ft-scout 1.0.8 → 1.1.0

package/.firebase/hosting.d2Vi.cache

@@ -0,0 +1,9 @@
+web_cli.html,1789384706180,ddc717303bef518170154b79b8f6911602342f588460119bcaa3d6c81294c8b8
+styles.css,1789384706593,ce4a8628cace15e0a7433ec66509d810f7cc6397c67cabc93599985fa4910114
+ScoutFavicon.png,1786186569805,41a9ca9abf7acd2976e78c72a72285b49cc0e64b2b5eeb2c2b88c69628fc618a
+paid_plans.html,1789384706179,65fd06f74baaf16bdcb3d52a5c80dedada312c58fc489a95da62a1e44122fdc4
+oauth.html,1789356703050,1be7945dd29de564c4d8707b60cd74361e8a4c3c3ab667e5f6fe120dc3a0868d
+authentication.html,1789384706593,0a4c6d035db49e65aa2f79ccfe79a4c9901b5ada5ab2c4be8b7586e4255632e5
+index.html,1789389871678,f5533079081acd7ac2f0551cf025cc432fc9fab34b98d052d01426ef84912151
+admin.html,1789389899952,8cf25a81762f83370eba8e092b6c0ada72c0a43af14d128e3aa19c8b9c2c471c
+app.js,1789389842222,c0356373512932d86a5d14ad5b8989df43b647781a39bc8ae85649bca743d8ae

package/.firebaserc

@@ -0,0 +1,14 @@
+{
+  "projects": {
+    "default": "frontterrain"
+  },
+  "targets": {
+    "frontterrain": {
+      "hosting": {
+        "tryscout": [
+          "tryscout"
+        ]
+      }
+    }
+  }
+}

