overview.txt

================================================================================
                         PromptSync PRO v3.1.1
                     COMPLETE TECHNICAL OVERVIEW
================================================================================

  "Your context. Synced with AI."
  "Less typing, smarter AI, better results."

  PromptSync is a privacy-first Chrome Extension (Manifest V3) that
  automatically injects your personal context into AI chats. It remembers
  your goals, preferences, and important notes, then silently prepends
  them to every message — so you never repeat yourself and always get
  better, more relevant responses.

  Works on: ChatGPT · Claude · Gemini · Perplexity · DeepSeek

================================================================================
  1. CORE CONCEPT
================================================================================

  PROBLEM:  Every new AI chat starts from zero. You type "I'm a developer,
            I use React, keep answers short" hundreds of times.

  SOLUTION: PromptSync stores your profile and context notes locally, then
            auto-injects them into the input field before your message sends.
            The AI sees your context + your question as a single prompt.

  PIPELINE: User Input → Parse Commands → Build Context → Inject → Track Tokens

  PRIVACY:  All data stays in chrome.storage.local. Zero network requests.
            No analytics. No telemetry. Nothing leaves your device. Ever.

================================================================================
  2. FEATURE BREAKDOWN (16 FEATURES)
================================================================================

  [F1] MULTI-PROFILE SYSTEM
       Store multiple profiles (Default, Student, Developer, Custom).
       Each profile has: name, role, goals, skills, preferences, notes.
       Switch profiles from popup or let auto-switch handle it.

  [F2] PROFILE AUTO-SWITCH PER WEBSITE
       Map domains to profiles: claude.ai → Student, chatgpt.com → Developer.
       On page load, detect domain and auto-switch profile.
       Manual mapping only (no AI detection). Toggle on/off.

  [F3] SMART CONTEXT INJECTION
       Builds injection using customizable template with placeholders:
       {user_name}, {user_role}, {user_goals}, {memories}, {instruction}
       Section toggles: enable/disable name, role, goals, memory, instruction.
       Wraps in [USER CONTEXT START]...[USER CONTEXT END] markers.

  [F4] TOKEN-SAVING DEFAULT SYSTEM
       Injects: "Answer as short as possible. No explanation unless asked."
       Facts → direct answer. Best choice → only best option.
       Disabled automatically when task mode is active.
       Per-message override with /explain or /detail commands.

  [F5] COMMAND SYSTEM (BUILT-IN + CUSTOM)
       /research      — Switch to Researcher role
       /dev           — Switch to Developer role
       /design        — Switch to Designer role
       /business      — Switch to Business role
       /writer        — Switch to Writer role
       /no-memory     — skip injection for this message
       /only-profile  — inject only profile (no saved notes)
       /strict        — ultra-minimal (name + pinned only)
       /explain       — allow full explanation
       /detail        — full context + detailed answer
       /code          — coding assistant mode
       /short         — force shortest answer
       /reset         — revert to default
       /temp <text>   — add session-only memory
       + User-defined custom commands

  [F6] CUSTOM COMMANDS
       Users create commands like /fast, /deep.
       Each triggers predefined actions:
       - Mode change (coding/study/short/custom)
       - Enable/disable memory
       - Inject custom instruction text
       No arbitrary code execution. Only preset actions.
       Create/edit/delete from Modes tab.

  [F7] ROLE-BASED ENGINE
       Researcher: "Think deeply, analyze critically..."
       Developer: "Provide clean, optimized, production-ready code."
       Designer: "Focus on modern, clean, user-friendly design."
       Business: "Strategic business expert..."
       Writer: "Write clearly, naturally..."
       Strict injection format: [ROLE] ... [/ROLE] and [STYLE] ... [/STYLE].
       Fail-safe: skips injection if user manually types "act as..." or "you are...".

  [F8] SAVED CONTEXT NOTES
       Add/edit/delete notes with text, tags, priority, pin status.
       Enable/disable individual notes.
       Drag-and-drop reorder. Bulk delete.
       Smart keyword matching: notes relevant to your question auto-selected.
       Priority scoring: high > medium > low. Pinned always included.

  [F9] MEMORY DECAY
       Tracks lastUsed timestamp on every note.
       After configurable days (default: 7), suggests unused notes for removal.
       Decay banner in Saved tab with one-click bulk remove.
       Pinned notes are exempt from decay.

