Creative Content Automation v7 (CCA v7)

Notion → ChatGPT → Gemini → Notion image pipeline. v7 adds humanlike mouse + typing for Gemini (ghost-cursor + 0–50ms randomized typing + 2s cooldown + rotating mouse-movement profiles), keeping ChatGPT typing fast.

FETCH ─→ REFINE ─→ PROMPTS ─→ IMAGES ─→ UPLOAD
Notion    ChatGPT    ChatGPT    Gemini     Notion (zip + refined MD)
                       │            │
                  fast typing   ghost-cursor Bezier mouse +
                                randomized 0-50ms typing per char +
                                2s cooldown between submissions

What's new vs v6

• ghost-cursor integrated in scripts/submit_prompts.cjs — mouse moves through Bezier paths with overshoots and micro-corrections instead of teleporting to click coordinates.
• humanType() in submit_prompts.cjs types each char with Math.random() * 50 ms delay (0–50ms uniform). Spec: "typing speed should be randomized from the range of 0 to 50".
• SUBMIT_COOLDOWN_MS = 2000 between successive prompt submissions.
• 4 rotating movement profiles in tools/browser/gemini.py (calm / energetic / deliberate / casual). One picked per Python session at module-import time. Each profile changes Bezier curve swing, move duration, step variance, click jitter, and click delay — so movement signature varies across sessions.
• ChatGPT typing stays fast (tools/browser/chatgpt.py:276 uses delay=1). Per spec: "in chatgpt you can type as fast as you can".

Carry-over from v6

• Multi-account accounts.json with auto-rotation on Gemini 1095 / quota
• auto_login.py Python wrapper signing in to ChatGPT + Gemini at startup
• tools/accounts.py rotator with persistent state in .cca/active_accounts.json
• scripts/save_images.cjs detectBlocker() recognizing 1095 + quota error UI
• scripts/run_autonomous.cjs triggerRotation() cycle: kill children → rotate accounts.json → sign-out → sign-in next account → respawn

Quick start

:: 1. Extract / clone
:: 2. setup.bat (one-time — installs deps + launches the two Chrome windows)
:: 3. Sign in to ChatGPT + Gemini (or fill accounts.json for auto-login)
:: 4. Edit lessons.txt with chapters
:: 5. start.bat

See RUN.md for the full runbook.

Trade-off

Per-prompt overhead increases by ~10s vs v6:

• humanType on a 100-char prompt: ~2.5s
• 2 cursor.click() Bezier moves: ~3-4s
• 2s cooldown: 2s

For an 80-prompt chapter, that's roughly +13 min wall-clock. Trade-off accepted for human-like behavior.

License

MIT.