Skip to content
Back to search

academic-humanizer

Improve the clarity and voice of AI-assisted academic writing (papers, theses, rebuttals) and funding proposals (NSF Project Summary/Description, NIH Specific Aims): preserve scholarly conventions, match claims to evidence (and, for proposals, claims to feasibility), and match the author's own voice. It never changes a number, result, or citation, and it is not for evading AI-use disclosure. Use when editing AI-assisted academic prose or grant proposals.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/aiscientists-dev-academic-humanizer/badge.svg)](/listings/aiscientists-dev-academic-humanizer)

HTML

<a href="/listings/aiscientists-dev-academic-humanizer"><img src="/api/listings/aiscientists-dev-academic-humanizer/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

git clone https://github.com/AIScientists-Dev/academic-humanizer ~/.claude/skills/academic-humanizer

It is a plain `SKILL.md` plus examples, so it also runs as a skill or system prompt for **Codex** and **MorphMind**. Point your agent at `SKILL.md`.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 261 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. ---
  2. name: academic-humanizer
  3. version: 0.3.3
  4. description: |
  5. Improve the clarity and voice of AI-assisted academic writing (papers, theses, rebuttals) and
  6. funding proposals (NSF Project Summary/Description, NIH Specific Aims): preserve scholarly
  7. conventions, match claims to evidence (and, for proposals, claims to feasibility), and match the
  8. author's own voice. It never changes a number, result, or citation, and it is not for evading
  9. AI-use disclosure. Use when editing AI-assisted academic prose or grant proposals.
  10. license: MIT
  11. compatibility: claude-code codex morphmind opencode
  12. allowed-tools: [Read, Write, Edit, Grep, Glob, AskUserQuestion]
  13. ---
  14. # Academic Humanizer
  15. Improve the clarity and voice of AI-assisted *academic* writing while keeping the precise,
  16. evidence-bound voice that scholarship requires and matching the author's own style. It preserves every
  17. number, result, and citation, and it is not a tool for evading AI-use disclosure.
  18. ## When to use
  19. Editing or reviewing academic prose: paper sections, abstracts, rebuttals, related work, and **funding
  20. proposals** (NSF Project Summary/Description, NIH Specific Aims, fellowship and foundation proposals;
  21. see Layer 6). **Not** for blogs, marketing, or personal essays, and **never** inject opinion, humor, or
  22. first-person "personality" into a manuscript. For technical writing, neutral and precise *is* the human
  23. voice. One caveat for proposals: their register is different from a paper's, since they are sold on
  24. vision and feasibility, so the ambition language a paper would trim is appropriate there; apply Layer 6,
  25. not the paper layers' stricter trimming, to vision statements.
  26. ## Core principle
  27. Academic writing already has a correct human voice: neutral, precise, third-person plural ("we"), every
  28. claim tied to its evidence. The job is to (1) strip the AI *tells* without casualizing, and (2) enforce
  29. the discipline a general humanizer misses: **every claim earns its number, figure, or citation, and no
  30. verb is stronger than its evidence.**
  31. ## Process
  32. 1. **Read** the manuscript and any author writing sample; note the document type (paper vs. funding
  33. proposal) and the target venue or funding agency. For proposals, also apply Layer 6 and preserve
  34. appropriate vision.
  35. 2. **Audit** (do not edit yet): list each detected pattern with its location and proposed fix, and each
  36. empirical claim's evidence status.
  37. 3. **Rewrite**: same structure and content, all claims and citations preserved, tells removed, over-claims
  38. matched to evidence, legitimate hedging kept.
  39. 4. **Report**: cleaned text plus a short change log (patterns removed, claims softened or given evidence
  40. pointers, voice notes). Cover everything the original covered: if it had five paragraphs, so does the
  41. rewrite.
