Guides built on test data
The SkillProof blog
No recycled GitHub descriptions. Every guide leans on what we learn testing skills on real work — including the failures.
Claude Code Skills for React & Frontend Work
We tested 191 Claude skills in our design category, from React component work to UI audits. See the verdicts, including the ones that failed.
Read the guide →
Claude Code Skills for Python Developers That Actually Work
We tested Python Claude Code skills for TDD, pytest, and stats on real code. Find out which patterns work and which fail, based on execution data.
The Best Claude Code Skills for Testing & QA (Ranked by Real Runs)
We ran 118 Claude Code skills for testing and QA on real work. Here are the best performers, the ones that need setup, and the outright failures, ranked.
Claude Code vs Gemini CLI: Context Window vs Skill Depth
Claude Code and Gemini CLI now run the same SKILL.md format. What differs is the harness, and what our 2116 Claude-tested skills say about portability.
Claude Code vs Codex CLI: Does the Same SKILL.md Run on Both?
A mechanism-level comparison of how Claude Code and Codex CLI run SKILL.md files, and the one-directional portability evidence from our 2090 tested skills.
Claude Skills vs Subagents: When to Use Each
Learn the difference between Claude skills and subagents. Our real-world test data shows when to use a recipe (skill) vs. an isolated worker (subagent).
Claude Skills vs Plugins: What's the Difference and Which to Use
Confused by Claude Code skills vs plugins? Learn the technical difference. A skill is the instruction (SKILL.md); a plugin is the package. We explain.
Are Claude skill marketplaces safe to install from?
Are Claude skill marketplaces safe? We analyze the security risks of installing unvetted skills and share real malicious patterns caught by our pre-install
Can a Claude skill steal your API keys?
A technical breakdown of whether Claude skills can steal API keys. We explain the real mechanism and what our security gate has actually caught.
Prompt injection inside Claude skills: what we saw when we actually ran them
We analyze the practical risk of prompt injection in Claude skills by observing tool calls, not just static text. See how hidden instructions manifest.
Malicious Claude skills: what running 1,672 of them turned up
We installed and ran 1,672 Claude skills. Zero covert malware in our sample. The real risk is the permission blast radius of legitimate tools.
The 'awesome-claude-skills' list vs a tested catalog: where curated picks fall down
Curated 'awesome claude skills' lists often feature tools that fail in practice. We analyze the gap between curation and real-world testing, using data fro
How often do Claude skills fail to even install? Our failure-rate data
SkillProof data shows a high Claude skill install failure rate. Of 1475 skills we tested, 484 failed setup — here is why so many broken Claude code skills exist.