Review academic language style without modifying manuscript state. Use when the user asks for language style review, wording checks, grammar, AI-like phrasing, templated expression, Chinese-English translation accuracy, figure/table text wording, citation sentence style, sentence-level ambiguity, terminology acceptability, connector density by section, or operation-record residue checks in a paper, thesis, dissertation, review, rebuttal, chapter, section, paragraph, or sentence. May run independently or as an upstream report for revision-control or academic-paper-orchestrator. Chinese triggers: 语言风格审查, 检查语言, 用词检查, 语法检查, AI味, 模板化表达, 中文译文准确性, 图表文字, 引用句式, 逐句语言审查, 逻辑连接词, 正文机理段, 摘要过渡段.
Review academic language style without modifying manuscript state. Use when the user asks for language style review, wording checks, grammar, AI-like phrasing, templated expression, Chinese-English translation accuracy, figure/table text wording, citation sentence style, sentence-level ambiguity, terminology acceptability, connector density by section, or operation-record residue checks in a paper, thesis, dissertation, review, rebuttal, chapter, section, paragraph, or sentence. May run independently or as an upstream report for revision-control or academic-paper-orchestrator. Chinese triggers: 语言风格审查, 检查语言, 用词检查, 语法检查, AI味, 模板化表达, 中文译文准确性, 图表文字, 引用句式, 逐句语言审查, 逻辑连接词, 正文机理段, 摘要过渡段.
Language Style Review
Use this skill to audit scholarly wording and return scored findings, localized issues, candidate rewrites, and a handoff queue. It does not write the manuscript, update sentence status, or maintain revision state.
Boundaries
Do:
Review language at manuscript, chapter, section, paragraph, sentence, figure/table text, or citation-sentence level.
Score clarity, academic expression, terminology, and residue/mechanical-pattern risk.
Locate problems by object id or sentence when ids are available.
Provide candidate rewrites as suggestions only.
Explain why accepted professional terms or proper nouns should remain unchanged.
Create candidate project-standard items when the user asks to record a preference.
Distinguish between正文机理段 and摘要/引言/过渡段 when judging connector density, conceptual compression, and figure narration style.
Do not:
Update the latest draft, object library, sentence pass/fail state, revision count, or round directory.
Write to project supplemental review standards.
Perform full manuscript quality or benchmark readiness review; route to $post-manuscript-benchmark-review.
Perform formal sentence-level revision management; route to $revision-control.
Final-polish stable text as the primary task; route accepted rewrite execution to $academic-de-ai-polishing when appropriate.
Verify DOI, title, citation authenticity, or evidence sufficiency as final authority.
Core Rules
Return review output, not official modifications.
Respect user-specified disliked words, phrases, sentence rhythm, and translation concerns as high-priority signals.
Keep field-specific terms if they are accepted professional terminology; explain the reason instead of rewriting them away.
Flag operation records, prompt traces, project-management traces, TODOs, and file workflow residue in manuscript prose.
Separate language problems from evidence, structure, citation-identity, or manuscript-readiness problems.
When a problem implies structural rewriting, missing literature, or object-state changes, create a handoff item for $revision-control.
Never add a project rule directly; produce candidate_rules for $revision-control to confirm with the user.
Workflow
Identify scope, available object ids, manuscript language, field, target style, and any project standards path.
Use review-workflow.md to choose independent, orchestrated, or revision-control handoff mode.
Use scoring-and-output-schema.md for four-dimension scoring, issue tables, accepted-term notes, candidate rewrites, and handoff queues.
Use project-standards-handling.md when project supplemental standards, terminology lists, problem-word lists, or user preferences are available.
If benchmark-quality judgment is requested, route the quality portion to $post-manuscript-benchmark-review before language review.
If official modification is requested, send the localized task queue to $revision-control.
Output Contract
Return:
review scope and source limits;
project standards used, if any;
four dimension scores to one decimal place;
main issues;
sentence/object-level issue table;
accepted terms/proper nouns that should remain;
candidate rewrites marked as non-final;
suggested $revision-control task queue;
candidate_rules when the user asks to record a preference.
Reference
Read review-workflow.md for activation modes, sequencing with benchmark review, and handoff boundaries.
