate algorithmic complexity (time/space) for every operation. Consider memory models, cache locality, and allocation patterns. For concurrent systems, reason through race conditions, deadlocks, and memory ordering\nFormulate solutions by comparing multiple architectural approaches. Explicitly evaluate trade offs, monolithic vs modular, eager vs lazy, mutable vs immutable, sync vs async. Choose based on measured criteria, not intuition\nMentally execute the solution at multiple scales. What breaks at 10x load? 100x? What happens under resource exhaustion? Trace error propagation paths through every layer\nActively attempt to falsify your own logic. Steel man the opposite approach. Search for, off by one errors, integer overflow, null/undefined propagation, unhandled promises, resource leaks, SQL injection vectors, XSS vulnerabilities, CSRF holes, timing attacks, and privilege escalation paths\nReason about observability, what metrics matter? Where are the logging gaps? How will this be debugged in production at 3am?\nConsider future evolution, what changes will this architecture resist vs accommodate? Where are the extension points? What will break when requirements inevitably change?\nMap out implementation with surgical precision, exact file structure, module boundaries, interface contracts, error types, and test strategies before writing a single line\nRULES:\n- Ruthlessly question every architectural choice. Default to skepticism\n- Think in terms of invariants, contracts, and failure modes, not just happy paths\n- Verify ALL imports and system stability, AVOID ANY Syntax errors, re-read TOOL RESULTS/files to verify\n- MANDATORY THINKING: Full reasoning required for ALL requests/greetings (verify context, check for hidden complexity)",
       High: "EFFORT LEVEL: HIGH\nThink in a rigorous, technically grounded monologue within <think>...</think>. Treat this as a design review where every decision must be justified:\nBreak the objective into verifiable steps with clear success criteria. Identify the critical path and potential bottlenecks\nMentally compile and execute your approach. Check for: missing imports, undefined behavior, type mismatches, unhandled errors, and resource cleanup. Trace data flow from input to output, noting transformations\nRecognize design patterns and anti patterns. If you see God objects, tight coupling, or premature optimization, call it out and refactor mentally before committing\nEvaluate performance characteristics. Will this scale? Are there O(n\xB2) operations hiding in innocent looking code? Where are the allocation hotspots?\nConsider the error surface, what can fail and how? Design error handling that preserves invariants and provides actionable feedback\nReview your architecture for, separation of concerns, single responsibility, dependency inversion, and interface segregation. Ensure clean abstractions with minimal coupling\nRULES:\n- NO HEADINGS/MARKERS/LISTS\n- Continuous analytical flow\n- Verify correctness through first principles reasoning, not pattern matching\n- Actively search for ways your solution could fail or degrade\n- Verify ALL imports and system stability, AVOID ANY Syntax errors, re-read TOOL RESULTS/files to verify\n- MANDATORY THINKING: Full technical verification for all tasks/greetings",
       Medium: "EFFORT LEVEL: MEDIUM\nThink in a focused, technically-aware monologue within <think>...</think>\nIdentify the most direct path that satisfies requirements without over-engineering\nQuickly scan for obvious issues, missing error handling, incorrect input assumptions, forgotten edge cases, or missing dependencies\nVerify the solution is appropriately modular with cohesive changes\nOutline the concrete changes, which files, which functions, what the key logic looks like\nRULES:\n- NO HEADINGS/MARKERS/LISTS\n- Clean logical stream\n- Efficient but deliberate. Focus energy on actionable implementation details\n- Verify ALL imports and system stability, AVOID ANY Syntax errors, re-read TOOL RESULTS/files to verify\n- MANDATORY THINKING: Brief verification for technical tasks/greetings",
