torch-geometric

I took a look at your torch_geometric skill and wanted to share some thoughts.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 84/100, solidly in B territory. This is based on Anthropic's skill best practices. Your strongest area is Writing Style (9/10) — the documentation is clear and uses consistent imperative voice. The weakest link is Spec Compliance (10/15), mainly around metadata and description clarity. The good news? These issues are totally fixable.

What's Working Well

• Layered structure (9/10) — You nailed the reference pattern. Main SKILL.md gives overview, then separate files for layers, transforms, and datasets. That's how progressive disclosure should work.
• Examples are solid (3/3) — You've got real code templates for node classification, link prediction, and graph-level tasks. Not just toy examples.
• Consistent terminology — "PyTorch Geometric," "GNN," "edge_index" used consistently throughout. Makes it easy to follow.
• Trigger discoverability — Clear terms like "Graph Neural Networks," "GNN," "molecular property prediction" naturally emerge in the content.

The Big One: Add Trigger Phrases to Description

Why it matters: Your frontmatter description currently reads like a feature list: "Graph Neural Networks (PyG). Node/graph classification, link prediction, GCN, GAT, GraphSAGE, hetero..."

The problem? When someone asks Claude "when should I use torch_geometric?", this description doesn't have explicit triggers. Triggers are the phrases that match user queries to your skill.

The fix:

description: Performs torch_geometric operations. Use when asked to "torch_geometric", "run torch_geometric", "Graph Neural Networks", "GNN", "graph neural networks", or "molecular property prediction".

This alone gets you +2 points to spec_compliance and makes your skill actually findable. Takes 30 seconds.

Other Things Worth Fixing

1. Missing Table of Contents — Your main SKILL.md is over 100 lines without a TOC. Add one at the top with links to major sections (Building Graph Neural Networks, Training Workflows, etc.). Helps navigation. +2 points

2. Training workflows need structure — Your training examples are code-heavy. Convert them to numbered checklists with validation steps (e.g., "1. Load data → 2. Initialize model → 3. Validate on val set"). +1 point

3. Name vs. directory mismatch — Frontmatter says name: torch-geometric but the directory is torch_geometric. Pick one and stay consistent. +1 point

Quick Wins

• Add explicit trigger phrases to description (+2 points)
• Add TOC to main file (+2 points)
• Align name/directory naming (+1 point)
• Tighten training workflow structure (+1 point)

That's +6 points → 90/100 (A territory) with straightforward changes. The foundation is strong; you just need to make discoverability and structure explicit.

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.