Content
78%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, information-dense skill body: executable core patterns, explicit validation checkpoints, honest gotchas, and a clean one-level-deep reference bundle. The main improvements are removing duplicated examples and caveats for token efficiency, making the illustrative folder/docking snippets self-contained, and surfacing the batch feedback-loop guidance (currently in a reference) as an explicit validate-fix-retry step in the body.
Suggestions
Consolidate the duplicated aspirin descriptors example (Quick start vs. Core usage pattern) into one canonical example and state the "hosted examples are illustrative, not authenticated" caveat once to save tokens (conciseness).
Make the Folders docking snippet self-contained by defining or clearly constructing `protein`, `pocket`, and `ligand` (e.g., via `retrieve_protein` and `Molecule.from_smiles`) so it is copy-paste ready (actionability).
Add an explicit validate-and-retry step for batch campaigns in the main body — e.g., check each submitted workflow's status on retrieval and resubmit or inspect failures — rather than deferring feedback loops entirely to references/batch_and_webhooks.md (workflow_clarity).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious, API-specific gotchas ("Folder.model_rebuild" workaround, "MW is exact/monoisotopic mass, not average molecular weight", pagination and non-recursive-listing semantics) and assumes Claude's competence throughout. Minor trimming is possible: the aspirin descriptors example appears in both Quick start and Core usage pattern, and the "hosted examples are illustrative" caveat is repeated several times. Fits the efficient-with-minor-trimmable-instances anchor; not 5 due to this duplication, not 3 because there is no padding explaining concepts Claude already knows. | 4 / 5 |
Actionability | Mostly executable, concrete guidance: pinned install command, auth setup with verification via whoami(), submission calls with typed result access, and SMILES validation. Minor gaps remain — the Folders section submits with undefined `protein`, `pocket`, and `ligand` variables, which is labeled illustrative but not copy-paste ready. Matches the mostly-executable anchor; not 5 because a few snippets cannot be run as written. | 4 / 5 |
Workflow Clarity | The submit → wait → retrieve pattern is clearly sequenced, and explicit validation checkpoints exist: SMILES validation with a raised ValueError, a None-check on `best_taut` before chaining workflows, redocking a known ligand to validate docking, and verifying PDB accessions. However, the main body stops short of explicit fix-and-retry feedback loops for batch/destructive operations (those are deferred to references/batch_and_webhooks.md). Fits the clear-sequence-with-most-checkpoints anchor; not 5 because feedback loops are implicit rather than spelled out. | 4 / 5 |
Progressive Disclosure | The SKILL.md body keeps core usage, decision trees, and common pitfalls inline while splitting the workflow catalog, batch/webhooks, pricing, a worked example, and troubleshooting into five real, clearly signaled one-level-deep reference files. Each reference section summarizes exactly what it contains, and the bundle contains no nested (2+ level) references. Matches the clear-overview, well-signaled, appropriately split anchor. | 5 / 5 |
Total | 17 / 20 Passed |