diffdock

Predict small-molecule binding poses with DiffDock-L (Corso et al. 2023/2024, github.com/gcorso/DiffDock) — blind diffusion docking that places a ligand into a protein pocket without a predefined search box and ranks the samples with a learned confidence model. Reach for this skill to dock a SMILES or SDF against a PDB, to generate ranked 3D poses for a small fragment library, or to get a starting pose for downstream rescoring. DiffDock predicts geometry, not affinity.

Quality

93%

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

Quality

Content

87%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.

An exceptionally lean, high-value body: executable commands, exact batch CSV schema, and hard-won operational gotchas (silent precompute, RAM ceiling, YAML-over-CLI override, flag-name pitfall) with nothing wasted. The one gap is the absence of an output-validation checkpoint for batch runs, which the rubric caps workflow clarity at 3 for.

Suggestions

Add a short verification step after running (e.g., check that rank1.sdf exists for every complex_name in the batch and sanity-check that the reported confidence is a logit only comparable within the same complex) — this would lift workflow clarity above the batch-operation validation cap.

In the batch path, mention what an empty/failed row looks like in the output directory so a long library run can be checked mid-flight instead of only after completion.

Consider stating in the batch section how many samples per complex the default YAML produces, since the YAML-overrides-CLI gotcha makes the default sampling depth easy to get wrong silently.

Dimension	Reasoning	Score
Conciseness	Every section carries non-obvious, DiffDock-specific information (YAML config silently overwriting CLI flags, the ~11-minute silent precompute and 32 GB RAM requirement, the '--ligand' argparse prefix-match accident) with zero filler and no explanation of concepts Claude already knows. Lean and efficient — matches the 5 anchor.	5 / 5
Actionability	Fully executable guidance: a copy-paste-ready inference command with real flag values and an example SMILES, the exact four-column batch CSV schema, the output naming convention ('rank{N}_confidence{score}.sdf'), concrete fixes ('sed the constant in inference.py to min(64000, rlimit[1])', 'provider_params.modal.memory: 65536'), and an error-recognition table. Covers the common single-complex and batch cases.	5 / 5
Workflow Clarity	The run → output-interpretation → error-recovery flow is clear and the error table gives symptom-to-fix mapping, but the skill supports batch operations ('--protein_ligand_csv batch.csv', fragment-library screening) with no output-validation or verification checkpoint, and the rubric explicitly caps batch workflows without validation at 3.	3 / 5
Progressive Disclosure	The body is a lean overview that handles the single-complex path inline and clearly signals the one-level-deep split: 'that path and a larger-library screening recipe are in references/workflows.md' — a real, relevant file (verified). Content is appropriately divided with easy navigation.	5 / 5
	Total	18 / 20 Passed

Description

100%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.

An exemplary description: concrete third-person actions, explicit 'Reach for this skill to…' trigger guidance covering the single-complex, fragment-library, and starting-pose use cases, and an explicit capability boundary. Every token is informative with no fluff or over-claiming.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'places a ligand into a protein pocket', 'ranks the samples with a learned confidence model', 'dock a SMILES or SDF against a PDB', 'generate ranked 3D poses for a small fragment library', 'get a starting pose for downstream rescoring' — in third person with no padding. Coverage is comprehensive rather than having minor gaps, so it matches the 5 anchor rather than 4.	5 / 5
Completeness	Explicitly answers both: 'what' ('blind diffusion docking that places a ligand into a protein pocket… and ranks the samples with a learned confidence model') and 'when' ('Reach for this skill to dock a SMILES or SDF against a PDB, to generate ranked 3D poses…, or to get a starting pose for downstream rescoring'). It also states the boundary ('DiffDock predicts geometry, not affinity'), exceeding the 4 anchor's 'when could be more explicit'.	5 / 5
Trigger Term Quality	Natural domain phrasing a user would actually say: 'dock', 'binding poses', 'SMILES or SDF against a PDB', 'fragment library', 'rescoring', '3D poses', including format names and synonyms. Comprehensive coverage including file-format terms, matching the 5 anchor.	5 / 5
Distinctiveness Conflict Risk	Clear niche — DiffDock-L blind diffusion docking — with distinct triggers and an explicit boundary statement ('predicts geometry, not affinity') that steers affinity/co-folding requests elsewhere, so conflict risk with related skills (boltz, chai1) is minimal.	5 / 5
	Total	20 / 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
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: aipoch/open-science
Path: resources/skills/diffdock/SKILL.md
Commit: bf35648

Reviewed: 8 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.