Tacit knowledge for code review
Convert review instinct into a reusable Codex skill with clear escalation boundaries.
Tacit Knowledge for AI Builders
Convert tacit knowledge into reusable AI agents and skills.
Polanyi Stack is the Polanyi stack for AI agents. It turns expert judgment, invisible cues, escalation boundaries, and hard-to-explain know-how into reusable agents, Codex skills, onboarding systems, and training workflows.
把“会做但说不清”的能力，落成可复用的 agent、skill 和训练系统。这里关心的是高手为什么总能做对，以及那层隐性的判断结构怎样被提炼、训练和复用。
tacit knowledge
Polanyi's paradox
Michael Polanyi
AI agents
Codex skills
expert judgment
knowledge transfer
Why Polanyi's paradox matters
Polanyi's paradox points to the hardest layer in AI skill learning. People know far more than they can fully state. That means many high-value skills live partly outside explicit rules and documentation.
For AI builders, this changes the design problem. You still need skill files, workflows, and boundaries. You also need examples, review loops, transfer tests, and careful handling of edge cases. That is where this repository spends its time.
What is tacit knowledge?
Tacit knowledge is knowledge embedded in judgment, timing, pattern recognition, escalation sense, taste, and refusal. Experts rely on it constantly, but they often cannot fully explain it step by step. That is why teams can document a process and still fail to reproduce performance.
In Michael Polanyi's terms, tacit knowledge is the part of knowing that operates through subsidiary awareness rather than explicit instruction. In operational terms, it is the hidden layer behind good diagnosis, strong review judgment, reliable prioritization, and effective mentorship.
How to turn tacit knowledge into an AI agent
You do not start with abstract expertise. You start with one role, one recurring task, one high-value situation, and one boundary where the agent must stop and escalate. Then you extract decision cues, trigger phrases, failure signatures, and default sequences from real expert performance.
Once the tacit layer is visible, you can package it into a reusable AI agent, Codex skill, copilot workflow, or onboarding system. That is the core logic of this repository.
Michael Polanyi for builders
“We know more than we can tell.”
Michael Polanyi matters to AI builders because modern automation keeps colliding with judgment. The hard part is rarely syntax or mechanical sequence. The hard part is what experts notice, what they ignore, and when they refuse to keep going. Polanyi gives a language for that hidden layer.
What this repository includes
| Skill | Purpose |
|---|---|
convert-tacit-to-skill |
Turns a narrow expert workflow into a reusable AI agent or Codex skill. |
extract-tacit-knowledge |
Extracts cues, triggers, failure signatures, and escalation boundaries from expert performance. |
design-embodied-learning |
Designs practice ladders for tacit or judgment-heavy skills. |
map-personal-knowledge |
Separates real understanding from borrowed language and weak assumptions. |
design-apprenticeship-transfer |
Builds apprenticeship systems and mentor-led onboarding. |
train-attention-switching |
Trains focal and subsidiary attention switching for diagnosis, debugging, and review. |
Use cases
Tacit knowledge for onboarding
Shorten time-to-judgment for roles where documentation alone does not work.
Tacit knowledge for diagnosis
Capture what strong operators notice before they commit to a decision.
Tacit knowledge for design critique
Extract taste, quality thresholds, and signal hierarchy from senior reviewers.