agentleFS
CuiMuxuanCuiMuxuan/academic-codex-skills/skills/academic-de-ai-polishing/SKILL.md

academic-de-ai-polishing

Skill8 starsChanged 5 months agoMITView on GitHub

Use when polishing stable paper, review, thesis, rebuttal, or other scholarly prose that reads mechanically, over-smoothed, obviously AI-assisted, or contaminated by internal project notes, operation logs, process records, TODOs, handoff language, or draft-management traces in the manuscript body. Best after content, structure, evidence, citations, and claim strength are stable, for section-by-section rewriting that preserves meaning, citations, and technical rigor while reducing templated transitions, repetitive cadence, formulaic academic phrasing, and non-manuscript operational residue. Also use when a paragraph needs section-appropriate rhetorical calibration between正文机理密度 and摘要/引言/过渡 compactness. Chinese triggers: 论文降AI痕迹, 去AI味, 去模板化润色, 降低机械感, 学术润色, 最终润色, AIGC痕迹评估, 正文机理升级, 摘要过渡压缩, 内部项目记录, 操作记录痕迹, 主文日志化.

The file

Show its rules and text

We split this skill into the rules it gives an agent

32 rules. About 0 are common, 29 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈29

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill to reduce mechanical or AI-like academic prose while preserving the core argument, evidence balance, citation logic, …

    line 8
  2. Distinctive:

    Remove or recast internal project notes, operation logs, prompt traces, TODO markers, and draft-management residue.

    line 16
  3. Very specific:

    Use polish to bypass missing evidence, unresolved LIT_GAP markers, or unstable manuscript structure.

    line 27
  4. Distinctive:

    Prefer paragraph-level rewrites over whole-section regeneration.

    line 36
  5. Distinctive:

    Do not make prose more confident than the evidence permits.

    line 37
  6. Distinctive:

    Keep Chinese, English, and bilingual terminology stable unless the user asks for translation.

    line 38
  7. Distinctive:

    Treat accurate, direct, concise, understandable writing as the baseline.

    line 40
  8. Distinctive:

    Use the shared cross-disciplinary language review gate before finalizing formal body-text or rebuttal rewrites with abstract terms, local …

    line 42
  9. Distinctive:

    When the target is a正文机理段, allow denser conceptual compression and explicit logic chaining if the evidence and terminology support it.

    line 43
  10. Very specific:

    When called by $revision-control, only return the confirmed local rewrite and hand state updates back to $revision-control.

    line 44
  11. Distinctive:

    Before rewriting, confirm or infer:

    line 48
  12. Distinctive:

    If source text is missing, ask for it.

    line 60
  13. Distinctive:

    Use final-polish-boundary.md before broad final polishing or when the request may hide evidence, structure, or claim-strength work.

    line 78
  14. Distinctive:

    Use high-risk-patterns.md for pattern diagnosis and rewrite-tactics.md for revision tactics.

    line 79
  15. Distinctive:

    Use section-playbooks.md when the rewrite problem is section-dependent.

    line 80
  16. Distinctive:

    Use manuscript-residue-cleanup.md when main text contains project, operation, prompt, or file-management traces.

    line 81
  17. Distinctive:

    Use main-text-and-rebuttal-claim-support-gate.md when final main-text or rebuttal wording contains factual, boundary, subjective, …

    line 82
  18. Distinctive:

    After each rewrite, verify meaning, citations, terminology, claim anchors, support wording, and technical scope.

    line 84
  19. Distinctive:

    Before assessment or rewriting, ask for missing materials that would materially improve quality:

    line 88
  20. Distinctive:

    Ask for source text, target examples, section type, paragraph role, fixed claims, fixed citations, do-not-change terms, allowed edit …

    line 97
  21. Distinctive:

    Read edit-contract-template.md when the edit boundary is unclear.

    line 101
  22. Distinctive:

    Read final-polish-boundary.md when deciding whether the manuscript is ready for final polish or whether the task should route back to …

    line 103
  23. Distinctive:

    Read high-risk-patterns.md when diagnosing mechanical or AI-shaped sentence patterns.

    line 105
  24. Distinctive:

    Read rewrite-tactics.md for paragraph-level rewrite tactics, target-example calibration, and output patterns.

    line 107
  25. Distinctive:

    Read before-after-examples.md when the user wants examples or when calibration is needed.

    line 109
  26. Distinctive:

    Read section-playbooks.md for abstract, introduction, technology, discussion, review, or thesis-section handling.

    line 111
  27. Distinctive:

    Read project-notes-from-this-paper.md for long review-paper polishing sequences.

    line 113
  28. Distinctive:

    Read manuscript-residue-cleanup.md for internal project notes, operation logs, prompt traces, file-management language, handoff language, …

    line 115
  29. Distinctive:

    Read main-text-and-rebuttal-claim-support-gate.md before polishing unsupported main-text or rebuttal claims into final prose, and when …

    line 117
  30. Distinctive:

    Read cross-disciplinary-language-review-gate.md before strict language review or final polishing of formal body text or rebuttal prose that …

    line 119
  31. Very specific:

    Read revision-control-contract.md when a rewrite task arrives from $revision-control.

    line 121
  32. Distinctive:

    Read ai-trace-assessment.md when the user asks for AI-trace or AIGC-risk assessment.

    line 123

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Academic De-AI Polishing

Use this skill to reduce mechanical or AI-like academic prose while preserving the core argument, evidence balance, citation logic, terminology, technical scope, and section-appropriate rhetorical mode.