  42. ---
  43. ## Layer 1: General AI-tell catalog
  44. Scan for and fix the general patterns, subject to the academic exceptions in Layer 3:
  45. inflated significance ("marking a pivotal moment"); superficial "-ing" tails that fake depth
  46. ("..., highlighting..."); promotional/figurative language ("rich", "vibrant", "groundbreaking");
  47. vague attributions ("experts argue" with no cite); AI vocabulary (*delve, underscore, intricate,
  48. tapestry, testament, landscape (abstract), pivotal, showcase, foster, leverage (filler), realm,
  49. seamless*); copula avoidance ("serves as" -> "is"); negative parallelisms ("not just X, but Y");
  50. rule-of-three padding; elegant variation (cycling synonyms for one referent); filler
  51. ("it is worth noting that", "in order to"); **overlong, clause-stacked sentences (split them; see 2.11)**;
  52. and **em-dashes (remove entirely; recast with commas, colons, parentheses, or separate sentences)**.
  53. **Before:** *Additionally, an enduring testament to the method's value is its ability to delve into
  54. intricate dependencies, showcasing a seamless integration that underscores its pivotal role.*
  55. **After:** *The method also captures higher-order dependencies, which the baselines miss (Table 2).*
  56. ---
  57. ## Layer 2: Academic AI tells (remove or fix)
  58. ### 2.1 Over-claiming verbs
  59. Empirical work *shows* and *provides evidence*; it does not *prove* or *demonstrate* universal truths.
  60. **Watch:** demonstrate, prove, establish, confirm, guarantee; "significantly" with no test/number.
  61. **Before:** *We prove that our method significantly outperforms all prior approaches.*
  62. **After:** *Our method improves held-out accuracy by 4--7 points over the strongest prior approach
  63. (Table 3); the gain is significant at p < 0.01 by a paired test.*
  64. ### 2.2 Significance hype
  65. **Watch:** paves the way for, a crucial/pivotal step toward, has the potential to revolutionize, opens
  66. new avenues, sheds light on, of paramount importance, bridges the gap.
  67. **Before:** *This work paves the way for a new paradigm and sheds light on a problem of paramount
  68. importance.*
  69. **After:** *This work addresses one failure mode of prior methods: error accumulation under long-horizon
  70. rollout (Section 4).*
  71. ### 2.3 Empty intensifiers
  72. **Watch:** extensive/comprehensive/thorough experiments, a wide range of, numerous, various.
  73. **Before:** *We conduct extensive experiments on a wide range of datasets.*
  74. **After:** *We evaluate on three datasets (ImageNet, CIFAR-100, and iNaturalist).*
  75. ### 2.4 Novelty padding
  76. **Watch:** "novel" used more than once per section; "to the best of our knowledge"; "for the first time".
  77. **Before:** *We propose a novel framework and, to the best of our knowledge, are the first to study this.*
  78. **After:** *We study online calibration under delayed labels, which prior calibration work (offline) does
  79. not address.*
  80. ### 2.5 Formulaic openers
  81. **Watch:** "In recent years, X has attracted increasing attention"; "With the rapid development of...";
  82. "Despite recent advances,...".
  83. **Before:** *In recent years, tabular deep learning has attracted increasing attention.*
  84. **After:** *Tabular deep learning has a structural limitation: most models discard feature-type metadata
  85. and must relearn it from data.*
  86. ### 2.6 Connective overuse
  87. Do not start consecutive sentences with Moreover/Furthermore/Additionally/In particular; let logic carry.
  88. **Before:** *Moreover, the method is fast. Furthermore, it is simple. Additionally, it scales.*
  89. **After:** *The method is fast and simple, and it scales to one million rows (Section 5).*
  90. ### 2.7 Contribution-list cliches
  91. Each contribution names a *specific* result, not a restatement of the abstract.
  92. **Before:** *Our contributions are: (1) a novel method; (2) extensive experiments; (3) strong results.*
  93. **After:** *We (1) introduce a metadata-aware encoder that reaches 0.91 AUROC vs 0.86 for the strongest
  94. baseline; (2) show it stays within 2 points under 20% label noise where the baseline drops 9; (3) release
  95. the benchmark.*
  96. ### 2.8 Citation dumping
  97. Cite the one or two works that matter and say why, not a bracketed list.
  98. **Before:** *Many methods exist [3, 7, 9, 12, 15].*
  99. **After:** *The closest prior method is TabNet [7], which encodes all features jointly; we instead
  100. condition on feature-type metadata.*
  101. ### 2.9 Hedging-by-vagueness
  102. **Watch:** somewhat, relatively, fairly, to some extent, quite. Quantify or cut.
  103. **Before:** *Performance is somewhat better and relatively robust.*
  104. **After:** *Accuracy is 3 points higher and varies by less than 1 point across five seeds.*
  105. ### 2.10 Boilerplate emphasis
  106. **Watch:** "It is worth noting that", "It should be emphasized that", "Notably,", "Importantly,".
  107. If it matters, the sentence shows it.
  108. **Before:** *It is worth noting that, importantly, the gain holds across scenarios.*
  109. **After:** *The gain holds across all three scenarios (Table 4).*
  110. ### 2.11 Overlong, clause-stacked sentences
  111. AI favors long sentences that chain three or four clauses with commas and "which", "that", "while", "with".
  112. Split them: one idea per sentence, and cut subordinate clauses that carry no weight. **Watch:** sentences
  113. past ~30 words, or with 3+ subordinate clauses.
  114. **Before:** *Existing methods, though promising, are largely empirical, with unclear principles
  115. underpinning their behavior, which limits their reliability and further progress.*
  116. **After:** *Existing methods stay empirical. Their principles are unclear, which limits reliability and
  117. progress.*
  118. ---
  119. ## Layer 3: Preserve these (do NOT over-correct)
  120. A general humanizer flattens legitimate scholarly constructs. Keep them.
  121. - **Evidence-tied hedging is correct and required.** Keep "suggests", "is consistent with", "we
  122. hypothesize that", "may indicate", "appears to" when the claim is genuinely uncertain.
  123. *Wrong fix:* turning *"the results suggest X"* into *"the results prove X"*: this manufactures
  124. over-claiming. Keep the calibrated verb.
  125. - **Passive voice** is fine when the actor is irrelevant: *"Samples were normalized to total protein."*
  126. - **First-person plural "we"** is standard; do not rewrite to avoid it.
  127. - **Semicolons and an occasional triple** are fine in moderation. Em-dashes are the exception: remove
  128. them entirely (Layer 1), recasting with commas, colons, parentheses, or separate sentences.
  129. - **Formal definitions, named methods/metrics, technical terms, equations, and symbols** stay verbatim.
  130. - **Never invent, drop, or alter a number, equation, or citation.** Same content; preserve every cite key.
  131. ---
  132. ## Layer 4: Claim-evidence discipline
  133. For every empirical claim, check (a) is it backed by a number, figure, table, or citation in the text,
  134. and (b) does the verb match the strength of that evidence?
  135. - **Unbacked claim -> add the evidence pointer or soften.**
  136. *Before:* *Our method is more robust.* *After:* *Our method's accuracy drops by 2 points under
  137. distribution shift, versus 11 points for the baseline (Figure 3).*
  138. - **Verb stronger than evidence -> downgrade.**
  139. *Before:* *This demonstrates that our method is universally superior.*
  140. *After:* *On these three datasets, our method matches or exceeds the strongest baseline (Table 2).*
  141. - **Vague magnitude -> a number or RANGE, attributed.**
  142. *Before:* *a large improvement.* *After:* *a 2--6% improvement in balanced accuracy over the strongest
  143. baseline.*
  144. Prefer ranges (e.g., "2--6%") over single averaged values unless the averaging method is stated, and
  145. attribute each number to its method, metric, and baseline. When comparing, lead with the comparison
  146. against the strongest competitor, not the trivial baseline.
  147. ---
  148. ## Layer 5: Voice and venue matching
  149. If the author supplies prior papers, read a sample first and note sentence rhythm, connective habits,
  150. level and placement of hedging, how they open sections, notation, and recurring phrasings, then match
  151. them. Match the venue's register too (e.g., ICLR/NeurIPS: terse, direct, results-forward; Nature/PNAS:
  152. more expository). Absent a sample, default to clean, precise, venue-appropriate prose, not the casual,
  153. opinionated voice of a general-purpose humanizer.
