Korben00/stegano-llm-skill

stegano-llm is a Claude Code skill and CLI implementing Calgacus (Norelli & Bronstein) to hide a secret text inside another text of the same token length using an open-source LLM as an implicit codebook. It’s aimed at researchers, developers and experimenters working on covert communication or LLM steganography who can run Llama‑3‑8B via Ollama. Key features: - Claude Code skill and direct CLI (encode/decode) with example commands. - Setup script installs pyenv, creates a venv and builds llama‑cpp‑python (Metal on Apple Silicon). - Prefix-guided prompting to control style and token ranks. - Deterministic decoding requirements: identical GGUF/model/quantization/seed/hardware; any edit to stegotext breaks decoding. - Limitations: requires ~8B model (Metal/CUDA), not robust to paraphrase/edits, detectable by classifiers, potential rank-overflow errors. Reference: Norelli & Bronstein (2026) and original Calgacus repo.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.