After Claude Code’s source code leaked, we took a look at a few tools out there that might help developers overcome some of its shortcomings. Check out the video below or read on for the links.
caveman
A Claude Code skill that cuts output tokens by ~75% by making the agent respond in stripped-back “caveman” English. The landing page boasts the skill retain 100% technical accuracy while destroying up to 87% of output latency.
What at first seems like a gimmick actually solves a problem in a simple but effective way, so we’ll be keeping an eye on this one.
🔗 github.com/JuliusBrussee/caveman
claude-token-efficient
Just like caveman, claude-token-efficient is a remarkably effective considering how simple it is, as it’s a drop-in CLAUDE.md for your project root that reduces Claude Code output by ~63%, and doesn’t require any code changes.
It stops the model restating questions, adding unsolicited suggestions, and using sycophantic filler. Plus you can add multiple profiles for different tasks.
🔗 github.com/drona23/claude-token-efficient
🤖 apify/agent-skills
Apify’s agent skills give Claude Code a range of web scraping and data collection abilities, from lead generation to competitor intelligence, trend analysis and more across platforms like Google Maps, Instagram, Amazon and more.
With web scraping historically being so notoriously fragile, Apify’s platform combined with these skills make it surprisingly easy to create scraper-based applications that would have been far more difficult, or at least more time intensive, to build well in the past.
🔗 skills.sh/apify/agent-skills
Bonus from the video: We took a look at Advanced Search for YouTube, and while it’s not Claude related we’d be remiss if we didn’t give you all the link.