torch-geometric

Graph Neural Networks (PyG). Node/graph classification, link prediction, GCN, GAT, GraphSAGE, heterogeneous graphs, molecular property prediction, for geometric deep learning.

Quality

63%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/coding/torch_geometric/SKILL.md

Quality

Content

61%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A thorough, highly actionable reference body with executable examples for every major GNN task and a well-organized bundle. It is weakened by redundant boilerplate examples, sections covering knowledge Claude already has, and the absence of validation/feedback checkpoints in the training workflows.

Suggestions

Consolidate the triplicated GCN/GAT/GraphSAGE model boilerplate into one example plus a table of layer differences, and show dataset loading only once.

Remove or trim sections that re-teach standard PyTorch ('Save and Load Models', 'GPU Training', the generic message-passing theory) to respect the token budget.

Add a validation checkpoint pattern to the training workflows (e.g., checking loss for NaN, evaluating on a validation split each epoch with early stopping) and move the inlined 'Layer Capabilities' and 'Common transforms' catalogs fully into the existing reference files.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient, code-forward, and PyG-specific, but includes unnecessary padding: Planetoid loading is demonstrated three times, GCN/GAT/GraphSAGE boilerplate is triplicated with near-identical forwards, and sections like 'Save and Load Models' and 'GPU Training' re-explain basic PyTorch knowledge Claude already has.	3 / 5
Actionability	Nearly all guidance is concrete, executable code covering installation, graph creation, datasets, models, and full training loops. Minor gaps remain: the 'to_hetero' snippet uses a 'model = GNN(...)' placeholder, and some snippets reference 'F' or 'GCNConv' without the corresponding import.	4 / 5
Workflow Clarity	Training workflows are sequenced clearly (load dataset, define model, train, evaluate) but contain no validation checkpoints or error-recovery feedback loops (e.g., loss sanity checks, early stopping), matching the 'sequence present but checkpoints missing' anchor.	3 / 5
Progressive Disclosure	Good structure with real, well-signaled, one-level-deep references ('Check references/datasets_reference.md for a comprehensive list') and a Bundled References section covering all three reference files and three scripts. Minor gaps: SKILL.md inlines catalog-style content (the 'Layer Capabilities' flags and 'Common transforms' list) that duplicates the reference files and could be delegated.	4 / 5
	Total	14 / 20 Passed

Description

66%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A capability-dense, fairly specific description that clearly communicates the PyG domain and its concrete features. Its main weakness is the missing explicit 'Use when...' trigger guidance and the absence of common synonym terms like 'GNN', which cap completeness and trigger-term quality.

Suggestions

Append an explicit trigger clause, e.g. 'Use when the user mentions Graph Neural Networks, GNNs, node/graph classification, link prediction, or graph data in PyTorch.'

Add natural synonym keywords users actually say, such as 'GNN', 'graph learning', and 'knowledge graphs'.

Extend the capability list slightly to cover the skill's other documented strengths (training workflows, transforms, model explainability) for fuller coverage.

Dimension	Reasoning	Score
Specificity	Lists several specific concrete capabilities ('Node/graph classification, link prediction, GCN, GAT, GraphSAGE, heterogeneous graphs, molecular property prediction'), well above the 1-2-action anchor but with minor coverage gaps (no mention of training workflows, transforms, or explainability) that keep it below the comprehensive anchor.	4 / 5
Completeness	The 'what' is clear and concrete, but there is no 'Use when...' clause; the trailing 'for geometric deep learning' only weakly implies when to use the skill, so per the rubric cap completeness cannot exceed 3.	3 / 5
Trigger Term Quality	Good coverage of natural terms users would say (node classification, link prediction, GAT, molecular property prediction, geometric deep learning), but common variations like the 'GNN' abbreviation, 'graph learning', and 'knowledge graph' are missing, so it falls short of the comprehensive-synonyms anchor.	4 / 5
Distinctiveness Conflict Risk	Leads with a clear niche ('Graph Neural Networks (PyG)') and distinctive triggers (GCN, GAT, GraphSAGE), with only minor overlap risk against closely related skills like a general PyTorch or deep-learning skill, matching the 'mostly distinct' anchor rather than the minimal-conflict anchor.	4 / 5
	Total	15 / 20 Passed

Validation

81%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (683 lines); consider splitting into references/ and linking	Warning
metadata_version	'metadata.version' is missing	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/coding/torch_geometric/SKILL.md
Commit: 4082a2e

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

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.