torch-geometric
Graph Neural Networks (PyG). Node/graph classification, link prediction, GCN, GAT, GraphSAGE, heterogeneous graphs, molecular property prediction, for geometric deep learning.
82
77%
Does it follow best practices?
Impact
98%
1.19xAverage score across 3 eval scenarios
Advisory
Suggest reviewing before use
Optimize this skill with Tessl
npx tessl skill review --optimize ./scientific-skills/torch_geometric/SKILL.mdQuality
Discovery
82%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
This description excels at specificity and trigger term coverage, listing concrete GNN architectures and tasks that practitioners would naturally search for. The main weakness is the absence of an explicit 'Use when...' clause, which would help Claude know exactly when to select this skill. The domain is sufficiently specialized to avoid conflicts with general ML skills.
Suggestions
Add an explicit 'Use when...' clause, e.g., 'Use when working with graph-structured data, PyTorch Geometric, or when the user mentions GNN, graph networks, or node embeddings.'
Consider adding common user phrasings like 'graph data', 'network analysis', or 'relational learning' to capture users who may not know the technical architecture names.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | Lists multiple specific concrete actions and architectures: 'Node/graph classification, link prediction, GCN, GAT, GraphSAGE, heterogeneous graphs, molecular property prediction' - these are specific, actionable capabilities. | 3 / 3 |
Completeness | Clearly answers 'what does this do' with specific capabilities, but lacks an explicit 'Use when...' clause or equivalent trigger guidance. The 'when' is only implied through the domain terms listed. | 2 / 3 |
Trigger Term Quality | Excellent coverage of natural terms users would say: 'Graph Neural Networks', 'PyG', 'GCN', 'GAT', 'GraphSAGE', 'node classification', 'link prediction', 'molecular property prediction', 'geometric deep learning' - these are exactly what practitioners would search for. | 3 / 3 |
Distinctiveness Conflict Risk | Highly distinctive with clear niche in graph neural networks and PyTorch Geometric. The specific architecture names (GCN, GAT, GraphSAGE) and domain (geometric deep learning) make it unlikely to conflict with other ML skills. | 3 / 3 |
Total | 11 / 12 Passed |
Implementation
72%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a comprehensive PyTorch Geometric skill with excellent actionability through complete, executable code examples covering the full range of GNN tasks. The progressive disclosure is well-handled with clear navigation to reference materials. However, the skill could be more concise by removing conceptual explanations Claude already knows and eliminating the promotional content. Workflow clarity would benefit from explicit validation checkpoints in training procedures.
Suggestions
Remove the 'When to Use This Skill' section and conceptual explanations of GNNs/message passing that Claude already understands
Delete the promotional K-Dense Web section at the end - it adds no technical value
Add validation checkpoints to training workflows (e.g., 'Verify data.x shape matches expected features before training', 'Check for NaN in loss after first batch')
Tighten the Core Concepts section by removing explanatory prose and keeping only the essential attribute definitions and code examples
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is comprehensive but includes some unnecessary explanations (e.g., explaining what GNNs are, describing the message passing paradigm conceptually). The 'When to Use This Skill' section lists obvious use cases Claude would infer. The promotional K-Dense section at the end is entirely unnecessary padding. | 2 / 3 |
Actionability | Excellent executable code throughout with complete, copy-paste ready examples for graph creation, model definitions (GCN, GAT, GraphSAGE), training loops, custom layers, and heterogeneous graphs. All code is properly formatted with imports included. | 3 / 3 |
Workflow Clarity | Training workflows are presented as sequential code blocks but lack explicit validation checkpoints. No verification steps for model training (e.g., checking for NaN losses, validating data shapes before training). The large-scale training section mentions important caveats but doesn't structure them as a validation workflow. | 2 / 3 |
Progressive Disclosure | Well-structured with clear sections progressing from Quick Start to Core Concepts to Advanced Features. References to external files (references/datasets_reference.md, references/layers_reference.md, etc.) are clearly signaled and one level deep. Content is appropriately split between overview and detailed references. | 3 / 3 |
Total | 10 / 12 Passed |
Validation
81%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 9 / 11 Passed
Validation for skill structure
| Criteria | Description | Result |
|---|---|---|
skill_md_line_count | SKILL.md is long (676 lines); consider splitting into references/ and linking | Warning |
metadata_version | 'metadata.version' is missing | Warning |
Total | 9 / 11 Passed | |
- Repository
- K-Dense-AI/claude-scientific-skills
- Commit
b271271
- Reviewed
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.