skill-clear-scientific-writing

A Claude Code Skill that turns Claude into a scientific writing editor — for papers, abstracts, methods, results, discussions, figure captions, grants, and thesis chapters.

Most LLM writing help defaults to a generic "academic voice" that's wordy, hedged, and full of phrases like demonstrates the importance of and attracted considerable attention. This skill pushes hard against that default. It diagnoses specific problems (vague nouns, false-good writing, structural issues, citation misuse), prescribes specific fixes, and gives Claude a self-diagnosis loop to apply before returning text.

What it does

• Diagnoses, then edits. Reference files split scientific writing problems into categories — language, structure, content, citations, false-good writing — so Claude can name the specific issue (LANGUAGE.PASSIVE_HEDGE, STRUCTURAL.BURIED_CLAIM) before rewriting.
• Targets "false good writing". A whole reference file dedicated to text that sounds polished but says nothing: "underscores the importance of," "highlights the need for further research," "demonstrates the potential of." The skill flags these and replaces them with specific claims.
• Self-diagnosis loop. Before returning edits, Claude runs a checklist (SELF_DIAGNOSIS.md): are claims specific? are numbers used instead of adjectives? is hedging proportional to confidence? is the structural spine load-bearing?
• Treats craft as separable from grammar. CRAFT.md covers the things grammar checkers miss — sentence rhythm, claim density, when to use a short sentence after a complex one, when to let a number do the work an adjective is trying to do.
• Em-dash discipline. Promoted to its own section after repeated overuse — the skill restricts em-dashes to genuine asides, not as a default break.

Installation

From inside Claude Code:

/plugin marketplace add PEEKPerformer/skill-clear-scientific-writing
/plugin install skill-clear-scientific-writing

No external dependencies — this is a pure prose skill.

Usage

The skill auto-triggers on scientific writing tasks ("draft a methods section", "edit this paragraph", "review this abstract"). You can also invoke it explicitly:

/clear-writing

Paste your text, ask for what you want (draft / edit / review), and the skill takes over.

Repository layout

plugin.json                       Skill metadata (Claude Code plugin manifest)
CHANGELOG.md                      Version history (currently v1.4.3)
commands/
  clear-writing.md                Slash command + operating instructions
references/
  CRAFT.md                        Sentence rhythm, claim density, openings, conclusions
  SELF_DIAGNOSIS.md               Pre-return checklist Claude runs on its own edits
  FALSE_GOOD_WRITING.md           Polished-sounding empty phrases and their replacements
  LANGUAGE_PROBLEMS.md            Passive voice, nominalization, hedging, weasel verbs
  STRUCTURAL_PROBLEMS.md          Buried claims, paragraph spine, topic sentences
  CONTENT_PROBLEMS.md             Vague nouns, missing numbers, undefended claims
  CITATION_PROBLEMS.md            Citation misuse, "smoke citations", missing attribution
  FORMULAS.md                     Reusable templates for common writing moves
  ADAMSON_VOICE.md                Voice study — opening moves, reframes, blunt conclusions
                                  (drawn from published Adamson group papers)

Design notes

• Diagnose before editing. Naming the problem (STRUCTURAL.BURIED_CLAIM, LANGUAGE.NOMINALIZATION) makes feedback transferable. The author can fix the same pattern next time without Claude in the loop.
• Numbers over adjectives. "Stable after centrifugation at 300,000g" beats "remarkably stable" every time. The skill enforces this aggressively.
• One hedge per uncertain claim. "The data suggest X" is enough. Stacked hedges ("the data may possibly suggest that X could potentially") signal nothing except the writer's nervousness.
• No novelty claims by adjective. Don't write "novel"; demonstrate novelty by contrast. Describe what existed, then describe what's different.

Author

Brenden Ferland — Adamson Group, UConn (materials science / graphene composites).