  [F10] SECURITY GUARD
       Scans text for sensitive patterns before saving:
       - API keys: OpenAI (sk-), Stripe (pk_/sk_), AWS (AKIA), GitHub (ghp_)
       - Passwords: password:, passwd=, secret: patterns
       - Credit cards: 13-19 digit Luhn-validated numbers
       - Private keys: -----BEGIN PRIVATE KEY-----
       - Bearer tokens
       Shows warning popup. User can override or cancel save.

  [F11] CUSTOMIZABLE INJECTION TEMPLATE
       Edit the exact format of injected context.
       Placeholders: {user_name}, {user_role}, {user_goals}, {memories}, {instruction}
       Section toggles: individually enable/disable each section.
       Default wraps in [USER CONTEXT START]...[USER CONTEXT END].

  [F12] RESPONSE CONTROL
       Editable instruction appended to every injection.
       Default: "Keep answer short. No extra explanation. Only give what is asked."
       Users can customize to any instruction they want.
       Toggle on/off globally.

  [F13] FLOATING PANEL (TOKEN BADGE + CONTROL PANEL)
       Badge: always visible on AI pages, shows "120 tok" count.
       Click to expand into floating panel:
       - Memory toggle ON/OFF
       - Mode switcher dropdown
       - Token breakdown: input, memory, session
       - Context window progress bar with color coding
       - Session reset button
       Draggable (saves position). Outside click to collapse.
       Settings: show badge only, show panel, hide completely.

  [F14] REAL-TIME TOKEN TRACKING
       Uses o200k_base BPE tokenizer for accurate counting.
       Tracks: input tokens, injected memory tokens, session total.
       Shows percentage of context window used per platform:
       ChatGPT: 128K, Claude: 200K, Gemini: 1M, Perplexity: 16K, DeepSeek: 64K.

  [F15] DARK MODE / THEME
       Light, Dark, System (auto-detect OS preference).
       Cycle with header button. CSS variables for smooth switching.
       Glassmorphism effects on floating panel and toasts.

  [F16] PREVIEW SYSTEM
       Optionally intercept send action.
       Injects context into the textbox but waits for manual user confirmation.
       Allows editing the generated prompt.

  [F17] SANDBOXED CLAUDE COUNTER
       Independent script targeting ONLY claude.ai.
       Complete token awareness without conflicting with PromptSync Core.

================================================================================
  3. ARCHITECTURE
================================================================================

  manifest.json (Manifest V3)
  ├── background/background.js    — Service worker
  │   ├── Context menu ("Save to PromptSync")
  │   ├── Keyboard shortcut (Ctrl+Shift+M)
  │   ├── Default settings, profiles, modes on first install
  │   ├── Content script re-injection on update
  │   └── Memory decay alarm (daily)
  │
  ├── storage/storage.js          — Central data layer
  │   ├── Multi-profile CRUD (create/read/update/delete/switch)
  │   ├── Memory items CRUD with migration
  │   ├── Task modes CRUD
  │   ├── Custom commands CRUD
  │   ├── Settings with deep-merge defaults
  │   ├── Injection template management
  │   ├── Domain → profile auto-switch lookup
  │   ├── Export/import (all data, all formats)
  │   └── Settings cache with TTL for performance
  │
  ├── utils/
  │   ├── tokenizer.js            — BPE token counter (o200k_base + heuristic fallback)
  │   ├── contextEngine.js        — Smart context selection + template engine
  │   │   ├── Keyword extraction (stop word filtered)
  │   │   ├── Relevance scoring (keyword match + priority + pin bonus)
  │   │   ├── Template-based build with placeholder replacement
  │   │   ├── Task mode instruction injection
  │   │   ├── Response control injection
  │   │   ├── Session memory management
  │   │   └── lastUsed timestamp tracking
  │   ├── commandParser.js        — Slash command parser
  │   │   ├── 10 built-in commands
  │   │   ├── Dynamic custom command loading from storage
  │   │   ├── Regex-based extraction without state issues
  │   │   └── Human-readable command list generator
  │   └── securityGuard.js        — Sensitive data scanner
  │       ├── 10 detection patterns (API keys, passwords, CC, etc.)
  │       ├── Luhn checksum for credit card validation
  │       └── Quick-check fast path
  │
  ├── content/
  │   ├── content.js              — Core injection pipeline
  │   │   ├── Platform detection (5 platforms)
  │   │   ├── Input field discovery (multi-selector per platform)
  │   │   ├── Send button interception (click + Enter)
  │   │   ├── Injection pipeline (parse → build → inject → track)
  │   │   ├── Auto-detect user info patterns
  │   │   ├── Text selection save button
  │   │   ├── SPA navigation detection
  │   │   ├── Profile auto-switch per domain
  │   │   ├── Security scanning on save
  │   │   └── Debug mode logging
  │   ├── content.css             — All injected UI styles
  │   ├── floatingPanel.js        — Combined token badge + control panel
  │   └── onboarding.js           — First-run overlay
  │
  ├── popup/
  │   ├── popup.html              — 5-tab dashboard
  │   │   ├── Profile tab (switcher, form, preview, token bar)
  │   │   ├── Saved tab (search, tags, priority, bulk ops, decay)
  │   │   ├── Modes tab (task modes + custom commands)
  │   │   ├── Settings tab (template, response, security, decay, auto-switch, platforms, advanced)
  │   │   └── Help tab (commands, tips, onboarding reset)
  │   ├── popup.js                — All popup logic
  │   └── popup.css               — CSS variables, light/dark, all components
  │
  └── vendor/o200k_base.js        — GPT tokenizer library

