Content
35%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill suffers primarily from extreme verbosity—it reads like a textbook on scientific writing rather than a concise instruction set for Claude. It explains numerous concepts Claude already knows (what bar graphs are for, what active voice means, basic writing principles) and includes encyclopedic sections on field-specific terminology that add little actionable value. The strongest sections are the outline-to-prose conversion example and the LaTeX formatting guidance, but these are buried in hundreds of lines of general writing advice.
Suggestions
Cut the content by 60-70%: Remove sections explaining concepts Claude already knows (basic writing principles, what different chart types are for, field-specific terminology conventions). Focus only on project-specific tools, workflows, and conventions that Claude wouldn't know.
Move the field-specific terminology section (Section 10), detailed figure type descriptions, and writing principles entirely to reference files—these are exactly the kind of content that should be progressively disclosed rather than inlined.
Add explicit validation checkpoints to the manuscript workflow: e.g., 'Run reporting guideline checklist before proceeding to Stage 3' with a concrete command or checklist reference, and 'If checklist reveals gaps, return to the relevant section in Stage 2.'
Consolidate the visual enhancement section: the minimum figure requirements table and extensive lists of when to use each tool could be reduced to a brief rule ('Generate figures liberally using scientific-schematics for technical diagrams and generate-image for visual content') with a reference to detailed guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | This skill is extremely verbose at ~600+ lines, extensively explaining concepts Claude already knows (what IMRAD is, what active voice means, what a scatterplot is, basic writing advice like 'eliminate redundant words'). Entire sections like field-specific terminology (biomedical, chemistry, ecology, physics, neuroscience) are encyclopedic padding that Claude inherently understands. The document explains basic concepts like 'bar graphs: comparing discrete categories' and 'use precise, unambiguous language.' | 1 / 5 |
Actionability | The skill provides some concrete guidance (LaTeX code examples for scientific_report.sty, bash commands for figure generation, example outline-to-prose conversion), but much of the content is descriptive rather than instructive. Many sections are abstract guidance ('establish research context, identify gaps, state objectives') rather than executable steps. The writing process example with the outline-to-prose conversion is genuinely useful and concrete, but it's buried in hundreds of lines of vague advice. | 3 / 5 |
Workflow Clarity | The manuscript development workflow (Stages 1-4) provides a reasonable sequence, but validation checkpoints are mostly implicit ('check logical flow,' 'verify consistency') rather than explicit with concrete verification steps. There are no feedback loops for error recovery—for example, no step says 'if reporting guideline checklist reveals gaps, return to Stage 2 to address them.' The two-stage writing process (outline → prose) is clearly sequenced but lacks validation. | 3 / 5 |
Progressive Disclosure | The skill references multiple external files (references/imrad_structure.md, references/citation_styles.md, etc.) and assets, which is good progressive disclosure in principle. However, no bundle files were provided to verify these exist, and the SKILL.md itself inlines enormous amounts of content that should be in those reference files (e.g., the entire field-specific terminology section, detailed LaTeX examples, figure generation guidance). The references section at the end properly signals external files, but the body doesn't actually defer to them—it duplicates and expands on the content. | 3 / 5 |
Total | 10 / 20 Passed |