Hey {{first_name | there}},
So I've been seeing people install 20, 30, sometimes even 50 Claude skills from GitHub.
And honestly, I get it. Some of them are genuinely useful.
But I came across a study that made me rethink how casually we're installing them.
Researchers found security issues in more than one in four AI agent skills they analyzed.
And some of those issues could potentially expose your private data.
Let me explain what they found and how you can check your own skills.
Your Claude Skills Might Not Be as Safe as You Think
And according to their security study:
26.1% contained at least one identified security vulnerability.
13.3% had potential data-exfiltration risks.
5.2% showed high-severity patterns suggesting malicious intent.
Now, this doesn't mean a quarter of all Claude skills are deliberately stealing data.
But here's what makes these skills risky. Unlike ordinary prompts, some skills contain scripts that Claude can run on your computer.
A malicious skill could try to read your files, access credentials, or send information somewhere you never intended.
And there's a name for one way attackers manipulate AI into doing this.
Prompt injection.
Here's what I'd do before installing another skill.
First, scan it.
Nvidia has a free, open-source tool called SkillSpector.
It checks agent skills for suspicious instructions, data-exfiltration patterns, dangerous code, and other risks.
You can also use Snyk Agent Scan to check skills and MCP servers you've already installed. It requires some setup.
Second, check what the skill actually needs.
A writing skill probably shouldn't need access to your passwords, browser cookies, or unrelated private files.
And if you can't understand why a skill needs certain permissions, don't give them automatically.
Security scanners aren't guarantees, but they're a much better starting point than blindly installing whatever's trending on GitHub.

Google Just Made Research a Lot More Interesting
Okay, here's something you can actually try today.
On November 16, Google announced a new Gemini Notebook feature.
You can now turn your documents, notes, and research into interactive websites.
Not just a summary.
Actual interactive pages with timelines, dashboards, trackers, and visual explanations.
And you don't need to code them.

PwC Wants 85,000 Employees to Learn How to Question AI
Here's a development I found pretty interesting.
After years of companies encouraging employees to use AI, PwC is now introducing a program to teach more than 85,000 employees and partners critical thinking and judgment.
According to PwC's announcement, the training will begin in December 2026.
Why?
Because when AI starts doing the research, analysis, and other basic tasks for you, you get fewer opportunities to practice those skills yourself.
And that creates a problem.
You might become really good at getting answers from AI without becoming any better at knowing whether those answers are right.

What I'd Try This Week
Here's my suggestion.
If you use Claude Code, spend a few minutes reviewing the skills you've installed. Remove the ones you don't use and check unfamiliar ones before running them again.
Then try Google's new interactive-site feature with a report or set of notes you've been putting off reading.
And finally, use that counterargument prompt on one actual decision you're making this week.
Three small experiments, but each teaches you something different: how to use AI more safely, how to make information easier to understand, and how to think more critically about its answers.

See you next time!
- AP
New AI tools launch every day. A few will change how you work. The rest will be forgotten by next month.
TLDR AI is the free newsletter that tells you which is which. Every morning, Anthropic and ex-Google engineers pick the new tools, updates, and open source releases worth trying, with a quick note on what each one does and who it's for.
You find the good ones early, without testing all of them.
One email, 5 minutes a day, read by 1.1M+ people. Get tomorrow's picks.

