Looking at your senior data engineer skill—I'm curious how you're thinking about the progression from spec to actual implementation, since there's a gap there that's affecting the score.
Links:
The TL;DR
You're at 43/100, which puts you in F territory. This is based on Anthropic's best practices for agentic skills. Your strongest area is Spec Compliance (12/15)—the YAML frontmatter and metadata are solid. But Progressive Disclosure Architecture (8/30) is where you're losing the most points. The reference files are basically empty templates, and the main file is bloated with marketing language instead of concrete guidance.
What's Working Well
- Metadata is clean. Your frontmatter validates, name follows conventions, and the description has solid trigger terms (designing data architectures, building pipelines, etc.)
- Spec compliance is tight. You nailed the required fields and file structure—no friction there.
- Breadth of scope. Covering ML, LLM, ETL, and streaming shows you're thinking about the full data engineering landscape.
The Big One: Empty Reference Files
This is your biggest issue. You've got three reference files (data_modeling_patterns.md, data_pipeline_architecture.md, dataops_best_practices.md) that are basically identical boilerplate placeholders. They say things like "World-class data pipeline architecture" and "Core Principles" but provide zero actual content—no schema patterns, no architecture diagrams, no CI/CD strategies, nothing.
Why it matters: References are supposed to be your PDA leverage. They let you keep SKILL.md lean while providing depth on demand. Right now they're just adding noise.
Concrete fix: Replace the boilerplate with actual content:
data_pipeline_architecture.md: Lambda vs Kappa architectures, stream vs batch tradeoffs, backpressure handling, exactly-once semantics with code snippetsdata_modeling_patterns.md: Star/snowflake schemas, SCD types, slowly changing dimensions with SQL examplesdataops_best_practices.md: Data testing frameworks, data contracts, pipeline observability, sample health check queries
Estimate: +12 points if you nail this.
Other Things Worth Fixing
-
Kill the marketing speak. "World-class" appears 6+ times. "Senior-level" gets repeated. Remove all of it—just say what the skill does. "Build scalable data pipelines" beats "World-class capabilities for building scalable data pipelines" every time. Impact: +5 points
-
Add workflows, not just commands. Your Quick Start shows command examples but no step-by-step workflows. Add numbered checklists for common tasks (e.g., "Building a Batch ETL Pipeline: 1) Design data model, 2) Validate schema, 3) Create pipeline, 4) Test"). Impact: +4 points
-
Narrow your scope or add decision frameworks. You're covering ML, LLMs, ETL, and streaming—that's huge. Either focus on one domain, or add a decision tree ("Use streaming for real-time events, batch for historical aggregations"). Right now it feels scattered. Impact: +3 points
-
Add examples and troubleshooting. Include before/after scenarios, sample configs, and "If X fails, try Y" sections. This bridges the gap between concept and execution. Impact: +3 points
Quick Wins
- Strip marketing language from every section
- Replace reference file boilerplate with one complete, concrete example per file
- Add a numbered workflow for at least one common data engineering task
- Include a troubleshooting section in SKILL.md
- Consider a table of contents at the top—227 lines is long without navigation
This is solid foundational work. The scaffolding is there. You just need to fill in the substance and cut the fluff.
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.
Looking at your senior data engineer skill—I'm curious how you're thinking about the progression from spec to actual implementation, since there's a gap there that's affecting the score.
Links:
The TL;DR
You're at 43/100, which puts you in F territory. This is based on Anthropic's best practices for agentic skills. Your strongest area is Spec Compliance (12/15)—the YAML frontmatter and metadata are solid. But Progressive Disclosure Architecture (8/30) is where you're losing the most points. The reference files are basically empty templates, and the main file is bloated with marketing language instead of concrete guidance.
What's Working Well
The Big One: Empty Reference Files
This is your biggest issue. You've got three reference files (
data_modeling_patterns.md,data_pipeline_architecture.md,dataops_best_practices.md) that are basically identical boilerplate placeholders. They say things like "World-class data pipeline architecture" and "Core Principles" but provide zero actual content—no schema patterns, no architecture diagrams, no CI/CD strategies, nothing.Why it matters: References are supposed to be your PDA leverage. They let you keep SKILL.md lean while providing depth on demand. Right now they're just adding noise.
Concrete fix: Replace the boilerplate with actual content:
data_pipeline_architecture.md: Lambda vs Kappa architectures, stream vs batch tradeoffs, backpressure handling, exactly-once semantics with code snippetsdata_modeling_patterns.md: Star/snowflake schemas, SCD types, slowly changing dimensions with SQL examplesdataops_best_practices.md: Data testing frameworks, data contracts, pipeline observability, sample health check queriesEstimate: +12 points if you nail this.
Other Things Worth Fixing
Kill the marketing speak. "World-class" appears 6+ times. "Senior-level" gets repeated. Remove all of it—just say what the skill does. "Build scalable data pipelines" beats "World-class capabilities for building scalable data pipelines" every time. Impact: +5 points
Add workflows, not just commands. Your Quick Start shows command examples but no step-by-step workflows. Add numbered checklists for common tasks (e.g., "Building a Batch ETL Pipeline: 1) Design data model, 2) Validate schema, 3) Create pipeline, 4) Test"). Impact: +4 points
Narrow your scope or add decision frameworks. You're covering ML, LLMs, ETL, and streaming—that's huge. Either focus on one domain, or add a decision tree ("Use streaming for real-time events, batch for historical aggregations"). Right now it feels scattered. Impact: +3 points
Add examples and troubleshooting. Include before/after scenarios, sample configs, and "If X fails, try Y" sections. This bridges the gap between concept and execution. Impact: +3 points
Quick Wins
This is solid foundational work. The scaffolding is there. You just need to fill in the substance and cut the fluff.
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.