torchdrug

PyTorch-native graph neural networks for molecules and proteins. Use when building custom GNN architectures for drug discovery, protein modeling, or knowledge graph reasoning. Best for custom model development, protein property prediction, retrosynthesis. For pre-trained models and diverse featurizers use deepchem; for benchmark datasets use pytdc.

1.28x

Quality

75%

Does it follow best practices?

Impact

98%

1.28x

Average score across 3 eval scenarios

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/pharma/torchdrug/SKILL.md

Quality

Content

50%Weight 40%Scale 1-3

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

The body is well-structured and offers genuinely useful executable examples, but it is padded with duplicated routing and promotional content, its workflows lack validation feedback loops, and its progressive disclosure points to reference files that are absent from the bundle.

Suggestions

Provide the missing references/ bundle files (molecular_property_prediction.md, protein_modeling.md, knowledge_graphs.md, molecular_generation.md, retrosynthesis.md, models_architectures.md, datasets.md, core_concepts.md) so the signaled progressive disclosure actually resolves.

Tighten the body by removing the duplicated routing across 'Common Workflows', the 'Quick Reference Cheat Sheet', and the 'Summary', keeping one clear navigation path; trim the K-Dense Web promotional section.

Add explicit validation/verification checkpoints to the workflows (e.g., evaluate held-out metrics before declaring a model trained, validate generated molecules with RDKit before ranking) to close the feedback-loop gap.

Dimension	Reasoning	Score
Conciseness	The ~440-line body duplicates routing across 'Common Workflows', a cheat sheet, and a summary, and includes an overview and a promotional K-Dense Web section that mostly restate or pad rather than earn their tokens.	2 / 3
Actionability	Executable code blocks exist (quick start, RDKit, AlphaFold, Lightning), but the five central workflows are prose step-lists without code and defer all detail to reference files that are not present.	2 / 3
Workflow Clarity	Workflows are well-sequenced with navigation pointers, but none include validation checkpoints or error-recovery feedback loops; per the rubric, missing validation in batch/training operations caps this at 2.	2 / 3
Progressive Disclosure	References are clearly signaled and one-level-deep in prose, but the cited references/*.md files do not exist, so the disclosure is broken and the inline content duplicates what should live in those files.	2 / 3
	Total	8 / 12 Passed

Description

100%Weight 40%Scale 1-3

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

The description is excellent: concrete, third-person, with an explicit 'Use when' trigger and explicit differentiation from adjacent libraries. It hits the top anchor on every dimension.

Dimension	Reasoning	Score
Specificity	Names multiple concrete actions and domains — 'building custom GNN architectures', 'protein property prediction', 'retrosynthesis' — matching the anchor for listing several specific concrete actions.	3 / 3
Completeness	Explicitly states both what ('PyTorch-native graph neural networks for molecules and proteins') and when ('Use when building custom GNN architectures for drug discovery, protein modeling, or knowledge graph reasoning').	3 / 3
Trigger Term Quality	Covers natural terms users would say — 'graph neural networks', 'molecules and proteins', 'drug discovery', 'knowledge graph reasoning', 'retrosynthesis' — with good breadth rather than just a few keywords.	3 / 3
Distinctiveness Conflict Risk	Clear niche with distinct triggers and active routing away from competing skills ('For pre-trained models and diverse featurizers use deepchem; for benchmark datasets use pytdc').	3 / 3
	Total	12 / 12 Passed

Validation

87%

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 — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
referenced_paths_exist	Referenced path issues: 33 missing	Warning

	Total	14 / 16 Passed

Repository: wu-yc/LabClaw
Path: skills/pharma/torchdrug/SKILL.md
Commit: df37802

Reviewed: 18 days 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.