proteinmpnn

Inverse-fold a protein backbone (PDB structure) into amino-acid sequence with ProteinMPNN (Dauparas et al. 2022, github.com/dauparas/ProteinMPNN). Reach for this skill to run sequence design on RFdiffusion backbones, to redesign one chain of a PDB while holding interface residues fixed, or to generate a temperature-swept set of sequences for downstream folding.

Quality

91%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

90%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 exemplary single-file skill: dense with genuinely non-obvious operational knowledge, fully executable commands, and hard-won gotchas (quoting semantics, the outer PDB-stem key in fixed_positions_jsonl) presented as compact tables. The only real improvements are an explicit output-verification step after a run and a decision on whether the two tables belong in a reference file.

Suggestions

Add an explicit post-run validation step — e.g., "check the score= and seq_recovery= headers in out/seqs/<stem>.fa, and confirm fixed positions were held" — to close the workflow-clarity gap.

If the skill grows, move the checkpoint-selection and error tables into a references/ file and keep SKILL.md as a lean overview with clearly signaled links.

Dimension	Reasoning	Score
Conciseness	Every section earns its tokens with non-obvious, repo-specific knowledge: the string-parsing quirks of --sampling_temp and --pdb_path_chains, the silent fixed-positions failure mode, the checkpoint-noise table, and the error table. Nothing re-explains concepts Claude already knows — the first paragraph's model background exists only to route between proteinmpnn/ligandmpnn/solublempnn, which is decision-relevant rather than padding.	5 / 5
Actionability	The skill provides a copy-paste-ready clone-and-run command block, the exact output location (out/seqs/<pdb_stem>.fa), a concrete chain-diagnostic grep (grep '^ATOM' file.pdb \| cut -c22 \| sort -u), and the named helper script make_fixed_positions_dict.py — fully executable guidance covering the common cases.	5 / 5
Workflow Clarity	The run → interpret output → recover-from-errors flow is clear, and the silent-redesign gotcha plus the errors table give real feedback loops (including the silent all-positions-redesigned case). It falls short of the top anchor only because there is no explicit post-run verification step, e.g., checking score=/seq_recovery= headers or confirming fixed positions took effect before proceeding.	4 / 5
Progressive Disclosure	No bundle files exist and the ~73-line body is well-sectioned and self-contained with nothing that clearly demands a separate file. It sits just past the under-50-line simple-skill exception and the checkpoint/error tables are borderline candidates for a reference file, so structure is good but not the textbook clear-overview-plus-signaled-references shape.	4 / 5
	Total	18 / 20 Passed

Description

92%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 strong description: concrete actions, explicit use-when triggers, and a distinct niche with negligible conflict risk. The only gap is minor keyword coverage — a .pdb extension mention and a couple of everyday synonyms would make the triggers fully comprehensive.

Suggestions

Add the .pdb file extension and a common synonym such as "design sequences for a backbone" to broaden natural trigger coverage.

Optionally name the sibling tools (ligandmpnn for ligand/metal interfaces) in the description so users with non-protein interfaces route correctly at trigger time.

Dimension	Reasoning	Score
Specificity	The description lists multiple concrete, distinct actions — "Inverse-fold a protein backbone (PDB structure) into amino-acid sequence", "redesign one chain of a PDB while holding interface residues fixed", "generate a temperature-swept set of sequences for downstream folding" — with no generic filler, matching the comprehensive anchor.	5 / 5
Completeness	It explicitly answers what ("Inverse-fold a protein backbone (PDB structure) into amino-acid sequence with ProteinMPNN") and when ("Reach for this skill to run sequence design on RFdiffusion backbones, to redesign one chain...") with concrete trigger phrases, matching the top anchor.	5 / 5
Trigger Term Quality	Natural domain phrasings like "inverse-fold", "protein backbone", "sequence design", "RFdiffusion backbones", and "temperature-swept" give good keyword coverage, but the .pdb file extension and common synonyms (e.g., "design sequences for this backbone") are missing, so it falls just short of the comprehensive anchor.	4 / 5
Distinctiveness Conflict Risk	It occupies a clear niche — ProteinMPNN inverse folding of protein backbones — with trigger contexts (RFdiffusion backbones, fixed-interface redesign, temperature sweeps) that would not plausibly fire a different skill.	5 / 5
	Total	19 / 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/proteinmpnn/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.