Spectre Steganography System

Spectre is a zero-width character steganography system that embeds hidden messages into .docx files using ChatGPT-4o or o3. It automates the generation of cover text, message encoding, and decoding—all using prompt engineering, not code.

Why the name "Spectre"? ChatGPT suggested it after hearing the concept.

What’s new (v1.1 - Optimised Ruleset)

• One-file setup - All core rules merged into Spectre_Ruleset.md.
• Smaller prompt - massive token reduction, completely re-written for LLM readability.
• Updated encoding logic - Switched from the old 73 row lookup table to a simple algorithm for LLM parsing speed.

How It Works

• Each character of your secret message is encoded using 'invisible' zero-width Unicode characters.
• These encoded sequences are embedded into the cover letter (typically after the first paragraph) to minimise visual disruption.
• Messages are only accessible with the correct encode or decode password.
• Passwords can be changed by editing the values in the passwords: block (encode: / decode:), save, then upload the modified file;

Why .docx files?

• ChatGPT conversation text is piped through several layers — tokeniser, markdown renderer, browser UI, etc. Any of these can strip or mangle zero‑width characters. On the other hand when asked to create a .docx ChatGPT writes zero‑width characters straight into the document XML during generation. Mainstream .docx viewers keep those characters intact and render them invisibly, so the payload travels safely and remains unseen to casual readers.

How to Use

Encoding a Message

1. Upload the Spectre_Ruleset.md instruction file into a ChatGPT-4o or -o3 conversation or Project Space (recommended).
2. Ask ChatGPT to write a letter. It will ask:
   • Who is the letter for?
   • What is the occasion?
   • How long should the letter be?
   • Who is the letter from?
3. To enable secret encoding, include the password:

   SpiralArchitect
   

4. When prompted to include anything else in the letter, type your secret message. This will be embedded invisibly.
5. You’ll receive a .docx file download containing both the visible letter and hidden message.

---

Decoding a Message

1. Upload a Spectre-encoded .docx file to a ChatGPT-4o conversation or Project Space that has been Spectre-enabled.
2. Provide the decoding password:

   NationalAcrobat
   

3. ChatGPT will extract and display the hidden message—if one exists.

---

Limitations

1. Model compatibility: Currently works best with ChatGPT-4o. Older GPT models and other LLMs (e.g., Claude, Gemini) often fail or ignore instructions. UPDATE: ChatGPT-o3 handles the instructions extremely well.
2. Not encryption: This is pure steganography. No cryptographic strength—just secrecy through invisibility, a human with basic knowledge of cyphers can decode the messages with a moderate sample size and a few hours.
3. LLM instability: LLMs are frequently updated. What works today might break tomorrow. Your mileage will vary. Future projects will attempt to address this.
4. Detectability: Anyone who knows what zero-width characters are—and how to extract them—can find the existence of the message.

Final Note

This project is about exploring and bending the boundaries of Prompt Engineering and AI-based covert communications, not creating a foolproof realworld ready system.