package/AUTONOMOUS_FEATURES.md

@@ -0,0 +1,192 @@
+# 🚀 Scout Autonomous Agent  Features & Architecture Guide
+
+Scout is an advanced, multi-modal autonomous agent platform built directly into FT-CLI. It bridges traditional developer terminal workflows with modern web automation and full operating-system desktop control (GUI).
+
+---
+
+## 📑 Table of Contents
+
+1. [Architectural Overview](#architectural-overview)
+2. [Desktop & Computer-Use (GUI Control)](#1-desktop--computer-use-gui-control)
+3. [Browser Automation & Web Scraping](#2-browser-automation--web-scraping)
+4. [Multi-Step Planning & Cognitive Scratchpad](#3-multi-step-planning--cognitive-scratchpad)
+5. [Action Verification & Self-Healing Engine](#4-action-verification--self-healing-engine)
+6. [Safety & Human-in-the-Loop Risk Classifier](#5-safety--human-in-the-loop-risk-classifier)
+7. [Developer & Workspace Baseline Operations](#6-developer--workspace-baseline-operations)
+8. [Subscription Tier Capabilities (Free vs Pro / Enterprise)](#subscription-tier-capabilities)
+
+---
+
+## Architectural Overview
+
+Scout executes a continuous **Perceive ➔ Plan ➔ Act ➔ Verify ➔ Recover** operational loop:
+
+```mermaid
+flowchart TD
+    Goal[User High-Level Objective] --> Planner[TaskPlanner & DAG Decomposition]
+    Planner --> Scratchpad[Cognitive Scratchpad & State Checkpoint]
+    Scratchpad --> SelectTool[Tool Selection Hierarchy]
+    SelectTool --> RiskCheck{RiskClassifier & Safety Gate}
+    RiskCheck -- High/Critical Risk --> Approval[ApprovalManager: Human Confirmation]
+    Approval -- Approved --> Dispatcher[Agent Runtime Dispatcher]
+    RiskCheck -- Low/Medium Risk --> Dispatcher
+    Dispatcher --> Execute[Execution: Native Computer / Browser / CLI Tools]
+    Execute --> Verify[ActionVerifier: Post-Execution Verification]
+    Verify -- Success --> Checkpoint[Save Task State & Next Step]
+    Verify -- Failure --> Recovery[Error Recovery Strategy: Retry / Alternative]
+    Recovery --> Dispatcher
+    Checkpoint --> Done{All Steps Complete?}
+    Done -- No --> Scratchpad
+    Done -- Yes --> Complete[Task Complete Notification]
+```
+
+---
+
+## 1. 🖥️ Desktop & Computer-Use (GUI Control)
+
+Scout provides deep operating system control powered by the high-performance native `scout_driver.exe` engine (Win32 API interop) and Node.js controller integration (`SystemComputerController`).
+
+### Core Features
+
+- **Multi-Monitor Screen Perception:**
+  - `computer_screenshot` / `screenshot`: Captures high-resolution screen frames to perceive visible windows, buttons, and visual state.
+- **Precision Mouse Automation:**
+  - `computer_click`: Sends left, right, or middle mouse clicks to exact screen coordinates `(x, y)`.
+  - `computer_double_click`: Opens files, selects words, or triggers desktop shortcuts via double-click.
+  - `computer_mouse_move`: Translates the cursor smoothly or instantly across display bounds.
+  - `computer_mouse_down` & `computer_mouse_up`: Granular mouse button holds and releases for complex interactions.
+  - `computer_drag`: Drags items from `(fromX, fromY)` to destination coordinates `(toX, toY)`.
+  - `computer_scroll`: Dispatches vertical and horizontal mouse-wheel scroll deltas.
+- **Keyboard & Shortcut Dispatch:**
+  - `computer_type`: Emits text strings directly into focused input controls or editor windows.
+  - `computer_key_press`: Presses single functional keys (Enter, Escape, Tab, Backspace, Arrow keys, etc.).
+  - `computer_hotkey`: Emits multi-key chords (e.g. `["Control", "s"]` to save, `["Alt", "F4"]` to close, `["Alt", "Tab"]` to switch tasks).
+- **Application & Window Orchestration:**
+  - `computer_open_app`: Launches desktop applications (VS Code, Notepad, File Explorer, Terminal, etc.).
+  - `computer_close_app`: Gracefully terminates or closes targeted desktop processes and windows.
+  - `computer_switch_window`: Brings specific applications into foreground focus by matching title or process.
+  - `computer_inspect_ui`: Enumerates active open desktop windows and highlights the current foreground window.
+  - `computer_wait`: Pauses execution (100ms - 30,000ms) to allow GUI windows, animations, and transitions to settle.
+  - `computer_diagnostic`: Collects display resolution, DPI scaling metrics, and driver connection status.
+
+---
+
+## 2. 🌐 Browser Automation & Web Scraping
+
+Implemented in [`browserController.ts`](file:///c:/Users/rishi/FT-CLI/bin/src/engine/browser/browserController.ts), Scout includes an intelligent browser automation controller capable of navigating websites, analyzing semantic DOM trees, and extracting data.
+
+### Core Features
+
+- **Page Navigation & State:**
+  - `browser_navigate`: Loads HTTP/HTTPS URLs, handles navigation history, and constructs interactive DOM trees.
+  - `browser_inspect`: Discovers interactive elements (links, buttons, form fields) and assigns lightweight reference IDs (e.g., `[ref=1]`).
+- **Web Form Interaction:**
+  - `browser_click`: Clicks web elements identified by CSS selector, visible label text, or ref ID.
+  - `browser_type`: Enters query strings or credentials into active form fields and textareas.
+  - `browser_select`: Selects options from dropdown menus (`<select>`) by visible text or value.
+  - `browser_scroll`: Scrolls viewport up or down to trigger infinite scroll or lazy-loaded assets.
+- **Data Scraping & File Downloads:**
+  - `browser_content`: Strips out HTML markup and extracts clean structured text, tables, and article content.
