Tacit Knowledge — AI skill for Claude Code

Tacit Knowledge

A cognitive protocol for AI agents. Makes any LLM-based agent think more like an experienced practitioner — commit to judgments, surface hidden signals, and give actionable advice instead of safe platitudes.

Works with **any** AI agent framework: Claude Code, Codex, Gemini, OpenCode, Cline, Cursor, or raw system prompts.

What it does

Default AI output has a fundamental flaw: it optimizes for "least likely to be criticized" instead of "most useful." This produces balanced-but-empty responses, lists without priorities, and advice that sounds smart but doesn't help.

Tacit Knowledge fixes this by injecting a cognitive protocol — a set of thinking rules that change **how** the agent reasons, not what domain it works in.

Before (default agent)

"Microservices and monoliths each have pros and cons. Microservices offer independent deployment, tech stack flexibility, and fault isolation. Monoliths offer simplicity, easier debugging, and simpler deployment. The choice depends on your team size, business complexity, and operational maturity."

After (with Tacit Knowledge)

"Stay on the monolith. With 8 people, the coordination cost of microservices will eat all the gains. If adding a feature touches 3+ modules, the problem isn't the monolith — it's that your module boundaries are wrong. Fix those first. Revisit microservices when you hit 20 engineers or one module needs 10x the traffic of the others."

How it works

The protocol applies four cognitive shifts to every response:

Default Mode	Tacit Knowledge Mode
Balance — list pros/cons, let user decide	Commit — give a judgment with conditions and reversal points
Flat — all points equally weighted	Hierarchy — one focal point, 2-3 supporting signals
External — generic advice from outside	Indwelling — think from inside the user's situation
Reductive — break into parts first	Holistic — diagnose the whole pattern before fixing parts

P