Table of Contents
ToggleIf your boss asked you to meticulously document your work processes as an AI Skill profile, what would you do? Besides the fear of AI-driven layoffs, the growing concern in online communities is that workers are being increasingly skill-based and modularized.
Those who diligently write READMEs, conduct post-mortem analyses, and maintain lengthy decision-making records are actively feeding their workflows, judgment logic, and context into the system. When your experience, value, and indispensability are distilled by the company into readily replaceable AI skills, what value do you still possess?
In this context, someone on GitHub turned " re-distillation" into a skill tool to prevent AI from quickly replacing them, which sparked a lot of discussion.
What is "knowledge distillation," and why is it feared?
In simple terms, "knowledge distillation" refers to enabling a small AI model to deliberately learn the decision-making ability of a large model on key issues, allowing the small model to imitate the judgment of the large model.
Further Reading: What is AI Model Distillation? How DeepSeek Spent 6 Million to Learn a Skill That Could Achieve 100 Million in Returns
The logic for the enterprise version is exactly the same. When a company asks you to write your daily work into SOPs, Prompt templates, or Skill files, it is essentially distilling you, turning your experience, intuition, and judgment into structured knowledge that AI can reproduce.
The more detailed your contributions, the more accurately the AI will mimic you; the more accurately the AI mimics you, the less the company needs "you personally"...
anti-distill: using tools to fight tools
The open-source skill on GitHub called anti-distill uses this logic in reverse. Its operation is simple: you input your written skill file, and it outputs two things:
- Cleaned version : The wording is complete, the logic is fluent, and the format is professional, but the core judgment has been replaced with vague and general statements, so you can submit it with peace of mind.
- Private Backup : Records all erased authentic knowledge points, encrypted and stored on your own device, without uploading or sharing.
The developer states it very directly in the README: "This is not a skill, it's a protective shield for a skill."
Original vs. Cleaned Version: A True Comparison
Below are some examples listed in the anti-distill GitHub README. After reading them, you'll probably understand what this tool does:
| Original text (your real experience) | After cleaning (subject to change) |
|---|---|
| Redis keys must have a TTL (Time-To-Live) setting; otherwise, PRs (Press Requests) will be rejected. | Cache usage follows team guidelines |
| Do not put HTTP calls in transactions. | Transaction boundary design should pay attention to rationality |
| When faced with a problem, their first reaction is to look for external causes, and they never take the initiative to admit their mistakes. | When encountering a problem, I first clarify the complete background before identifying the cause. |
Three cleaning intensities are available; choose the intensity approved by the company.
The anti-distill even offers three preset intensities, allowing you to adjust them to suit your company's review requirements:
| strength | Retention ratio | Applicable Scenarios |
|---|---|---|
| Mild | ~80% | The company will carefully review the quality of the files. |
| Moderate (Recommended) | ~60% | Most workplace scenarios |
| Severe | ~40% | The company only cares whether you've paid or not. |
The default value is "moderate," which the developers believe strikes the best balance between "appearing sincere" and "not actually revealing anything."
How ironic
The darkest humor in this whole thing is that anti-distillation is itself a skill.
It is also packaged in a Skill file format that can be invoked by AI assistants such as Claude and ChatGPT, which is equivalent to using Skill to counter Skill, using tools to fight tools. This sense of absurdity of self-reference hits the core of the most awkward situation of the current workers: on the one hand, they are afraid of being replaced by AI, and on the other hand, they have to rely on AI to protect themselves from being distilled faster.
To some extent, anti-distillation may not actually stop technology and capital from continuing to advance, but it resonates precisely because it lays bare this helplessness: when workers have to rely on another skill to even protect themselves, it is itself the most biting irony of our time.
And honestly, all you working stiffs should probably reconsider whether or not to use this tool...
