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.

Quality

79%

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/chemistry/torchdrug/SKILL.md

Quality

Content

71%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 well-structured overview with real executable examples and a genuinely effective one-level-deep reference bundle (all 8 files exist and match their in-body descriptions). Main weaknesses are redundancy in navigation (four overlapping pointer sections) and validation checkpoints that are listed but not enforced as gates.

Suggestions

Consolidate the redundant navigation into one section: drop the per-workflow 'Navigation:' lines and either the 'Quick Reference Cheat Sheet' or the closing 'Summary', which duplicate each other and the per-capability 'Reference:' blocks.

Add explicit validation gates to the workflows, e.g. 'check training/validation loss curves before evaluating' in Workflow 1 and 'validate predicted reaction centers against known splits before synthon completion' in Workflow 5.

Fix the minor code gaps for copy-paste readiness: add the missing torch import to the quick example, and complete the PyTorch Lightning example's validation_step metric logging.

Dimension	Reasoning	Score
Conciseness	The ~440-line body is mostly dense domain signal (datasets, models, task types) rather than filler, but navigation content is repeated four times — per-capability 'Reference:' blocks, per-workflow 'Navigation:' lines, the 'Quick Reference Cheat Sheet', and the closing 'Summary' — which is padding that could be consolidated.	3 / 5
Actionability	Concrete, near-executable code throughout: install commands, a full GIN/PropertyPrediction training loop, RDKit conversion, residue-graph construction, and a Lightning wrapper. Minor gaps: the quick example uses torch.optim without importing torch, and the Lightning validation_step returns a dict without the metric plumbing to make it runnable as-is.	4 / 5
Workflow Clarity	Five numbered workflows give clear sequences with concrete API choices (datasets.BBBP, RGCN center ID, GIN synthon completion) and most include evaluation/validation steps ('Evaluate using AUROC and AUPRC', 'Validate chemistry and filter'). Not 5 because validation checkpoints are mostly implicit — e.g., no step to confirm loss convergence before evaluating, or to verify predicted routes before 'apply recursively'.	4 / 5
Progressive Disclosure	Verified against the actual bundle: all 8 referenced files exist in references/ (169–565 lines each), references are one level deep, each is explicitly signaled with a 'See references/X.md for:' bullet list of its contents, and the body stays an overview while the detail lives in the bundle.	5 / 5
	Total	16 / 20 Passed

Description

87%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.

Strong description: third-person, concrete, with explicit 'Use when' triggers and explicit boundary guidance against overlapping skills. Keyword and capability coverage has only minor gaps (SMILES, molecular generation/property prediction terms).

Dimension	Reasoning	Score
Specificity	Quotes several concrete actions — 'building custom GNN architectures for drug discovery, protein modeling, or knowledge graph reasoning', 'protein property prediction, retrosynthesis' — but coverage has minor gaps (molecular property prediction and molecular generation are core capabilities of the skill yet go unmentioned in the description).	4 / 5
Completeness	Explicitly answers 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'), with additional 'Best for...' trigger guidance.	5 / 5
Trigger Term Quality	Good natural keyword coverage — 'PyTorch', 'graph neural networks', 'GNN', 'molecules', 'proteins', 'drug discovery', 'retrosynthesis' — but a few natural terms users would say are missing, e.g. 'SMILES', 'chemistry', 'molecular property prediction'.	4 / 5
Distinctiveness Conflict Risk	Clear PyTorch-native custom-GNN niche plus explicit de-confliction against sibling skills ('For pre-trained models and diverse featurizers use deepchem; for benchmark datasets use pytdc'), minimizing wrong-skill triggering.	5 / 5
	Total	18 / 20 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
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/chemistry/torchdrug/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.