excel-parser

I took a look at your excel-parser skill and wanted to share some thoughts.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 67/100, which puts you in D territory — solid foundation, but needs meaningful work before it's production-ready. Your strongest area is Spec Compliance (12/15), meaning the YAML frontmatter and naming conventions are spot-on. The real drag is Progressive Disclosure Architecture (17/30) — your skill is 371 lines with a lot of repeated logic, and that bloat is hurting discoverability and usability.

What's Working Well

• Spec compliance is tight. Your YAML frontmatter is valid, naming follows conventions, and required fields are all there.
• Real problem-solving approach. The Scout Pattern for intelligent Excel routing addresses an actual pain point — knowing whether to use Pandas or HTML parsing based on file complexity is genuinely useful.
• Clear decision boundaries. The complexity rules (row count thresholds, merge detection) give users a concrete way to decide their path without ambiguity.
• Good trigger terms. 'Excel/CSV', 'complexity analysis', 'merged cells' — these will help people find your skill when they need it.

The Big One: Bloat is Killing Your Architecture Score

Your 371-line SKILL.md has the same HTML conversion logic explained twice, and you're repeating workflow steps in a full implementation template at the bottom that's already documented above. This is a Progressive Disclosure fail — you're not layering complexity, you're duplicating it.

Here's the fix: Move your implementation templates to a separate references/smart_excel_router.py file. Keep SKILL.md focused on the why and when — when to use Scout Pattern, how complexity rules work, decision tree. The reference file contains the actual code. You also mention scripts/complexity_analyzer.py but don't provide it — that should be references/complexity_analyzer.py so it's actually available.

Impact: This alone gets you +4-6 points and makes your skill way easier to scan and actually use.

Other Things Worth Fixing

1. Add a Table of Contents. At 371 lines, readers need navigation. Add a "## Contents" section linking to Overview, Workflow, Complexity Rules, Path A, Path B, Best Practices. +2 points.

2. Pick a language. Your JSON examples and code comments mix Chinese and English ("行数过多 (>1000), 强制使用 Pandas 模式") without explanation. Either go bilingual consistently or stick to English. Confuses people trying to understand output. +2 points.

3. Complete the LLM integration. Your code template has a placeholder comment "you need to implement call_llm function" but zero guidance on how. Either provide a full implementation or reference Claude's tool usage pattern. +3 points.

4. Add error handling guidance. No mention of failure scenarios: corrupted files, invalid JSON from LLM, missing sheets. Add a "Troubleshooting" section with common errors and recovery steps. +2 points.

Quick Wins

• Move code to references/ — eliminate duplication, get +4-6 points
• Add TOC — helps navigation, +2 points
• Standardize language — clarity boost, +2 points
• Complete LLM guidance — makes skill actually usable, +3 points

Hit those four items and you're looking at 78-80/100 — solid C to B range.

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.