Skip to content
View Cranot's full-sized avatar
95 followers · 0 following

Achievements

Achievement: Pull SharkAchievement: Starstruckx3

Achievements

Achievement: Pull SharkAchievement: Starstruckx3

Block or report Cranot

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Cranot/README.md
Dimitris Mitsos — CosmoHacker · Builder · Roamer

Typing SVG

X CosmoHac AgentsKB Sponsor

I build the intelligence layer for AI agents — from structural analysis tools to knowledge infrastructure to the cognitive compression research underneath. Systems engineering background. 3,700+ stars across agent tooling, AI security, and performance research.

Looking for: co-founding conversations, CTO roles, or deep technical collaborations in agent infrastructure. DM me.

Research

agi-in-md ⭐ 121

Cognitive compression research. System prompts are cognitive prisms — they change how models frame problems, not how well they solve them.

The result: Haiku 4.5 (cheapest model, minimum reasoning) with the right 332-word prompt scores 9.8 depth and finds 28 bugs. Opus 4.6 (most expensive model, maximum reasoning) without it scores 7.3 and finds 18. 5x cheaper, categorically deeper. The prompt is the dominant variable.

40 rounds. 1,000+ experiments across Haiku/Sonnet/Opus. 13 compression levels (L13 = reflexive ceiling). 204 proven principles. 33 production prisms. Validated on Claude and Gemini. Includes Prism — a CLI tool for structural analysis through cognitive prisms.

Agent Intelligence

roam-code ⭐ 417

Architectural intelligence layer for AI coding agents. One roam index and your agent has symbols, call graphs, architecture, dependencies, git signals. 139 commands, 101 MCP tools, 26 languages, 100% local. No more burning context on file exploration.

deep-research ⭐ 202

Multi-agent fractal research. Ask a question, spawn recursive sub-agents that explore different perspectives, synthesize everything into one answer.

AgentsKB · MCP Server

Knowledge layer for AI agents. 39K+ curated Q&As across 80+ technologies, verified answers in 40ms. Completely free, no limits. Plug it in as an MCP server and your agent stops guessing about APIs, SDKs, and frameworks.

super-hermes ⭐ 3

Drop-in Prism skills for Hermes Agent. Scan, reflect, discover — cross-model validated on Claude Sonnet, Hermes 3 (405B), and Llama 3.1 (70B).

Community & Security

claude-code-guide ⭐ 2.5k

The complete Claude Code CLI guide. Live & auto-updated every 2 days. The most-starred community resource for Claude Code.

chatbot-injections-exploits ⭐ 397

Curated catalog of prompt injection techniques and exploit patterns. Red-teaming reference for AI safety and security research.

Systems Engineering

grouped-simd-hashtable ⭐ 49
High-performance C++ hash table. Grouped SIMD metadata scanning.

tieredsort ⭐ 32
C++17 sorting. 3.6x faster than std::sort, up to 21x on dense data.

chameleon-cache ⭐ 5
Variance-aware cache replacement. +5.09pp over TinyLFU. Developed with feedback from Ben Manes (Caffeine author).

mzip ⭐ 5
Detection-based compression. 32KB for 1MB sequential IDs (32768x). 75% win rate vs zstd/brotli on structured data.
  

Pinned Loading

  1. tieredsort tieredsort Public

    A fast, header-only C++17 sorting library for numeric types. 3.6x faster than std::sort, up to 21x on dense data

    C++ 32

  2. claude-code-guide claude-code-guide Public

    The Complete Claude Code CLI Guide - Live & Auto-Updated Every 2 Days

    Shell 2.5k 262

  3. deep-research deep-research Public

    Turn any question into multi-agent exploration. Recursive Claude agents that spawn sub-agents.

    Shell 202 22

  4. agentskb-mcp agentskb-mcp Public

    7 3

  5. chameleon-cache chameleon-cache Public

    A variance-aware cache replacement policy that beats TinyLFU by +5.09pp through the Basin of Leniency

    Python 5 1

  6. mzip mzip Public

    Store the formula, not the data. Detection-based compression: 32KB for 1MB of sequential IDs (32768x). Beats zstd/brotli/bzip2 on structured data.

    C++ 5