Content
38%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.
The skill's core content (inputs, five-step PubMed workflow, Vancouver citation quality rules, NCBI credentials) is sound, but it is buried in ~130 lines of generic boilerplate with duplicated and conflicting validation sections, and every reference to the implementing scripts points at files absent from the bundle. The body reads as padded template output rather than lean, executable guidance.
Suggestions
Cut the generic boilerplate sections (When to Use, Key Features, Recommended Workflow, Output Contract, Validation and Safety Rules, Failure Handling, Required Inputs, Deterministic Output Rules, Completion Checklist) — they restate what Claude already knows and duplicate each other; keep only the skill-specific Inputs, Workflow Process, Quality Rules, and Usage sections.
Merge the two conflicting validation sections ("Validation Shortcut" running `scripts/validate_skill.py --help` and "Quick Validation" running `python scripts/pubmed_api.py --help`) into one canonical pre-flight check, and include the concrete end-to-end invocation instead of the vague "Then invoke the skill with disease and molecule parameters".
Fix broken references: `scripts/pubmed_api.py`, `scripts/validate_skill.py`, and the `cd "20260316/scientific-skills/Academic Writing/molecular-review-workflow"` path do not exist in this bundle — either ship the scripts or replace the references with inline, self-contained commands; also correct "See `## Usage` above" in Example Usage, since the Usage section appears later in the file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose with several padded boilerplate sections that restate what Claude already knows: "Validate the request against the skill boundary", "Keep the output safe, reproducible", a "Key Features" section that only re-quotes the description, and duplicated generic checklists ("Recommended Workflow", "Output Contract", "Validation and Safety Rules", "Completion Checklist"). It avoids the severely-verbose score-1 pattern of explaining basic external concepts, but the padding is substantial. | 2 / 5 |
Actionability | Some concrete guidance exists (NCBI env-var exports, `python scripts/pubmed_api.py --help`, required inputs disease/molecule, explicit Vancouver citation rules), but there is no runnable end-to-end example, "Then invoke the skill with disease and molecule parameters" gives no actual invocation, and the referenced scripts (`scripts/pubmed_api.py`, `scripts/validate_skill.py`) do not exist in the bundle, so the documented commands are not executable as shipped. This lands at concrete-but-incomplete rather than the mostly-executable guidance of a 4. | 3 / 5 |
Workflow Clarity | A clear 5-step sequence is listed (Input Translation, Search Term Generation, PubMed Search, Result Processing, Review Generation), but validation is scattered and contradictory: "Validation Shortcut" runs `scripts/validate_skill.py --help` while "Quick Validation" runs `python scripts/pubmed_api.py --help`, with no concrete error-recovery loop, and actual execution is delegated to scripts missing from the bundle. Checkpoints exist but are duplicated and implicit, matching the steps-listed-but-checkpoints-missing anchor rather than the clean sequence of a 4. | 3 / 5 |
Progressive Disclosure | The bundle contains no scripts/, references/, or assets/ directories, yet the body points to `scripts/pubmed_api.py`, `scripts/validate_skill.py`, "1 additional script(s)", and a `cd "20260316/scientific-skills/..."` path — none of which exist, so navigation to the real implementation is impossible. All content is inlined monolithically in SKILL.md with generic filler that belongs in no file, matching the minimal-structure/broken-references anchor of 2. | 2 / 5 |
Total | 10 / 20 Passed |