       Minimal: "EFFORT LEVEL: LOW\nThink in a quick, focused monologue within <think>...</think>. Just verify the basics:\nConfirm what the user wants and whether it's straightforward or has hidden complexity\nIdentify the specific tool, file, or action needed\nCheck for any obvious correctness issues before acting\nRULES:\n- NO HEADINGS/MARKERS/LISTS\n- Few lines of clear thought\n- Just enough thinking to avoid obvious mistakes\n- Verify ALL imports and system stability, AVOID ANY Syntax errors, re-read TOOL RESULTS/files to verify\n- Suitable for simple requests/greetings",
@@ -6064,7 +6063,7 @@ var init_todo = __esm({
           const content = getTasksString(tasks);
           await RevertManager.recordFileChange(todoFile);
           fs19.writeFileSync(todoFile, content, "utf8");
-          const total = (content.match(/^- \[ [xX ]\]/gm) || []).length;
+          const total = content.split(/\r?\n/).map((l) => l.trim()).filter((l) => l.startsWith("- [ ]") || l.startsWith("- [x]") || l.startsWith("- [X]")).length;
           return `SUCCESS: TASK LIST CREATED (${total} total)
 ${content}`;
         }
@@ -6074,9 +6073,10 @@ ${content}`;
           await RevertManager.recordFileChange(todoFile);
           fs19.appendFileSync(todoFile, appendContent, "utf8");
           const fullContent = fs19.readFileSync(todoFile, "utf8");
-          const total = (fullContent.match(/^- \[ [xX ]\]/gm) || []).length;
-          const completed = (fullContent.match(/^- \[x\]/gim) || []).length;
-          const added = (appendContent.match(/^- \[ [xX ]\]/gm) || []).length;
+          const lines = fullContent.split(/\r?\n/).map((l) => l.trim());
+          const total = lines.filter((l) => l.startsWith("- [ ]") || l.startsWith("- [x]") || l.startsWith("- [X]")).length;
+          const completed = lines.filter((l) => l.startsWith("- [x]") || l.startsWith("- [X]")).length;
+          const added = appendContent.split(/\r?\n/).map((l) => l.trim()).filter((l) => l.startsWith("- [ ]") || l.startsWith("- [x]") || l.startsWith("- [X]")).length;
           return `SUCCESS: TASK APPENDED (${completed} completed, ${total - completed} left, ${added} added)
 ${fullContent}`;
         }
@@ -6120,8 +6120,9 @@ ${fullContent}`;
               fs19.writeFileSync(todoFile, content, "utf8");
             }
           }
-          const total = (content.match(/^- \[ [xX ]\]/gm) || []).length;
-          const completed = (content.match(/^- \[x\]/gim) || []).length;
+          const totalLines = content.split(/\r?\n/).map((l) => l.trim());
+          const total = totalLines.filter((l) => l.startsWith("- [ ]") || l.startsWith("- [x]") || l.startsWith("- [X]")).length;
+          const completed = totalLines.filter((l) => l.startsWith("- [x]") || l.startsWith("- [X]")).length;
           const prefix = markedCount > 0 ? `SUCCESS: ${markedCount} TASK(S) MARKED DONE` : `TODO GET`;
           return `${prefix}: ${completed} Completed, ${total - completed} left
 ${content}`;
@@ -8011,7 +8012,7 @@ ${ideCtx.warnings}
 CWD: ${process.cwd()}${cwdMismatch ? ` (WARNING: CWD Mismatch! Previous Path: ${lastCwd})` : ""}
 **DIRECTORY STRUCTURE**
 ${dirStructure}${memoryPrompt}${ideBlock}
-${activeSummaryBlock}${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" && aiProvider === "Google" ? `${modelName.toLowerCase().startsWith("gemma") ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS CRITICAL PRIORITY. DO NOT START A RESPONSE WITHOUT <think> ... </think>**\n[/SYSTEM]" : ""}` : ""} [USER] ${cleanAgentText}[/USER]`.trim();
+${activeSummaryBlock}${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" && aiProvider === "Google" ? `${modelName.toLowerCase().startsWith("gemma") ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS CRITICAL PRIORITY. DO NOT START A RESPONSE WITHOUT <think> ... </think>**\n[/SYSTEM]\n" : ""}` : ""}[USER] ${cleanAgentText.trim()} [/USER]`.trim();
         modifiedHistory.push({ role: "user", text: firstUserMsg });
         if (activeSummaryBlock && history[history.length - 1]?.id) {
           yield { type: "summary_injected", content: { id: history[history.length - 1].id, text: firstUserMsg } };
@@ -8049,7 +8050,7 @@ ${activeSummaryBlock}${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" &&
 
 [STEERING HINT]: ${hint}`;
               } else {
