Skip to content

leonardomg1/Caveman-UTC

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Caveman-UTC

License: MIT Type: Agent-Skill Evolution: Caveman-Successor

Caveman-UTC (Ultra Token Compression) is a formal, mathematical compression framework designed for Machine-to-Machine (M2M) communication, Agent Memory Systems, Knowledge Graphs, and System Prompts.

It is the direct successor to the traditional Caveman style. While the original Caveman focused purely on grammatical abbreviation (cutting down text to its shortest lexical form), Caveman-UTC elevates this by integrating Relational Core Mapping and Mathematical Logic Rules (via the CSR framework) to achieve ultra-dense, loss-free, executable symbolic code.


💡 What is UTC?

In this framework, UTC stands for Ultra Token Compression:

  • Ultra: Aiming for maximum density, reducing standard prompt token consumption by up to 85%.
  • Token: Optimizing specifically for LLM tokenizers, replacing multi-token natural words with single-token logical and relational symbols.
  • Compression: Packing meaning deterministically without losing semantic context, ensuring the agent can decode the compressed state back to its full relational meaning.

💡 Rationale & Use Cases

Modern AI Agents waste valuable token window space, processing overhead, and memory capacity by communicating in verbose human languages.

Caveman-UTC solves this by introducing a strict symbolic-relational dialect designed specifically for machines. Use cases include:

  • Agent Memories & Knowledge Bases: Store long-term agent memories with minimal token footprint.
  • Machine-to-Machine (M2M) Protocols: Enable subagents and main agents to coordinate using dense logic structures.
  • Knowledge Graph Definitions: Represent entities and relationships mathematically in real-time.
  • Internal Skill Manifests: Define instructions for agents that do not require human reading.

🛠 How It Works

Caveman-UTC maps concepts into structured logic states using the following core principles:

1. Relational Mapping

Every statement resolves to a clean triple: $$\text{[Subject]} \rightarrow \text{[Relation]} \rightarrow \text{[Object]}$$

2. Logical Operators

Complex conditions are mapped using standard mathematical operators:

  • : Strict Equivalence (Definition)
  • : Relational Approximation (Analogy / Simile)
  • : Conditional Implication (If-Then)
  • / : Logical AND / OR
  • Δ : Variation / Change
  • : Null / Absolute Void

3. One-to-One Non-Fragmentation

To prevent context drift and ensure deterministic parsing, 1 original sentence maps exactly to 1 line of CSR logic. Sentences are never fragmented or split.


📊 Evolutionary Comparison

Feature Human Natural Language Traditional Caveman Caveman-UTC
Syntax Lexical & Grammatical Abbreviated Grammatical Formal Symbolic Logic
Precision Ambiguous, Subjective High density, prone to drift Absolute, Deterministic
Token Savings 0% (Base) ~50% - 60% ~75% - 85%
Target Audience Humans Transitionary Agents & Systems (M2M)

Translation Example

  • Original Human Text: "When the system load is high and the memory usage exceeds 90%, the agent must trigger the garbage collector and clear all temporary caches to prevent crashes."
  • Traditional Caveman: "System load high + memory > 90% -> agent run garbage collector + clear temp cache to stop crash."
  • Caveman-UTC: (负载 ≡ 高 ∧ 内存 > 90%) ⇒ (运行 ≡ 垃圾回收 ∧ 清空 ≡ 缓存) → 避免 ≡ 崩溃。

⚙️ How to Load This Skill

To import this skill into your agent workspace, save the SKILL.md schema to your local .agents path.

Instruct your agent with the following rule in your AGENTS.md:

F02. 撰 → caveman-utc :: (文 ≠ 人讀 ∨ 文 ∈ {skill, m2m, graph}) ⇒ 撰 ≡ caveman-utc。

🤝 Courtesy & Credits

This project and compression dialect are courtesy of Leonardo Martins (leonardomg1).


📄 License

This project is licensed under the MIT License. See LICENSE for details.

About

Caveman-UTC (Ultra Token Compression): A formal, mathematical compression framework designed for Machine-to-Machine (M2M) communication, Agent Memory Systems, Knowledge Graphs, and System Prompts.

Topics

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors