scalability-advisor

I found your scalability-advisor skill while reviewing distributed systems tooling—the way you've structured the guidance around architectural patterns rather than just optimization tricks is genuinely useful. Scored it an 83, and I think the main lift for a 90+ would be tightening up the PDA structure to reduce token overhead in the explanatory sections.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 83/100, solid B territory. This evaluation's based on Anthropic's skill best practices—specifically looking at Progressive Disclosure Architecture, ease of use, spec compliance, writing style, and utility. Your strongest area is Writing Style (9/10)—the content is concise and technically sound. Weakest is PDA (18/30)—mostly because you've packed everything into one 466-line file instead of using a layered reference structure.

What's Working Well

• Trigger phrase specificity: You nailed the discovery angle with concrete scenarios like "planning growth," "diagnosing bottlenecks," and "10x-1000x load changes." That's exactly how developers search for skills.
• Stage-based framework: Breaking scaling into Startup → Growth → Scale stages with clear metrics (10k users, 100k users, etc.) makes it actionable rather than abstract advice.
• Copy-paste ready content: Your SQL queries, formulas, and ASCII architecture diagrams are genuinely useful—the kind of thing someone can actually use without modification. That earned you a +2 bonus.
• Decision criteria: Each "When to Move to Stage X" section has concrete thresholds. That's rare and valuable.

The Big One: Missing Reference Files Breaking PDA

Your SKILL.md references two files that don't exist:

• bottleneck-diagnosis.md
• capacity-calculator.md

Why this matters: Progressive Disclosure Architecture is about splitting content across reference files so the main SKILL.md stays under ~150 lines. You're at 466 lines, which means a Claude instance loading your skill burns tokens upfront instead of lazily loading details. It also violates the one-level-deep requirement.

The fix: Either (1) create those two reference files with detailed diagnostic checklists and capacity planning templates, OR (2) remove the references and keep the content inline since your skill is already comprehensive. I'd suggest option 2 here—you don't have enough overflow content to justify splitting. Remove lines 464-465 and you're cleaner. This alone gets you +4 points to 87.

Other Things Worth Fixing

1. Add a table of contents — At 466 lines, you need a TOC after the description. Add one after line 10 with links to each Stage and major section. +2 points.

2. Make checklists actionable — Lines 241-249 say "Stateless application servers" but don't explain how to verify or what to change. Expand to: "Verify stateless servers: Check for local file writes, in-memory session storage; move state to Redis/database." +2 points.

3. Add a run→check→fix feedback loop — You provide metrics but no clear iteration pattern. Add a short section showing: "1. Implement (add caching) 2. Measure (cache hit rate) 3. Adjust (modify TTL if issues)." +1 point.

4. Sharpen sharding guidance — Your sharding strategies (hash-based, range-based, directory-based) list pros/cons but don't say when to pick each. Add decision rules: "Use hash-based for: even distribution. Use range-based for: time-series data." +1 point.

Quick Wins

• Remove broken references (464-465) → +4 points
• Add TOC after intro → +2 points
• Expand checklists with how-to steps → +2 points
• Add feedback loop pattern → +1 point

Hit those first three and you're at 92/100 territory.

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.