Conciseness

Every section carries non-obvious operational knowledge: flag parsing quirks ('--sampling_temp is parsed as a space-separated string... commas never split'), the silent fixed-positions failure mode, checkpoint selection by training noise, and a compact error table. Nothing explains what proteins or inverse folding are; no padding anywhere.

Actionability

The run block is copy-paste executable end to end (pip install, shallow clone, full command with flags), output location and header fields are spelled out ('out/seqs/<pdb_stem>.fa', 'score=', 'global_score=', 'seq_recovery='), and the error table prescribes exact diagnostic commands like "grep '^ATOM' file.pdb \| cut -c22 \| sort -u".

Workflow Clarity

The sequence (install → clone → run → read outputs → fold downstream) is clear, and the 'Errors worth recognizing' table gives recovery guidance for the main failure modes. It stops short of explicit validation checkpoints — e.g. there is no proactive step to confirm fixed positions were actually held or to sanity-check seq_recovery before spending compute on folding — matching 'most checkpoints present; minor validation gaps'.

Progressive Disclosure

This is a self-contained single-file skill (no references/, scripts/, or assets/ bundle exists) of ~70 lines, and all content — one run command, two flag gotchas, a checkpoint table, an error table — is appropriately inline rather than split out. Sections are well-headed and navigable, satisfying the simple-skill exception for a skill with no need for external references.

Specificity

Lists multiple concrete actions — 'Inverse-fold a protein backbone (PDB structure) into amino-acid sequence', 'run sequence design on RFdiffusion backbones', 'redesign one chain of a PDB while holding interface residues fixed', 'generate a temperature-swept set of sequences' — with comprehensive coverage of the tool's use cases. All are specific, executable capabilities rather than generic claims.

Completeness

Explicitly answers both: what — inverse-folding a backbone into an amino-acid sequence with ProteinMPNN — and when — 'Reach for this skill to run sequence design on RFdiffusion backbones, to redesign one chain of a PDB... or to generate a temperature-swept set of sequences'. Concrete trigger phrases are attached to each use case.

Trigger Term Quality

Strong natural phrasing users would say: 'sequence design', 'redesign one chain of a PDB', 'RFdiffusion backbones', 'temperature-swept', 'downstream folding', plus the tool name and citation. A few common synonyms are missing (e.g. 'protein design', 'inverse folding' unhyphenated, 'fixed residues'), keeping it just below the comprehensive anchor.

Distinctiveness Conflict Risk

A clear niche (protein-only inverse folding with named triggers like RFdiffusion backbones and interface-residue fixing) with minimal conflict risk. Minor overlap with closely related skills (ligandmpnn for ligand interfaces, solublempnn) that the description does not explicitly disambiguate, which matches the 'mostly distinct; minor overlap risk' anchor.