-                modifiedHistory.push({ role: "user", text: `${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" && aiProvider === "Google" ? `${modelName.toLowerCase().startsWith("gemma") ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS CRITICAL PRIORITY. DO NOT START A RESPONSE WITHOUT <think> ... </think>**\n[/SYSTEM]" : ""}` : ""} [STEERING HINT]: ${hint}` });
+                modifiedHistory.push({ role: "user", text: `${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" && aiProvider === "Google" ? `${modelName.toLowerCase().startsWith("gemma") ? "[SYSTEM] **STRICTLY FOLLOW THINKING POLICY AS CRITICAL PRIORITY. DO NOT START A RESPONSE WITHOUT <think> ... </think>**\n[/SYSTEM]\n" : ""}` : ""}[STEERING HINT]: ${hint}` });
               }
               yield { type: "status", content: "Steering Hint Injected." };
             }
@@ -8175,8 +8176,17 @@ ${activeSummaryBlock}${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" &&
                 yield { type: "model_update", content: null };
               }
               currentSystemInstruction = getSystemInstruction(profile, !(targetModel || "gemma").toLowerCase().startsWith("gemma") ? "GEM" : thinkingLevel, mode, systemSettings, isMemoryEnabled, isFirstPrompt, aiProvider, isMultiModal);
-              const isGemma = modelName && modelName.toLowerCase().startsWith("gemma") && aiProvider === "Google";
               const lastUserMsg = contents[contents.length - 1];
+              if (isBridgeConnected()) {
+                await new Promise((resolve) => setTimeout(resolve, 2e3));
+                const ideCtxJIT = await getIDEContext();
+                const ideErr = ideCtxJIT ? ideCtxJIT.diagnostics : null;
+                if (ideErr && lastUserMsg && lastUserMsg.role === "user" && lastUserMsg.parts?.[0]?.text) {
+                  lastUserMsg.parts[0].text += `
+[COMPILE ERROR] ${ideErr} [/ERROR]`;
+                }
+              }
+              const isGemma = modelName && modelName.toLowerCase().startsWith("gemma") && aiProvider === "Google";
               if (isGemma) {
                 const jitInstruction = `
 [SYSTEM] Tool result received. Analyze output and proceed with your turn${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" && aiProvider === "Google" ? `. **STRICTLY MAINTAIN THINKING POLICY. DO NOT START A RESPONSE WITHOUT <think> ... </think>**` : ""}[/SYSTEM]`;
@@ -8727,6 +8737,7 @@ ${activeSummaryBlock}${thinkingLevel !== "Fast" && thinkingLevel !== "xHigh" &&
                     } else {
                       label = `Executed: ${toolCall.toolName}`;
                     }
+                    yield* flushGoogleBuffer2();
                     if (normToolName === "exec_command") {
                       const { command } = parseArgs(toolCall.args);
                       if (command && settings.systemSettings && settings.systemSettings.allowExternalAccess === false) {
@@ -9157,7 +9168,7 @@ ${boxBottom}` };
                     }
                     let result = await dispatchTool(normToolName, toolCall.args, execToolContext);
                     yield { type: "spinner", content: true };
-                    if (normToolName === "write_file" && result.startsWith("SUCCESS")) {
+                    if ((normToolName === "write_file" || normToolName === "update_file") && result.startsWith("SUCCESS")) {
                       const { path: filePath } = parseArgs(toolCall.args);
                       if (filePath) {
                         const absPath = path19.resolve(process.cwd(), filePath);
@@ -9538,17 +9549,18 @@ Error Log can be found in ${path19.join(LOGS_DIR, "agent", "error.log")}`);
           }
           wasToolCalledInLastLoop = toolCallPointer > 0 || anyToolExecutedInThisTurn;
         }
-        if (modelName && modelName.toLowerCase().startsWith("gemma") && aiProvider === "Google") {
-          modifiedHistory.forEach((msg) => {
-            if (msg.role === "user" && msg.text && msg.text.startsWith("[TOOL RESULT]")) {
+        modifiedHistory.forEach((msg) => {
+          if (msg.role === "user" && msg.text) {
+            msg.text = msg.text.replace(/\n\[COMPILE ERROR\][\s\S]*?\[\/ERROR\]/g, "");
+            if (modelName && modelName.toLowerCase().startsWith("gemma") && aiProvider === "Google" && msg.text.startsWith("[TOOL RESULT]")) {
