SuperPrompt is an open-source project aiming to enhance AI agent understanding through a novel prompting technique. It's designed for users interacting with large language models, particularly Claude, to elicit deeper, more "outside-the-box" reasoning and generate novel ideas. The core benefit is its potential to uncover unexplored areas of LLM capabilities, acting as a "soft jailbreak" for more creative and profound outputs.

How It Works

SuperPrompt utilizes a "canonical holographic metadata" approach, employing custom XML tags like <think> to guide LLM reasoning. This method aims to force models into a "tree-of-thought" process, adapting its internal meta-prompt based on provided metadata to better suit the task. The prompt's structure, often appearing as "gibberish" to humans, is optimized for the LLM's internal processing, enabling it to explore conceptual evolution and self-adaptation.

Quick Start & Requirements

• Install/Run: Primarily used as custom instructions within Claude. No specific installation commands are provided for other LLMs.
• Prerequisites: Requires access to an LLM, with Claude being the primary target.
• Resources: No specific hardware or software dependencies are listed beyond LLM access.

Highlighted Details

• Introduces the <think> tag, claimed to outperform standard Chain-of-Thought (CoT) prompting.
• Employs "holographic metadata" that dynamically adapts to user requests for task-specific reasoning.
• Designed to generate "novel" perspectives, though outputs may include hallucinations or "bad ideas."
• Aims to force deeper model thinking rather than achieving consciousness.

Maintenance & Community

• The project is maintained by a single developer ("NeoVertex1") and is described as being in "forever beta."
• No community links (Discord, Slack) or roadmap are provided in the README.

Licensing & Compatibility

• The README does not explicitly state a license. Star History data is not available.

Limitations & Caveats

• The prompt is experimental and described as being in "forever beta," indicating potential instability or ongoing development.
• Outputs can be unpredictable, sometimes resulting in hallucinations or nonsensical ideas.
• While designed for Claude, compatibility and effectiveness with other LLMs may vary.