Boundaries

Do:

  • Rewrite local paragraphs or adjacent paragraph blocks.
  • Reduce templated transitions, repetitive cadence, vague academic filler, and over-smooth balance.
  • Remove or recast internal project notes, operation logs, prompt traces, TODO markers, and draft-management residue.
  • Preserve or strengthen evidence-grounded authorial judgment.
  • Enforce final-polish boundaries after writing, evidence, and structure are stable.
  • Execute confirmed local rewrite tasks from $revision-control when the task is language-level and state boundaries are already set.

Do not:

  • Fabricate evidence, citations, results, or conclusions.
  • Change claim strength beyond what the evidence supports.
  • Claim detector percentages or guarantee AIGC-score outcomes.
  • Regenerate whole sections unless the user explicitly asks.
  • Use polish to bypass missing evidence, unresolved LIT_GAP markers, or unstable manuscript structure.
  • Act as a language-style review skill; route diagnosis-only requests to $language-style-review.
  • Maintain sentence-level revision status, object libraries, latest drafts, or round logs; route those tasks to $revision-control.
  • Handle literature补充, structural reorganization, or large-scale manuscript rewriting as the owner skill.

Core Rules

  1. Preserve technical meaning, evidence balance, citation logic, section purpose, and terminology.
  2. Fix argument, evidence scope, and claim precision before fixing cadence.
  3. Prefer paragraph-level rewrites over whole-section regeneration.
  4. Do not make prose more confident than the evidence permits.
  5. Keep Chinese, English, and bilingual terminology stable unless the user asks for translation.
  6. Move unresolved author decisions to User-review items instead of hiding them in polished prose.
  7. Treat accurate, direct, concise, understandable writing as the baseline; style elevation is optional and must stay evidence-safe.
  8. In main text and rebuttal prose, do not polish unsupported claims into final language; request support or soften the claim.
  9. Use the shared cross-disciplinary language review gate before finalizing formal body-text or rebuttal rewrites with abstract terms, local labels, operation-record residue, unclear sentence purpose, proxy wording, or bilingual inconsistency.
  10. When the target is a正文机理段, allow denser conceptual compression and explicit logic chaining if the evidence and terminology support it; when the target is摘要、引言、结论 or a long transition, keep the chain compact and reduce connector stacking.
  11. When called by $revision-control, only return the confirmed local rewrite and hand state updates back to $revision-control.

Pre-Edit Gate

Before rewriting, confirm or infer:

Source text present:
Section type:
Paragraph role:
Fixed claims and citations:
Claim anchors, if available:
Allowed edit intensity:
Do-not-change terms:

If source text is missing, ask for it. If citation or evidence boundaries are unclear, keep the rewrite conservative and mark author verification needs.

Argument-First Gate

Classify each target paragraph:

StateAction
sound_argument_mechanical_styleRewrite for cadence, vocabulary, transitions, and authorial voice.
overclaiming_or_missing_evidenceFlag the claim problem first, then soften or request evidence before polishing.
unclear_section_jobAsk for section role or propose a local role before rewriting.
citation_boundary_unclearPreserve citations conservatively and mark claims that need author verification.
internal_project_or_operation_traceRemove or recast workflow residue using manuscript-residue-cleanup.md.

Workflow

  1. Identify medium- and high-risk paragraphs.
  2. Diagnose the dominant risk: connector overload, repetitive cadence, templated dialectic, vague vocabulary, missing judgment, overclaiming, or workflow residue.
  3. Use final-polish-boundary.md before broad final polishing or when the request may hide evidence, structure, or claim-strength work.
  4. Use high-risk-patterns.md for pattern diagnosis and rewrite-tactics.md for revision tactics.
  5. Use section-playbooks.md when the rewrite problem is section-dependent.
  6. Use manuscript-residue-cleanup.md when main text contains project, operation, prompt, or file-management traces.
  7. Use main-text-and-rebuttal-claim-support-gate.md when final main-text or rebuttal wording contains factual, boundary, subjective, contribution, or limitation claims.
  8. Route diagnosis-only language review, terminology review, and strict sentence-by-sentence checking to $language-style-review; use cross-disciplinary-language-review-gate.md only as a local safeguard during rewriting.
  9. After each rewrite, verify meaning, citations, terminology, claim anchors, support wording, and technical scope.

Material Request

Before assessment or rewriting, ask for missing materials that would materially improve quality:

Required or high-value materials:
Missing materials:
Can proceed without them: yes/no
Fallback if unavailable:

Ask for source text, target examples, section type, paragraph role, fixed claims, fixed citations, do-not-change terms, allowed edit intensity, and whether the user wants diagnosis only or rewrite.

Reference

Read edit-contract-template.md when the edit boundary is unclear.

Read final-polish-boundary.md when deciding whether the manuscript is ready for final polish or whether the task should route back to writing, research verification, or post-draft review.

Read high-risk-patterns.md when diagnosing mechanical or AI-shaped sentence patterns.

Read rewrite-tactics.md for paragraph-level rewrite tactics, target-example calibration, and output patterns.

Read before-after-examples.md when the user wants examples or when calibration is needed.

Read section-playbooks.md for abstract, introduction, technology, discussion, review, or thesis-section handling.

Read project-notes-from-this-paper.md for long review-paper polishing sequences.

Read manuscript-residue-cleanup.md for internal project notes, operation logs, prompt traces, file-management language, handoff language, TODO markers, or draft residue.

Read main-text-and-rebuttal-claim-support-gate.md before polishing unsupported main-text or rebuttal claims into final prose, and when replacing generic "evidence" wording with a more precise support term.

Read cross-disciplinary-language-review-gate.md before strict language review or final polishing of formal body text or rebuttal prose that may contain abstract wording, undefined local terms, operation-record traces, unclear sentence function, unsupported proxies, citation-distance problems, or bilingual claim-strength drift.

Read revision-control-contract.md when a rewrite task arrives from $revision-control.

Read ai-trace-assessment.md when the user asks for AI-trace or AIGC-risk assessment.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.