ghostfont-cracklisted

# Ghost Font Cracker Recovers text hidden in an anti-motion random-dot clip via dense optical-flow accumulation, then reads the reconstructed still (Tesseract + local vision model). Self-contained, offline (except URL fetch / vision model on localhost). ## When to use - User points at a Ghost-Font-style clip (file / URL / on-screen) and wants the text. - Defensive research, CTF, or recovering your own known phrase. ## Input modes (auto-detected; force with `--mode`) | mode | trigger | how | |------|---------|-----| | `file` | local path ending in a video ext | `cv2.VideoCapture` | | `url` | `http(s)://…` | ffmpeg direct copy → yt-dlp fallback | | `live` | no source given, or `--mode live` | screen grab (mss → ffmpeg avfoundation) | ## Run ```bash # activate the bundled deps (opencv, pytesseract, mss, yt-dlp) — reuse the poc venv source ~/ghostfont-poc/.venv/bin/activate python ~/.claude/skills/ghostfont-crack/crack.py ghost.mp4 python ~/.claude/skills/ghostfont-crack/crack.py "https://site/ghost.mp4" python ~/.claude/skills/ghostfont-crack/crack.py --mode live --seconds 3 --region 400,300,900,300 ``` Options: `--axis x|auto|both` (motion axis; `auto`=PCA on flow, handles non-horizontal), `--winsize N` (fix Farneback window, else auto-scan 7/9/13), `--model <ollama vlm>` (`''` to skip vision), `--save PREFIX` (writes `PREFIX_recon.png`). ## Output ``` Tesseract : 'CATCN WE' Vision : 'CATCH ME' ``` Plus `PREFIX_recon.png` — the reconstructed text image (human-verifiab