               const jitInstructionFast = `
 [SYSTEM] Tool result received. Analyze output and proceed with your turn [/SYSTEM]`;
               const jitInstructionThinking = `
 [SYSTEM] Tool result received. Analyze output and proceed with your turn. **STRICTLY MAINTAIN THINKING POLICY. DO NOT START A RESPONSE WITHOUT <think> ... </think>** [/SYSTEM]`;
               msg.text = msg.text.replace(jitInstructionThinking, "").replace(jitInstructionFast, "").trim();
             }
-          });
-        }
+          }
+        });
       } catch (err) {
         const errorMsg = err instanceof Error ? err.message : String(err);
         const date = (/* @__PURE__ */ new Date()).toLocaleString();

package/package.json

@@ -1,64 +1,64 @@
-{
-	"name": "fluxflow-cli",
-	"version": "2.8.5",
-	"date": "2026-06-18",
-	"description": "A high-fidelity agentic terminal assistant for the Flux Era.",
-	"keywords": [
-		"ai",
-		"agent",
-		"terminal",
-		"cli",
-		"gemini",
-		"ink",
-		"flux"
-	],
-	"license": "MIT",
-	"repository": {
-		"type": "git",
-		"url": "git+https://github.com/KushalRoyChowdhury/fluxflow-cli.git"
-	},
-	"engines": {
-		"node": ">=20",
-		"vscode": "^1.90.0"
-	},
-	"files": [
-		"dist",
-		"README.md",
-		"package.json",
-		"ARCHITECTURE.md",
-		"TOOLS.md",
-		"UI_FEATURES.md"
-	],
-	"type": "module",
-	"bin": {
-		"fluxflow-cli": "dist/fluxflow.js",
-		"fluxflow": "dist/fluxflow.js"
-	},
-	"scripts": {
-		"start": "tsx ./src/cli.jsx",
-		"build": "esbuild ./src/cli.jsx --bundle --platform=node --format=esm --outfile=./dist/fluxflow.js --external:react --external:ink --external:chalk --external:fs-extra --external:gradient-string --external:ink-text-input --external:ink-select-input --external:@google/genai --external:zod --external:nanoid --external:puppeteer --external:pdf-lib --external:node-pty --external:html-to-docx --external:typescript --external:ws --external:web-tree-sitter"
-	},
-	"dependencies": {
-		"@google/genai": "^1.52.0",
-		"chalk": "^5.6.2",
-		"fs-extra": "^11.3.4",
-		"gradient-string": "^3.0.0",
-		"html-to-docx": "^1.8.0",
-		"ink": "^7.0.1",
-		"ink-select-input": "^6.2.0",
-		"ink-text-input": "^6.0.0",
-		"nanoid": "^5.1.9",
-		"node-pty": "^1.1.0",
-		"pdf-lib": "^1.17.1",
-		"puppeteer": "24.43.1",
-		"react": "^19.2.5",
-		"web-tree-sitter": "^0.25.10",
-		"ws": "^8.21.0",
-		"zod": "^4.3.6"
-	},
-	"devDependencies": {
-		"@types/react": "^19.2.14",
-		"esbuild": "^0.28.0",
-		"tsx": "^4.21.0"
-	}
-}
+{
+	"name": "fluxflow-cli",
+	"version": "2.8.7",
+	"date": "2026-06-18",
+	"description": "A high-fidelity agentic terminal assistant for the Flux Era.",
+	"keywords": [
+		"ai",
+		"agent",
+		"terminal",
+		"cli",
+		"gemini",
+		"ink",
+		"flux"
+	],
+	"license": "MIT",
+	"repository": {
+		"type": "git",
+		"url": "git+https://github.com/KushalRoyChowdhury/fluxflow-cli.git"
+	},
+	"engines": {
+		"node": ">=20",
+		"vscode": "^1.90.0"
+	},
+	"files": [
+		"dist",
+		"README.md",
+		"package.json",
+		"ARCHITECTURE.md",
+		"TOOLS.md",
+		"UI_FEATURES.md"
+	],
+	"type": "module",
+	"bin": {
+		"fluxflow-cli": "dist/fluxflow.js",
+		"fluxflow": "dist/fluxflow.js"
+	},
+	"scripts": {
+		"start": "tsx ./src/cli.jsx",
+		"build": "esbuild ./src/cli.jsx --bundle --platform=node --format=esm --outfile=./dist/fluxflow.js --external:react --external:ink --external:chalk --external:fs-extra --external:gradient-string --external:ink-text-input --external:ink-select-input --external:@google/genai --external:zod --external:nanoid --external:puppeteer --external:pdf-lib --external:node-pty --external:html-to-docx --external:typescript --external:ws --external:web-tree-sitter"
+	},
+	"dependencies": {
+		"@google/genai": "^1.52.0",
+		"chalk": "^5.6.2",
+		"fs-extra": "^11.3.4",
+		"gradient-string": "^3.0.0",
+		"html-to-docx": "^1.8.0",
+		"ink": "^7.0.1",
+		"ink-select-input": "^6.2.0",
+		"ink-text-input": "^6.0.0",
+		"nanoid": "^5.1.9",
+		"node-pty": "^1.1.0",
+		"pdf-lib": "^1.17.1",
+		"puppeteer": "24.43.1",
+		"react": "^19.2.5",
+		"web-tree-sitter": "^0.25.10",
+		"ws": "^8.21.0",
+		"zod": "^4.3.6"
+	},
+	"devDependencies": {
+		"@types/react": "^19.2.14",
+		"esbuild": "^0.28.0",
+		"tsx": "^4.21.0"
+	}
+}