senior-data-scientist

I just looked through senior-data-scientist and noticed the spec is pretty minimal—almost feels like it's still finding its voice. Given the 35/100, I'm curious what the original vision was before we started refining it.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 35/100, solidly F territory. This is based on Anthropic's skill evaluation rubric across five pillars. Your strongest area is Spec Compliance (12/15)—the YAML structure is valid and the naming convention is correct. But you're getting hit hard on Utility (4/20) and Progressive Disclosure (8/30), which are the backbone of a functional skill.

What's Working Well

• Clean metadata structure: Your YAML frontmatter is valid with required fields properly formatted—this is the foundation everything else builds on.
• Follows naming conventions: hyphen-case format is correct and consistent with the skill ecosystem standards.
• Reference architecture attempted: You're thinking layered—having separate files for statistical methods, experiment design, and feature engineering shows you understand the intent of Progressive Disclosure.

The Big One: Identical Reference Files Kill Utility

Here's the thing—all three reference files (statistical_methods_advanced.md, experiment_design_frameworks.md, feature_engineering_patterns.md) contain the identical boilerplate content. Same "Production-First Design" section, same "Pattern 1: Distributed Processing," everything copy-pasted.

Why this matters: The whole point of layered references is to provide progressively deeper, specialized knowledge. Right now they're just repetition, so a user gets zero additional value from clicking through.

The fix: Make these actually different. statistical_methods_advanced.md should cover hypothesis testing, causal inference, statistical significance. experiment_design_frameworks.md should cover A/B testing specifics, power analysis, randomization schemes. feature_engineering_patterns.md should cover transformations, feature selection, encoding strategies. Each file becomes a real resource instead of filler.

Impact: This alone could push you +10-12 points on Utility and PDA.

Other Things Worth Fixing

1. Fictional Quick Start scripts (SKILL.md:14-23): You're showing python scripts/experiment_designer.py --input data/ --output results/ but these don't exist anywhere. Either implement them in a scripts/ directory or replace with actual workflows like "1. Define hypothesis → 2. Calculate sample size → 3. Design randomization scheme" with links to references.

2. Marketing language everywhere: "World-class" appears 5+ times. This reads like a job posting, not instructional content. Strip it all out—just tell people what the skill does: "Statistical modeling, experimentation, and causal inference for production systems."

3. No actionable workflows: "Senior Responsibilities" is a job description. You need numbered checklists: "A/B Test Design: 1. Define metric 2. Calculate MDE 3. Run power analysis..." with actual steps users can follow.

4. Vague trigger phrases: Your description mentions "designing experiments, building models" but these are too generic. Get specific: "Use when setting up statistical tests, designing experiment randomization, calculating effect sizes."

Quick Wins

• Replace marketing terms (+5 points): Remove "world-class," "enterprise-scale," "senior-level" everywhere
• Differentiate your reference files (+10-12 points): Make each one domain-specific instead of identical boilerplate
• Add real workflows (+7 points): Convert lists into numbered, actionable checklists with clear steps
• Implement or remove Quick Start scripts (+6 points): Either create them or replace with conceptual workflow

Checkout your skill here: SkillzWave.ai | SpillWave We have an agentic skill installer that install skills in 14+ coding agent platforms. Check out this guide on how to improve your agentic skills.