================================================================================
  4. STORAGE SCHEMA
================================================================================

  asm_profiles: {
    default:   { name, role, goals, skills, preferences, customNotes },
    student:   { ... },
    developer: { ... },
    custom_1:  { ... }
  }

  asm_items: [
    {
      id:        string (base36 + random),
      text:      string,
      keywords:  string[],
      priority:  "high" | "medium" | "low",
      lastUsed:  timestamp,
      pinned:    boolean,
      enabled:   boolean,
      source:    "manual" | "selection" | "context-menu" | "auto-detect",
      timestamp: number,
      tags:      string[]
    }
  ]

  asm_settings: {
    injectEnabled, compactMode, autoDetect, saveSelection,
    showTokenCount, tokenSavingPrompt, focusMode, tokenIndicator,
    injectionMode, darkMode, activeProfile, activeMode,
    platforms: { chatgpt, claude, gemini, perplexity, deepseek },
    responseControl: { enabled, text, perMessage },
    security: { detectSensitive },
    decay: { enabled, days },
    advanced: { injectionDelay, maxMemories, strictMode, debugMode, maxTokenLimit },
    autoSwitch: { enabled, mappings: { domain → profileName } },
    onboarding: { completed, version },
    floatingPanel: { mode, visible }
  }

  asm_task_modes: {
    coding: { name, instruction, builtin },
    study:  { name, instruction, builtin },
    short:  { name, instruction, builtin },
    custom: { name, instruction, builtin: false }
  }

  asm_custom_commands: [
    {
      name:        "/fast",
      description: "Quick short answer",
      enabled:     true,
      actions: {
        taskMode:       "short",
        disableMemory:  true,
        skipSystemPrompt: false,
        instruction:    "Be extremely brief."
      }
    }
  ]

  asm_template: {
    format:   "[USER CONTEXT START]\nName: {user_name}\n...",
    sections: { name, role, goals, memory, instruction }
  }

================================================================================
  5. SECURITY MODEL
================================================================================

  - ALL data in chrome.storage.local (never leaves device)
  - NO external API calls, analytics, or telemetry
  - HTML-escaped output in all display contexts (XSS prevention)
  - SecurityGuard scans before every save operation
  - Custom commands limited to predefined actions only (no eval/exec)
  - Input sanitization on import (length limits, type validation)
  - Content script namespace isolation (asm- prefix on all DOM elements)

================================================================================
  6. PERFORMANCE
================================================================================

  - Settings cache with 2-second TTL (avoids repeated chrome.storage reads)
  - Token calculation debounced at 300ms
  - Auto-detect pattern matching debounced at 1500ms
  - BPE tokenizer result cache (200 entries LRU)
  - MutationObserver for DOM changes (no polling for input discovery)
  - setInterval polling only for SPA URL changes (800ms) and safety checks
  - Lazy loading: floating panel initializes only when setting enabled

================================================================================
  7. PLATFORM SUPPORT
================================================================================

  Each platform has custom selectors for:
  - Input field (textarea, contenteditable, ProseMirror)
  - Send button (various aria-labels, form submits, icon buttons)

  ChatGPT:    ProseMirror + data-testid send button
  Claude:     ProseMirror + aria-label Send Message
  Gemini:     rich-textarea contenteditable + send-button class
  Perplexity: textarea + Submit aria-label
  DeepSeek:   textarea#chat-input + div role=button

  SPA navigation handled via URL polling + popstate listener.
  Content scripts re-injected on extension update.

================================================================================
  PromptSync PRO v3.1.1 — Stop repeating. Start syncing.
================================================================================