Content
67%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 body is well structured with executable commands for every task and genuine validation checkpoint sections, but it carries substantial generic template boilerplate and never documents the CONFIG keys that constitute the skill's actual usage path. Trimming boilerplate and surfacing script configuration would lift it considerably.
Suggestions
Remove or collapse the generic template sections (Required Inputs, Output Contract, Failure Handling, User Checkpoints, Input/Quick Validation) into a few skill-specific lines — the Required Inputs table's "PMID, PDF, DOCX" examples are unrelated to spreadsheet operations and add pure token cost.
Document each script's CONFIG keys (input_file, output_file, sheet_name, target ranges, formula_template, etc.) inline in the Example Usage section, so users can configure tasks without opening each script to discover its parameters.
Link reference files at point of use (e.g., column-matching.md and dedup-conflict.md under the merge workflow, formulas.md under the formulas section) instead of a flat optional list, and reference the orphaned report-template.md or remove it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Task-relevant sections (When to Use, Key Features, Example Usage, Implementation Details) are tight, but ~50 lines of generic template boilerplate (Required Inputs with unrelated "PMID, PDF, DOCX" examples, Output Contract, Failure Handling, User Checkpoints, Input/Quick Validation) add tokens without skill-specific value. | 3 / 5 |
Actionability | Six concrete runnable commands each invoke a real bundled script that executes as-is with bundled defaults. However, the body instructs "configure parameters in CONFIG" without documenting any CONFIG key, so every real task requires opening a script to learn its configuration. | 4 / 5 |
Workflow Clarity | A clear four-step sequence (confirm inputs → choose task type → run script → produce output) is paired with checkpoint guidance in User Checkpoints (batch/overwrite confirmation) and Quick Validation (paths exist, output fields, assumptions marked). Validation lives in separate sections rather than embedded as explicit workflow steps with feedback loops, keeping it below 5. | 4 / 5 |
Progressive Disclosure | The body is well sectioned, all 8 referenced files exist in references/ and are one level deep with topic labels, and scripts map to tasks. Minor gaps: references are a flat "optional" bullet list rather than linked at point of use, and references/report-template.md is never referenced. | 4 / 5 |
Total | 15 / 20 Passed |