Conciseness

Every section carries non-obvious, decision-relevant knowledge — the numpy 1.24 sed patch with rationale, exact residue-token grammar ("A45 A46 B10", insertion codes "B82A"), ProDy compiler gotcha, and the silent ligand-drop failure mode. Not 4: no paragraph can be trimmed without losing actionable content; there is no explanation of concepts Claude already knows.

Actionability

Fully executable: copy-paste install line, clone, sed patch, checkpoint fetch, and a complete run.py invocation with real flags; plus an error table mapping exact messages to fixes and a description of the output file layout. Not 4: the common path has no gaps.

Workflow Clarity

The run sequence (install → clone → patch → fetch checkpoints → run → inspect outputs) is unambiguous and there is an explicit validation loop ("If ligand_confidence in the FASTA header is missing or zero across every design, the model never saw the ligand — fix the input, do not trust the sequences"), so the batch-operation cap does not apply. Not 5: the validation checkpoints live in separate troubleshooting sections rather than interleaved into an ordered validate-then-proceed flow.

Progressive Disclosure

No bundle files exist; the ~100-line body is entirely at overview level, cleanly sectioned with meaningful headers, and contains nothing that belongs in a separate file. Cross-links to sibling skills (proteinmpnn, solublempnn, boltz, chai1) are one level deep and clearly signaled. Not 4: navigation is easy and nothing is buried.

Specificity

Lists multiple concrete actions in third person — "Inverse-fold a backbone with ligand, nucleic-acid, and metal context", "redesign the residues lining a binding pocket around a bound small molecule or cofactor", "design metal-coordinating sites", "get threaded designed-sequence PDBs" — with comprehensive coverage of the tool's capabilities. Not 4: no meaningful coverage gap; the actions span the full surface of the skill.

Completeness

Explicitly answers both questions: what ("Inverse-fold a backbone with ligand, nucleic-acid, and metal context using LigandMPNN") and when ("Reach for this skill to redesign the residues lining a binding pocket… or to design metal-coordinating sites… or to get threaded designed-sequence PDBs") with three concrete trigger scenarios. Not 4: the when-clause is explicit and specific, not merely present.

Trigger Term Quality

Natural domain phrases a user would actually say are present with synonyms and format terms: "binding pocket", "bound small molecule", "cofactor", "metal-coordinating sites", "backbone", "ligand", "nucleic-acid", "metal", "PDB", "MPNN". Not 4: synonyms (ligand/cofactor/small molecule) and concrete artifacts (PDBs) are covered, not just a few terms.

Distinctiveness Conflict Risk

LigandMPNN names a clear niche with distinct triggers (ligand/NA/metal context), but "get threaded designed-sequence PDBs out of any MPNN run" deliberately claims territory that overlaps sibling proteinmpnn/solublempnn skills. Not 5: that overlap is real even if intentional; not 3: the ligand-context trigger is unmistakable and would not fire for generic skills.