Content
71%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 skill body with an explicit seven-step workflow, strong input/output validation gates, and exemplary reference-module organization. Its main weakness is redundancy: the no-fabrication and scope-boundary rules are restated across five or six sections, inflating token cost without adding guidance, and the body lacks a worked search-string or filled-record example.
Suggestions
Consolidate the repeated anti-fabrication and "not final inclusion / not screening" rules into the single Hard Rules section, and drop the duplicated lists in "This skill must never", "What This Skill Should Not Do", and "Quality Standard".
Add one worked example — an adapted PubMed search string and a filled-out sample candidate record showing the required field format — either inline or in the adaptation/normalization reference modules.
Trim the inline A–J output-section descriptions to names plus a pointer, since references/output-section-guidance.md already expands each section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The workflow steps themselves are lean and directive, but the same constraints are restated many times: anti-fabrication rules appear in "This skill must never", the Step 5 "Hard verification rule", "Hard Rules 2–4", "What This Skill Should Not Do", and "Quality Standard", and the "not final inclusion / not screening" boundary is repeated roughly five times. Not 2 because nothing explains concepts Claude already knows and every section is instruction rather than padded prose; not 4 because the cross-section repetition is extensive — the body could be cut nearly in half without losing guidance. | 3 / 5 |
Actionability | Concrete, executable instruction throughout: named databases (PubMed, Google Scholar, Web of Science, Embase...), an exact record field list (title, authors, PMID and/or DOI, evidence status), explicit tier labels (Tier 1/2/3/P), ten mandatory output sections A–J, an exact fallback phrase "DOI not available / not verified", and sample user inputs. Not 5 because the body never shows a worked example of an actual adapted search string or a filled-out example record; not 3 because the guidance is specific and directly executable, not high-level hints or pseudocode. | 4 / 5 |
Workflow Clarity | Seven steps are explicitly ordered ("always run in order"), each mapped to a reference module, with an input-validation gate and a verbatim redirect script for out-of-scope requests, plus a hard verification checkpoint in Step 5 ("If verification is incomplete, say so explicitly instead of filling placeholders") and vague-input handling ("If the user's question is too vague... recommend or trigger prior question clarification"). Not 5 because failure handling is "state it" only — no recovery/retry loop or verification mechanism (e.g., how to actually verify a DOI or link) is specified; not 3 because checkpoints are explicit and present at both the input and output boundaries. | 4 / 5 |
Progressive Disclosure | The body is an overview-plus-workflow-contract that points to nine real one-level-deep reference modules, each introduced with a descriptive purpose line ("Database choice and coverage logic → ...", "Search-term construction and logic expansion → ...") and tied to a specific workflow step; all nine referenced files exist in references/. Not 4 because navigation is genuinely clear, references are well signaled rather than buried, and the detailed per-topic rules are appropriately split out of the overview. | 5 / 5 |
Total | 16 / 20 Passed |