I Ran 60+ Automated Tests on My AI Skills Registry — Here's What Broke

A developer built an open registry indexing 5,090 AI agent skills from 200+ GitHub repos and ran 60+ automated tests on it. Key findings: auto-tagging was wrong 50% of the time due to keyword mismatches, the resolve API returned perfect results only 45% of the time (80% usable), and keyword search outperformed semantic search for structured technical queries. Security headers initially passed only 1 of 7 checks but were fixed to 7/7.