Today I'm sharing something that's great for turning methodologies into Skills.
Usually, after finishing a book or watching a video, I always feel like I've learned something, but I can't fully apply it to my own projects, wasting a lot of inspiration.
Today I want to recommend a GitHub project called Cangjie-skill, which can distill the books, videos, and blogs you've read into Skills, so you can invoke this knowledge in real scenarios.
You can take a look—this is the effect of distilling a lighting effects tutorial video and applying it to an image generation project:
01. Project Hands-On Test
case 1 Book Distillation Skill
We directly invoke this skill to distill:
Prompt:
Please invoke cangjie-skill to do a single Skill distillation on the book I provide.
Book title: The Copywriter's Handbook
Author: Robert W. Bly
Text path: “D:\360MoveData\Users\win\Desktop\文案创作完全手册-罗伯特·布莱.pdf”
After distillation is complete, we use this skill and not use the skill to do the same project:
Using the distilled skill:
Prompt:
Please invoke $feature-to-benefit to complete the following task.
Please write the product information below as landing page first-screen copy.
Product: Piaojia, an invoice organization software for small business finance teams.
Target users:
Finance teams of 2–5 people, processing 300–1,000 invoices per month.
Product information:
1. Can upload 30 electronic invoice PDFs at a time.
2. In internal testing with 500 standard electronic invoices, the comprehensive recognition accuracy for fields such as amount, date, and invoice code was 96.8%. Paper invoice photos were not tested.
3. The system prompts for duplicate invoices based on invoice code and invoice number.
4. Can export Kingdee standard templates; currently does not support Yonyou templates.
5. Administrators can set member permissions and view operation records.
6. Interviewed 8 finance personnel, 6 of whom said that manual entry and checking for duplicate invoices take up a large amount of time in invoice organization.
Please output:
1. A title of no more than 18 characters
2. A subtitle of no more than 50 characters
3. Three selling point cards
4. A trial button copy
Not using the skill:
Prompt:
Please write the product information below as landing page first-screen copy.
Product: Piaojia, an invoice organization software for small business finance teams.
Target users:
Finance teams of 2–5 people, processing 300–1,000 invoices per month.
Product information:
1. Can upload 30 electronic invoice PDFs at a time.
2. In internal testing with 500 standard electronic invoices, the comprehensive recognition accuracy for fields such as amount, date, and invoice code was 96.8%. Paper invoice photos were not tested.
3. The system prompts for duplicate invoices based on invoice code and invoice number.
4. Can export Kingdee standard templates; currently does not support Yonyou templates.
5. Administrators can set member permissions and view operation records.
6. Interviewed 8 finance staff, of whom 6 said that manual entry and checking for duplicate invoices took up a large amount of time in invoice organization.
Please output:
1. A title of no more than 18 characters
2. A subtitle of no more than 50 characters
3. Three selling-point cards
4. Copy for a trial button
The no-Skill version has target users and a sense of business volume, but there are several problems: the thousand invoices come from target users processing 300-1,000 invoices per month, and do not mean the invoice folder has already verified thousand-invoice processing capability; duplicate checking, collaboration, and export completed in one place stuffs in too many features; the button says free trial of the invoice folder, but the original materials do not have a free policy.
After calling feature-to-benefit, the title became: Recognize invoices in batches, do less manual entry, linking product features with the benefits finance staff get.
The three cards respectively handle entry, duplicate checking, and Kingdee export, and each benefit can be traced back to product facts. Duplicate checking is written as: flag suspected duplicates, avoiding the implication that the system can completely eliminate duplicate reimbursement.
case 2 video distillation skill
We directly call this skill to distill:
Prompt:
https://www.bilibili.com/video/BV1xQnPzdEiy/ Help me distill this lighting effect one
After distillation is complete, we use this skill and not use the skill to do the same project:
Using the distilled skill:
Prompt:
Please first call $prompt-lighting-director to generate a 16:9 psychological horror game promotional image. Subject: A young Asian woman with short black hair, wearing a plain light gray shirt, standing in front of a gray concrete wall. The person faces the camera directly, framed from above the chest, with a calm expression and no exaggerated movements.
Not using the skill:
Prompt:
Generate a 16:9 psychological horror game promotional image.
Subject:
A young Asian woman with short black hair, wearing a plain light gray shirt, standing in front of a gray concrete wall. The person faces the camera directly, framed from above the chest, with a calm expression and no exaggerated movements.
The lighting effect group shows the difference most clearly. The no-Skill image uses common cool light from above and slightly in front, with the forehead, bridge of the nose, and cheekbones brightest, and the shadows of the nose and chin falling downward. The person's face is clear, and the gray concrete wall also has texture, but overall it looks more like an interrogation room character photo, and the horror feel is not strong.
After calling prompt-lighting-director, the light source was placed below the person's chest. The chin, nostrils, and lower lip became the brightest areas, and the nose shadow, cheekbone shadow, and eye shadows moved upward abnormally, with the sense of pressure produced entirely by the shadow direction.
case 3 podcast distillation skill
We directly call this skill to distill:
Prompt:
Please call cangjie-skill to distill the following podcast into a Skill.
https://podscripts.co/podcasts/acquired/superhuman-with-ceo-rahul-vohra
After distillation is complete, we use this skill and not use the skill to do the same project:
Using the distilled skill:
Prompt:
Please call $superhuman-pmf-engine to complete the following task.
We are testing an AI meeting minutes product. Please determine whether the product has already achieved PMF, who it should prioritize serving, and what should be developed in the next six weeks.
Sample rules:
All respondents completed at least two real meetings within the last 14 days, and all have experienced the action item extraction feature. There are no duplicate responses and no missing data.
