claude-scientific-skills

I took a look at your claude-scientific-skills skill and wanted to share some thoughts.

Links:

• GitHub
• Your skill in the SkillzWave marketplace

The TL;DR

You're at 50/100, which lands you in failing territory. The grading is based on Anthropic's Progressive Disclosure Architecture framework – how efficiently skills guide Claude through use. Your strongest area is Utility (15/20) – the individual sub-skills like RDKit and Scanpy are legitimately excellent. The biggest drag is Progressive Disclosure Architecture (12/30) – the way this is structured breaks some fundamental expectations about how skills should work.

What's Working Well

• Sub-skill quality is chef's kiss. Your RDKit (764 lines), Scanpy, and ChEMBL implementations are domain-expert quality with real code examples, error handling, and practical workflows. This is the heavy lifting that makes the collection genuinely useful.
• Clear terminology consistency. Within each domain (cheminformatics, bioinformatics, etc.), you use consistent language and don't mix paradigms.
• Comprehensive coverage. 127 individual scientific skills across genomics, drug discovery, proteomics – you're addressing real capability gaps that researchers actually need.

The Big One: This Isn't a Skill, It's a Skill Collection – And The Architecture Reflects That

Here's the core issue: Your SKILL.md at the top level is 59 lines of marketing copy pointing to a subdirectory with 127 other skills. That's not a Progressive Disclosure Architecture – that's an index. And it costs you 8 points because:

1. Claude can't invoke this meaningfully. When someone asks "help me do sequence alignment," Claude sees a generic index with no trigger phrases. Should it load scanpy/SKILL.md? `biopyt...