Most content earns its place (exact tool signatures, SOAP/list quirks, fallback paths, console-noise caveats Claude would not know), but inline 'Why X matters' mini-essays and the 8-row ideal-range table explain ADMET significance that could be tightened. Not 3 because some explanatory prose could be trimmed; not 1 because it avoids basic-concept padding.

Provides copy-paste-ready tool calls with exact signatures (e.g. 'ADMETAI_predict_BBB_penetrance(smiles=["<SMILES>"])'), list-wrapping reminders, expected output fields, and concrete fallbacks. Not 2 because the guidance is fully executable rather than pseudocode.

A 5-phase sequenced workflow with an ASCII flow diagram, per-phase goals/steps, fallbacks, and a MANDATORY completeness checklist acting as an explicit validation checkpoint before reporting. Not 2 because validation and failure-handling steps are explicit, not merely implied.

Single monolithic ~310-line file with no bundle directories or external references, and large inline reference-style tables (8-row ideal-range table, 7 toxicity endpoints) that arguably belong in a separate reference file. Not 3 because the simple-skill carve-out does not apply to this long, multi-task skill; not 1 because sections are clearly organized rather than a wall of nested references.

Names concrete actions and integrations ('Integrates ADMET-AI predictions, SwissADME drug-likeness, PubChemTox experimental toxicity, ChEMBL clinical data, Lipinski rule-of-five, and CYP interaction data'), matching the multi-specific-action anchor; not the level below, which only names a domain and some actions.

Explicitly states what ('Comprehensive ADMET ... profiling for drug candidates') and when ('Use for ... before lab testing'), satisfying both halves; not 2 because the 'Use for' trigger clause is present and explicit.

The 'Use for ... BBB penetration, bioavailability, hepatotoxicity prediction, ADME/PK profiling, or screening compound libraries' clause covers natural terms users would say; not 2 because common variations are well represented rather than partial.

Clear ADMET/drug-likeness niche with distinct triggers unlikely to fire for unrelated skills; not 2 because the domain and triggers are specific enough to avoid overlap with general cheminformatics skills.