Conciseness

The body is dense and expert-level — it assumes knowledge of inverse folding and adds only operational facts (repo-relative imports, checkpoint availability at v_48_010/v_48_020 only, CPU sufficiency, no PyPI dist). A few trimmable spots remain: the intro's explanation of why the full-PDB model places surface hydrophobics ("because many of them are buried at crystallographic or membrane interfaces in the training set") partially duplicates the later note "that is the prior working, not a bug", fitting anchor 4 (efficient, minor over-explanation) rather than the every-token-earns-its-place anchor 5.

Actionability

The running section is copy-paste ready: a complete pip/git/python command sequence with concrete flags (--pdb_path_chains "A", --sampling_temp "0.1", --use_soluble_model), the exact output location and header fields ("out/seqs/<stem>.fa with score= and seq_recovery="), plus concrete remediation ("--omit_AAs \"CW\"", LigandMPNN's --model_type soluble_mpnn) and known failure modes with causes. This matches anchor 5: fully executable commands covering the common cases, exceeding anchor 4's 'minor gaps'.

Workflow Clarity

The single run action is unambiguous, and validation checkpoints are present ("check that the resulting designs still fold (via boltz or esmfold2) before assuming the constraint was free"; rank by an orthogonal predictor "before committing wet-lab slots"), with error-recovery guidance for the two known failure modes. However, the broader campaign flow (design, patch handling, filtering, folding, screening) is presented as caveat sections plus a "Next:" pointer rather than a clearly sequenced workflow, so it fits anchor 4 (clear sequence, most checkpoints, minor gaps) better than anchor 5's explicit ordered validate-fix-retry loop.

Progressive Disclosure

No bundle files exist, so this is a single-file skill; sections are well-organized (Running it, two caveat sections, Next) and shared detail is appropriately deferred to sibling skills ("see proteinmpnn for the parsing quirks", ligandmpnn, boltz, esmfold2) instead of being inlined. It exceeds anchor 3 (no misplaced inline bulk, deferral is clearly signaled) but the body runs ~60 lines — above the sub-50-line compact-skill ideal that would justify anchor 5 with section organization alone — so anchor 4's 'good structure, minor organization gaps' is the best fit.

Specificity

The description states concrete, parameterized actions — "Inverse-fold a backbone with SolubleMPNN — ProteinMPNN retrained on a soluble-PDB subset ... for sequences biased toward cytosolic expression and reduced aggregation" — rather than vague claims. It sits between anchor 3 (1-2 concrete actions, not comprehensive) and anchor 5 (comprehensive multi-action coverage): a single-purpose tool can only list one primary action, but that action is described with specific inputs and outputs, placing it noticeably above the midpoint.

Completeness

It explicitly answers both halves: what ("Inverse-fold a backbone with SolubleMPNN ... for sequences biased toward cytosolic expression and reduced aggregation") and when, via the explicit trigger clause "Reach for this skill when designs from vanilla ProteinMPNN are aggregating ..., when redesigning a membrane-adjacent fold ..., or when an E. coli expression screen is the next step". This matches the anchor-5 example's structure of concrete capability plus concrete trigger phrases, and clearly exceeds anchor 4's 'when could be more specific'.

Trigger Term Quality

Natural user phrases are well covered: "aggregating or going to inclusion bodies", "redesigning a membrane-adjacent fold for soluble expression", "an E. coli expression screen is the next step", plus "vanilla ProteinMPNN". A few natural variants (plain "solubility", "soluble yield", "heterologous expression") are absent, so it does not fully reach the comprehensive-synonym anchor 5.

Distinctiveness Conflict Risk

It carves out a clear niche and actively disambiguates from the closest sibling by naming "vanilla ProteinMPNN" as the alternative and tying triggers to solubility-specific failure modes (aggregation, inclusion bodies, E. coli expression). The triggers are unlikely to fire for a generic structure-design or expression-screening skill, matching the minimal-conflict anchor 5 rather than the minor-overlap anchor 4.