Scientific Revision Skill

This skill provides four tools for document review:

1. Citation Checker — cross-checks in-text citations against the bibliography
2. CrossRef Verifier — checks whether bibliography entries are real publications
3. Scholar Verifier — checks entries against Google Scholar (via scholarly) with Semantic Scholar as fallback
4. Writing Quality Checker — detects AI slop patterns, vague language, and terminology inconsistencies

Works with any document that has a bibliography/references section. Supports APA, MLA, Chicago, and language-native citation styles in English, German, French, and Italian — parenthetical and narrative styles.

Scripts

All scripts live in .agents/skills/scientific-revision/scripts/. Pass the document path as the first argument; all scripts default to docs/essay/essay.md if omitted.

1. Citation Checker

python .agents/skills/scientific-revision/scripts/check_citations.py [path/to/document.md]

• Finds citations in text missing from bibliography — and suggests alternatives from the bibliography that might be the intended reference.
• Finds bibliography entries never cited in the text.

2. Bibliography Verification (CrossRef)

python .agents/skills/scientific-revision/scripts/verify_crossref.py [path/to/document.md]

• Queries the CrossRef API to verify each bibliography entry is a real publication.
• Generates a citation_report.md in the same directory as the input file.
• Books and software get medium confidence by default (not indexed in CrossRef).

3. Bibliography Verification (Google Scholar / Semantic Scholar)

python .agents/skills/scientific-revision/scripts/verify_scholar.py [path/to/document.md]

• Uses the Semantic Scholar API via httpx (no library dependency issues).
• Optionally cross-checks with Google Scholar via scholarly (install with uv add scholarly).
• Also reports citation counts, useful to gauge publication impact.
• Generates a citation_report_scholar.md in the same directory as the input file.
• Complement CrossRef: use both to catch entries that appear in one index but not the other.
• Set SEMANTIC_SCHOLAR_API_KEY env var for a free key that raises the rate limit.

4. Writing Quality Checker

python .agents/skills/scientific-revision/scripts/check_writing.py [path/to/document.md]

• Flags AI slop phrases (e.g., "delve into", "it is worth noting", "seamlessly", "leverage").
• Detects terminology inconsistencies (e.g., mixing "dataset" and "data set").
• Flags repetitive sentence/paragraph starters.
• Reports line numbers so issues are easy to locate.

Supported Section Headers

The scripts auto-detect bibliography sections with any of these headings: References, Bibliography, Works Cited (EN) · Literaturverzeichnis, Literatur (DE) · Bibliographie, Références (FR) · Bibliografia, Riferimenti bibliografici, Opere citate (IT)

Workflow

When the user asks you to revise or verify their document, follow this sequence unless instructed otherwise:

1. Check Citations: Run check_citations.py. For each missing citation, review the suggested alternatives and ask the user if one of them is the intended reference. Report uncited bibliography entries too.

2. Verify References (CrossRef): Run verify_crossref.py. Share the generated citation_report.md with the user, highlighting any low-confidence entries that may be hallucinated or incorrect.

3. Verify References (Scholar): Run verify_scholar.py. Cross-reference the citation_report_scholar.md with the CrossRef results.

   • An entry flagged low by CrossRef but verified by Scholar (or vice-versa) warrants closer manual review.
   • Highlight citation counts for entries with zero citations — these may be obscure or fabricated.

4. Check Writing Quality: Run check_writing.py. For each flagged phrase or inconsistency, propose a concrete rewrite. Don't just list problems — offer the user improved versions of the flagged sentences.

After all four checks, give the user a concise summary of:

• Citation issues found and resolved
• Bibliography entries that could not be verified by either CrossRef or Scholar
• Any discrepancies between CrossRef and Scholar results
• Writing quality issues and suggested improvements