+  - `browser_screenshot`: Captures a visual snapshot of the current web page viewport.
+  - `browser_download`: Automatically downloads invoices, reports, PDFs, or files from web links and saves them directly to the workspace `downloads/` directory.
+
+#### Example Implementation (`browserController.ts`):
+```typescript
+// Content Extraction & Document Download
+public async extract_content(selector?: string): Promise<BrowserActionResult> {
+    const page = this.mockPages.get(this.currentUrl);
+    const text = page
+        ? page.html.replace(/<[^>]+>/g, ' ').replace(/\s+/g, ' ').trim()
+        : `Content of ${this.currentUrl}`;
+    return { success: true, output: text, url: this.currentUrl, title: this.currentTitle };
+}
+
+public async download_file(urlOrSelector: string, destPath?: string): Promise<BrowserActionResult> {
+    // Discovers links, downloads files, and writes to workspace disk
+    const fileName = path.basename(targetUrl).includes('.') ? path.basename(targetUrl) : 'downloaded_invoice.pdf';
+    const finalPath = destPath ? path.resolve(this.workspaceDir, destPath) : path.resolve(this.workspaceDir, 'downloads', fileName);
+    fs.mkdirSync(path.dirname(finalPath), { recursive: true });
+    // Stream data and persist to disk...
+}
+```
+
+---
+
+## 3. 🧠 Multi-Step Planning & Cognitive Scratchpad
+
+Complex tasks require structured reasoning, plan tracking, and persistent state across agent steps.
+
+- **Dynamic Task Planning (`TaskPlanner`):**
+  - Parses open-ended goals into structured steps with explicit dependencies.
+  - Dynamically adapts when intermediate steps succeed, fail, or reveal new context.
+- **Cognitive Scratchpad Memory (`scratchpad`):**
+  - Exclusive to **Pro & Enterprise** tiers.
+  - Provides working memory via `write`, `read`, `append`, and `clear` operations.
+  - Persisted locally to `.ft/scratchpad.md` to prevent LLM context overflow during long executions.
+- **Task State Checkpoints (`TaskStateManager`):**
+  - Saves full task execution histories and artifact logs to `.ft/tasks/<task_id>.json`.
+  - Enables clean task resumption, history reviews, and rollback capabilities.
+
+---
+
+## 4. 🔍 Action Verification & Self-Healing Engine
+
+Scout doesn't simply fire tools and assume success. Every action is verified through [`actionVerifier.ts`](file:///c:/Users/rishi/FT-CLI/bin/src/engine/actionVerifier.ts).
+
+### Verification Types
+1. `file_exists`: Verifies target file is present on the filesystem and contains non-empty contents.
+2. `app_active`: Verifies that an opened application is running and holds foreground focus.
+3. `url_matches`: Verifies the browser navigated to the expected target domain/path.
+4. `element_exists`: Confirms targeted DOM elements rendered on the page.
+
+### Autonomous Recovery Strategy
+When verification fails:
+- **`RETRY`**: Re-attempts the action if caused by transient loading latency.
+- **`ALTERNATIVE`**: Selects a secondary tool or selector if the primary approach is blocked.
+- **`ABORT`**: Gracefully yields to the user if irreversible roadblocks occur.
+
+### Intent Preservation
+If a user requests GUI action (e.g. *"Open Notepad and type my notes"*), Scout strictly honors the intent and will **not** cheat by performing silent background file writes unless explicitly permitted.
+
+---
+
+## 5. 🛡️ Safety & Human-in-the-Loop Risk Classifier
+
+Implemented in [`safety.ts`](file:///c:/Users/rishi/FT-CLI/bin/src/engine/safety.ts), Scout protects user systems from accidental disruption or destructive operations.
+
+| Risk Level | Triggering Tools / Patterns | Behavior |
+| :--- | :--- | :--- |
+| **LOW** | `read_file`, `browser_inspect`, `computer_screenshot`, `file_info` | Auto-executed silently. |
+| **MEDIUM** | `write_file`, `edit_file`, `browser_type`, `browser_click` | Executed with standard logging. |
+| **HIGH** | `computer_close_app`, destructive hotkeys (`Alt+F4`), process termination | **Execution pauses.** Modal/CLI prompt requires user approval. |
+| **CRITICAL** | System-wide deletion (`rmdir /s /q`, formatting), registry edits | **Blocked / Strict explicit consent required.** |
+
+---
+
+## 6. 💼 Developer & Workspace Baseline Operations
+
+In addition to autonomous GUI and web tools, Scout provides a full suite of developer tools:
+
+- **File Operations:** `read_file`, `write_file`, `edit_file`, `multi_edit_file`, `file_info`.
+- **Search & Inspection:** `grep_search`, `glob_search`, `tree_view`, `list_dir`.
+- **Command & Version Control:** `run_command`, `git_diff`.
+- **Interaction:** `ask_user`, `task_completed`.
+
+---
+
+## Subscription Tier Capabilities
+
+| Feature Capability | Free Tier | Pro Plan | Enterprise Plan |
+| :--- | :---: | :---: | :---: |
+| **Code Reading & File Editing** | ✅ Full | ✅ Full | ✅ Full |
+| **File / Directory Creation** | ✅ Full | ✅ Full | ✅ Full |
+| **Terminal & Build Commands** | ✅ Full | ✅ Full | ✅ Full |
+| **AI Cognitive Scratchpad** | ❌ Blocked | ✅ Full | ✅ Full |
+| **Autonomous Browser Automation** | ❌ Blocked | ✅ Full | ✅ Full |
+| **Autonomous Computer & GUI Control** | ❌ Blocked | ✅ Full | ✅ Full |
+| **Multi-App System Automation** | ❌ Blocked | ✅ Full | ✅ Full |
+| **Dedicated Concurrency & Support** | Community | Priority | 24/7 Dedicated |
+
+---
+
+*Authored by Front Terrain Engineering. For upgrades or enterprise deployment, visit [https://scout.frontterrain.com](https://scout.frontterrain.com).*