The week where skills to reduce AI-likeness took 1st place
This is the first edition of the “Weekly GitHub Trends,” which I will publish every Sunday.
The project that grew the most this week was pbakaus/impeccable.
From Sunday to Saturday, it gained 10,346 stars.
Both the 1st and 3rd place projects were skills for designs that reduce AI-likeness.
Weekly Graph
Weekly Trend Graph
I record and compare the GitHub trending page every morning.
The lines show how many stars were added since the first day it appeared, with that day set to 0.
impeccable gained 2,656 stars in a single day on Friday.
It was the biggest day of the week for impeccable.
The 2nd place project, cloudflare/security-audit-skill, grew by about 1,200 to 1,700 stars every day.
The 3rd place project, Nutlope/hallmark, disappeared from the list after Thursday, so its line stops there.
It reached 3rd place based solely on its growth up to Thursday.
On Wednesday, HKUDS/DeepTutor appeared for the first time and climbed to 4th place by Saturday.
1st Place: pbakaus/impeccable
The design vocabulary for your coding agent.
(Source: https://github.com/pbakaus/impeccable README introduction, accessed 2026-11-01)
It was created by Paul Bakaus.
When you only say “make it look good,” the AI can only guess.
This is a skill that provides the AI with specific terms like “make it bolder” or “finish it,” along with the corresponding instruction manuals.
The README includes pages from before and after using it, created with the same model and the same prompt (these are summaries of examples from the README, not results run by the author).
-
Before: Cream-colored background, thin serif font, round buttons. A safe look.
-
After: A page centered around a single painting of a mansion seen from the sea at night.
There are also 62 rules for identifying AI-likeness, which do not use AI or API keys.
-
Type: Skill
-
Compatibility: Claude Code, Cursor, Codex, etc.
-
Commands: 24
-
License: Apache 2.0
-
Installation: npx impeccable install
-
Weekly growth: +10,346 (recorded from 10/25 to 10/31)
The README for the 3rd place project, hallmark, mentions impeccable as a reference.
2nd Place: cloudflare/security-audit-skill
A coding-agent skill that turns your agent into a security auditor.
(Source: https://github.com/cloudflare/security-audit-skill README introduction, accessed 2026-11-01)
This is a skill that reads code to find potential security vulnerabilities.
Its feature is separating the roles of finding and verifying.
Another agent questions the findings to see if they are real.
The usage example in the README is to ask it to “security audit this codebase”.
Results are recorded in three categories: confirmed, needs_validation, and rejected.
The investigation process has 6 steps, ranging from preliminary research to report generation.
The author’s tests showed that a single run found about half the number of issues compared to running it multiple times.
It also notes that a safe, isolated environment is required when actually running the target code.
-
Type: Skill
-
License: MIT
-
Installation: npx skills add https://github.com/cloudflare/security-audit-skill –skill security-audit
-
Weekly growth: +8,764 (recorded 10/25–10/31)
It was 1st in the GitHub weekly ranking on Saturday (10/31) with 12,838 weekly stars.
Since this article counts growth from Sunday, it is ranked 2nd.
3rd Place: Nutlope/hallmark
A design skill … that refuses to look AI-generated.
(Source: https://github.com/Nutlope/hallmark README introduction, accessed 2026-11-01)
It was created by Together AI.
This is a skill that prevents the look that screams “AI-generated” when you have an AI build a screen.
The README lists these signs specifically.
-
Small labels above headings
-
Only one word in a heading colored differently
-
Email signup at the end of every page
hallmark opens the created screen in a real browser to check for these signs.
It generates three themes, builds the first screen to verify, and finally, a fresh agent reviews it.
There are 50 themes, and it can also redesign existing pages.
-
Type: Skill
-
Compatibility: Claude Code, Codex, Cursor, OpenCode
-
License: MIT
-
How to install: npx skills add nutlope/hallmark -g
-
Weekly growth: +8,134 (recorded from 10/25 to 10/29)
Notable Newcomer: HKUDS/DeepTutor
an agent-native learning workspace that connects tutoring, research, creation, and practice
(Source: https://github.com/HKUDS/DeepTutor README “Key Features”, accessed 2026-11-01)
It appeared for the first time on Wednesday.
It is created by HKUDS.
It is an app that allows you to load your own materials and learn by asking questions or solving problems.
For example, in “Books,” it first shows you a draft of the chapter structure based on the selected materials, and then creates the content after you confirm it.
The chapters can also include quizzes, flashcards, and timelines.
It also has a feature to consult with other AIs, such as Claude Code or Codex.
-
Type: App
-
How to install: Docker/pip (e.g., python -m pip install -U “deeptutor[full]”)
-
License: Apache 2.0
-
Weekly growth: +5,864 (recorded from 10/28 to 10/31)
According to the README, October 28th, when it first appeared, was the release date for version 1.7.0.
However, it has not been confirmed if this is the reason for its growth.
It already had over 45,000 stars and is not a newly created repository.
Before installing a skill, make sure to read the instruction manual (SKILL.md) inside it once.
4th to 10th Place
-
4th: HKUDS/DeepTutor – AI Learning Tool – +5,864
-
5th: togg53192-cmd/jailbreaks – +5,488 (Content not introduced due to policy)
-
6th: JetBrains/thinkrail – AI Development Environment – +4,837
-
7th: TauricResearch/TradingAgents – Financial AI Research – +4,099
-
8th: makecindy/cindy – AI Agent – +3,793
-
9th: Renhuai123/ziwei-doushu – Divination Engine – +3,647
-
10th: AgriciDaniel/claude-seo – SEO Skill – +3,644
Six out of the seven repositories ranked 4th to 10th appeared mid-week.
Data collection method
I save the GitHub trending pages (daily and weekly) every morning.
I set the star count to 0 on the day each repository first appears, and define the increase from there as the ‘weekly growth’.
Since I don’t know the numbers for days when they don’t appear on the list, I don’t generate values for those days.
Therefore, the numbers and rankings may differ from the ‘weekly stars’ on GitHub’s weekly list.
Main video
Behind the scenes
One week since preparation
Since it was the first one, it took more time than expected.
Next time, I should be able to release it on Sunday… hopefully.
