Guides built on test data · page 2 of 6
The SkillProof blog

Directories advertise 23,000+ Claude skills. How many did we find actually run?
Marketplaces claim 23,000+ Claude skills exist. We installed and ran 1416 to find a real count: only 63% pass. A data-driven look at skill quality.

GitHub stars vs our tested score: does popularity predict a good Claude skill?
We analyzed 1416 Claude skills to see if GitHub stars predict quality. The data shows popularity is a weak signal for performance and reliability.

A 100K+ star Karpathy-style CLAUDE.md: does that pattern beat plain Claude?
A technical review of the popular Karpathy-inspired CLAUDE.md. We tested the adversarial planning pattern it uses to see if it actually works.

What share of Claude skills actually beat plain Claude? We measured it across 1,416
We tested 1416 Claude skills against a no-skill baseline to find the real percentage that work. See the data on which skills actually improve output.

Token-efficiency skills promise big savings. Here is what we actually measured
Token efficiency skills for Claude claim up to 91% savings. Our benchmarks show they add ~29% overhead on one-shot tasks. Here's where they actually work.

We Built an MCP Server So Claude Can Check a Skill Before Installing It
Claude Code can now ask whether a skill was tested — and whether it works — before it runs a single install command. One line to add, two tools, 743 tested skills behind it. Free, MIT.

The Best Claude Skills for Coding, Tested (2026)
We tested 250 Claude skills for coding on real-world tasks. See our ranked list of which tools actually beat the baseline, and which ones failed completely

Claude Skill Frontmatter: Every Field, Explained
A complete technical reference for Claude SKILL.md frontmatter. We explain name, description, allowed-tools, user-invocable, and disable-model-invocation b

Claude Skills for Agencies and Freelancers, Tested
We tested 743 Claude skills on real agency work. Discover which skills for writing, SEO, and presentations actually speed up client delivery.

Claude Skills for DevOps and Platform Engineering, Tested
We tested Claude skills for DevOps, Kubernetes, and platform engineering. See which passed, which failed, and which require a paid backend. Honest verdicts

Claude Skills for Startup Founders, Tested
We tested hundreds of Claude skills on real startup tasks—from cap tables to launch copy. Here are the few that actually work and earn their place.

How to Update Claude Skills (and Know When They Break)
A technical guide on how to update Claude skills. Learn why skills silently break due to dependency drift and how to test if a skill is still effective.

Claude Skill allowed-tools: Scope a Skill's Permissions Right
Learn how the `allowed-tools` frontmatter field controls Claude skill permissions. We cover proper scoping, security risks, and real-world examples of fail

How Claude Decides Which Skill to Load
Claude skill triggering is controlled entirely by the description field. We explain how this mechanism works, why vague descriptions cause skills to fail,

Claude Skills for Security Review, Tested
We tested Claude skills for security review on real code with planted vulnerabilities. See which skills passed, which failed, and why a checklist can be a