foldseek-structural-search

Performs 3D structural searches of proteins against various databases (PDB, AlphaFold, CATH, MGnify, etc.) using the Foldseek API. Use ONLY when the user provides a physical 3D coordinate file (.cif, .mmcif, or .pdb) and wants to find structurally similar proteins. Do NOT use if the user only provides a protein sequence, gene name, or UniProt ID.

Quality

83%

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

77%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.

A largely actionable, well-sequenced workflow with executable commands and explicit validation gates. It loses points on conciseness due to duplicated rules and on progressive disclosure because the citation.bib reference is orphaned and no detail files are signaled.

Suggestions

Remove the duplicated file-requirement rule — keep it in either 'Core Rules' or Instructions step 1, not both — and merge the overlapping 'Do Not Parse the JSON' / 'No Raw Parsing' bullets to tighten conciseness.

Either reference references/citation.bib in the body (e.g., a 'Citation' note pointing to it for result attribution) or remove the orphaned file, so bundle structure is consistently signaled.

Integrate the dangling error-handling note (lines 97–99) into the numbered workflow as a proper validation/retry step rather than a stray line outside the list.

Dimension	Reasoning	Score
Conciseness	Mostly efficient and task-specific, but the file-type requirement is stated twice ('File Requirement' in Core Rules and 'Strict Input Validation' in Instructions) and 'Do Not Parse the JSON' / 'No Raw Parsing' overlap, so it cannot reach the lean every-token-earns-its-place level 3.	2 / 3
Actionability	Provides copy-paste-ready executable commands ('uv run scripts/search.py <path-to-file> -o <generated-filename.json> > <generated-filename.md>' and the --databases variant), an explicit database allowlist, and concrete file-naming examples, matching the fully-executable top anchor; not the pseudocode level 2.	3 / 3
Workflow Clarity	A clearly sequenced 9-step workflow with explicit validation gates ('halt immediately' on invalid input or unsupported database) and an API-error feedback path; the operation is a read-only search so the destructive/batch cap does not apply, reaching the clear-sequence-with-validation anchor.	3 / 3
Progressive Disclosure	The body is well-sectioned and correctly invokes scripts/search.py one level deep, but everything is inline with no navigation signaling and references/citation.bib exists as a bundle file yet is never referenced, so it falls short of the well-signaled one-level-deep-references top anchor.	2 / 3
	Total	10 / 12 Passed

Description

90%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.

A strong, third-person description that clearly states the capability, gives explicit positive and negative use-when triggers, and lists natural trigger terms. Its only weakness is that it describes one primary action rather than enumerating multiple concrete actions.

Dimension	Reasoning	Score
Specificity	Names the concrete action ('Performs 3D structural searches of proteins against various databases ... using the Foldseek API') and the tool/databases, but it is a single primary action rather than the multiple distinct actions the top anchor calls for, so it sits at 2 rather than 3.	2 / 3
Completeness	Explicitly answers both 'what' (Performs 3D structural searches ...) and 'when' ('Use ONLY when the user provides a physical 3D coordinate file ... and wants to find structurally similar proteins'), plus a negative trigger, matching the top anchor's what-and-when-with-explicit-triggers.	3 / 3
Trigger Term Quality	Covers natural terms a user would say — '3D structural searches', 'structurally similar proteins', '.cif, .mmcif, or .pdb', 'protein sequence, gene name, or UniProt ID' — giving good coverage of how the request actually surfaces; not the level-2 'Works with PDF files' style partial coverage.	3 / 3
Distinctiveness Conflict Risk	A clear niche (Foldseek structural search on coordinate files) with a distinct positive trigger and an explicit 'Do NOT use if the user only provides a protein sequence, gene name, or UniProt ID' guard, making conflict with sequence-based skills unlikely; well above the 'could still overlap' level-2 anchor.	3 / 3
	Total	11 / 12 Passed

Validation

100%

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 — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: google-deepmind/science-skills
Path: skills/foldseek_structural_search/SKILL.md
Commit: 0b42509

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.