A total of 48 valid responses:
Founders, 12 people:
– Very disappointed 8 people
– Somewhat disappointed 3 people
– Not disappointed 1 person
Sales managers, 12 people:
– Very disappointed 7 people
– Somewhat disappointed 4 people
– Not disappointed 1 person
Product managers, 12 people:
– Very disappointed 2 people
– Somewhat disappointed 5 people
– Not disappointed 5 people
Independent consultants, 12 people:
– Very disappointed 1 person
– Somewhat disappointed 5 people
– Not disappointed 6 people
Main benefits mentioned by the 18 "very disappointed" users:
– Accurate action items: 13 people
– Searchable meeting notes: 3 people
– Other: 2 people
Among the 17 "somewhat disappointed" users, 8 also recognized "accurate action items":
– 5 people were mainly blocked by calendar synchronization issues
– 2 people were mainly blocked by CRM synchronization issues
– 1 person needs custom export formats
Behavioral data for the same period:
– Founders' four-week retention rate 75%, paid conversion rate 50%
– Sales managers' four-week retention rate 67%, paid conversion rate 42%
– Product managers' four-week retention rate 33%, paid conversion rate 17%
– Independent consultants' four-week retention rate 25%, paid conversion rate 8%
– Among all registered users, the activation rate for completing two core actions was only 24%
Please output:
1. PMF judgment
2. Priority target users
3. Roadmap for the next six weeks
4. Whether customer acquisition should be expanded
5. Next round of survey plan
Without using skill:
Prompt: We are testing an AI meeting minutes product. Please determine whether the product has achieved PMF, whom it should prioritize serving, and what should be developed in the next six weeks. Sample rules: All respondents completed at least two real meetings within the last 14 days and have all experienced the action item extraction feature. There are no duplicate responses or missing data. A total of 48 valid responses: Founders, 12 people:
– Very disappointed 8 people
– Somewhat disappointed 3 people
– Not disappointed 1 person
Sales managers, 12 people:
– Very disappointed 7 people
– Somewhat disappointed 4 people
– Not disappointed 1 person
Product managers, 12 people:
– Very disappointed 2 people
– Somewhat disappointed 5 people
– Not disappointed 5 people
Independent consultants, 12 people:
– Very disappointed 1 person
– Somewhat disappointed 5 people
– Not disappointed 6 people
Main benefits mentioned by the 18 "very disappointed" users:
– Accurate action items: 13 people
– Searchable meeting notes: 3 people
– Other: 2 people
Among the 17 "somewhat disappointed" users, 8 also recognized "accurate action items":
– 5 people were mainly blocked by calendar synchronization issues
– 2 people were mainly blocked by CRM synchronization issues
– 1 person needs custom export formats
Behavioral data for the same period:
– Founder 4-week retention rate 75%, paid conversion rate 50%
– Sales manager 4-week retention rate 67%, paid conversion rate 42%
– Product manager 4-week retention rate 33%, paid conversion rate 17%
– Independent consultant 4-week retention rate 25%, paid conversion rate 8%
– Among all registered users, the activation rate for completing two core actions is only 24%
Please output:
PMF assessment
Priority target users
Roadmap for the next six weeks
Whether customer acquisition should be expanded
Next round of research plan
The no-Skill version directly labels the results of each group of 12 people as strongest PMF and strong PMF, without reminding that the subgroup samples are small. Information that was not provided also appears in the founder profile.
superhuman-pmf-engine retains the denominator for each judgment. The Skill also confirmed the common benefit of accurate action items from the 18 strong supporters, and then screened out the 8 people who also recognized this benefit from the 17 somewhat disappointed users. The roadmap arranges accurate action items, calendar sync, and the activation process.
02. Project introduction
Cangjie-skill is a methodology specification for "converting long-form content into AI Skills."Cangjie-skill It reads books, video subtitles, podcast transcripts, courses, or interviews, and breaks the reusable methodologies within them into multiple independent Skills, to be invoked by an AI Agent in specific problems.
Project address:https://github.com/kangarooking/cangjie-skill
Cangjie-skill will solve several practical problems:
- Which viewpoints qualify to become an independent Skill;
- When the user says what, the Agent should invoke the corresponding Skill;
- Which steps the Agent should execute after invocation;
- Which situations require stopping or switching to another method;
- How multiple Skills combine, and how to avoid competing with each other for invocation.
Cangjie-skill is more suitable for:
- Business, investment, management, and technical courses with high methodology density;
- Internal team training materials and expert interviews;
- Researchers who want to build a personal Agent tool library;
- Content distillation that needs to preserve sources, elimination records, and method boundaries.
To use this project, just have an AI Agent install it for you. Once installed, you can go distill the books, videos, or podcasts you like.
03. Dig a little
In the past, after finishing a good book or watching a video, what was left in the end was often a collection, notes, and a vague impression. When actually encountering a project, one still had to go through the materials again, and many useful methods were gradually forgotten.
Cangjie-skill gives good content a more practical way to be preserved. Copywriting methods from books, lighting experience from videos, and product judgment frameworks from podcasts can all be broken into Skills, preserved together with applicable conditions, execution steps, and usage boundaries.
The next time you write product copy, design lighting effects, or judge PMF, the Agent can directly invoke the corresponding Skill and work according to the same methodology. Individuals take fewer detours from repeated research, and teams can continue to use the experience left by experts.
In the future, when encountering books, videos, or podcasts worth repeatedly learning from, you can take one more step: turn the most useful methods within them into Skills.
Collections preserve content, Skills preserve the experience within the content, letting good content continue to play a role in one real project after another.
Original link:GitHub racks up 4.2K Stars, turning all knowledge into Skills