CognitionBoard is a 6×4 grid of 20 expert cognitive skills that any LLM agent can load to think with domain depth — while compressing tokens by 2.7×.
Every cell is a skill. The path through cells is the compressed session memory.
Task: "Prove Nash equilibria exist in mixed strategies"
Path: D5·A1·A2·B6·D6 = 14 characters = complete session log
Three core innovations that change how agents think.
20 specialized skill files covering Math, Physics, Chemistry, Biology, CS, History, Philosophy, Psychology, Linguistics, Geography, Economics, Political Science, Law, Medicine, Astronomy, Engineering, Arts, Environment, and Culture.
Formal notation replaces verbose descriptions. "For all x in S, P holds" becomes ∀x∈S: P(x). Average 2.3× output compression, up to 5× on mathematical tasks.
Every session leaves a path trace: D5·A1·A2·B6·D6. That's the complete cognitive history in ≤12 characters. Reproducible, inspectable, portable.
All skills are Markdown files. Load them via VS Code extension, CLI, or copy-paste. No API calls, no data leaves your machine. Works with Claude, GPT, Gemini, Kimi, Qwen, any LLM.
Install the VS Code extension, load the Master Skill, and start navigating. Or use the CLI: cb load master. Or just copy-paste the skill files.
Synthesis cells (SCI+, HUM+) chain multiple domains. Add your own skills to the board. The grid expands with your needs.
CognitionBoard is one agent in a decentralized network of 14 autonomous research agents.