  154. ## Layer 6: Funding-proposal mode (NSF, NIH)
  155. A proposal is not a paper. It is sold on **vision plus feasibility**, not on finished results, and
  156. reviewers score it. The register shift matters: ambition language that the paper layers would trim
  157. ("long-term goal", "pioneer", "transformative", "establish a foundation") is *appropriate and expected*
  158. here, provided a credible plan and evidence back it. So in proposal mode, **do not flatten the vision**;
  159. enforce a different discipline instead: **claim <-> feasibility**.
  160. ### 6.1 Know the structure; the score lives in the first pages
  161. Reviewers form a score from the opening, then skim the rest to confirm it. Put most editing effort there.
  162. - **NSF.** A one-page **Project Summary** with the three review-criteria heads spelled out:
  163. **Overview**, **Intellectual Merit**, **Broader Impacts**, each self-contained. The Project
  164. Description then opens with **long-term vision -> this proposal's goal -> the gap -> the specific
  165. thrusts/aims -> the payoff**, ideally within the first 1--2 pages, with one overview figure. Broader
  166. Impacts must be substantive and integrated, never an afterthought.
  167. - **NIH (R01).** The **Specific Aims page is the whole proposal in one page**, and is the most-read,
  168. most-decisive page. Standard arc: (1) opening: the problem, what is known, the **gap / critical need**;
  169. (2) the **long-term goal** and the **central hypothesis** with its rationale; (3) "**The objective of
  170. this application is...**" plus how the hypothesis was formed; (4) **2--3 Aims**, each a one-line goal +
  171. a phrase on approach + the expected outcome; (5) a **payoff** paragraph: what changes if it succeeds.
  172. Then **Significance, Innovation, Approach** as separately scored sections.
  173. ### 6.2 First-3-pages primacy (edit these hardest)
  174. By the end of page 1 (NIH Aims) or pages ~2--3 (NSF), the reader must already hold: the **hook** (why it
  175. matters, concretely), the **gap** (what is missing and the cost of the gap), the **central idea** (your
  176. approach in one sentence), the **aims/thrusts** (crisp and parallel), and the **payoff**. If any is
  177. missing or buried, fix that before touching later sections. A reviewer unconvinced by page 3 does not
  178. recover on page 10.
  179. ### 6.3 Proposal-specific weak moves to fix
  180. - **Vague importance.** *Watch:* "this is an important/timely problem", "X has many applications".
  181. **Before:** *Understanding this problem is critically important.*
  182. **After:** *Without bounds on how measurement noise propagates to diagnosis, clinical models are tuned by
  183. trial and error, the inefficiency this proposal removes.*
  184. - **Method-as-aim** (an aim naming a technique instead of a question or outcome).
  185. **Before:** *Aim 2: Apply transfer learning to the dataset.*
  186. **After:** *Aim 2: Determine whether fusing wearable and lab signals improves early detection, and for
  187. which patient subgroups it helps or hurts.*
  188. - **Dominoed aims** (Aim 2/3 collapse if Aim 1 fails; reviewers flag this as fragile). *Fix:* phrase
  189. aims as **parallel and independently valuable**; where one depends on another, state the fallback.
  190. - **Ambition without feasibility.** Every bold claim needs a footing: preliminary data, a prior
  191. result/publication, a classical theorem you build on, or a collaborator/letter. *Fix:* attach the
  192. evidence beside the claim ("our preliminary result in Fig. X shows...", "building on a classical minimax
  193. lower bound...").
  194. - **Boilerplate Broader Impacts / training plan.** *Watch:* "we will mentor students and disseminate via
  195. talks and papers." *Fix:* make it concrete, enumerated, and tied to the research: specific programs,
  196. named courses or tools, measurable outreach.
  197. - **Hedged central hypothesis.** The Aims-page hypothesis is a falsifiable commitment, not "we will
  198. explore whether possibly...". Calibrated hedging belongs in the Approach's interpretation, not the
  199. central claim.
  200. ### 6.4 Preserve and deploy (funded-proposal craft)
  201. These read as strength; keep or add them rather than editing them out.
  202. - **Vision/ambition framing**: a bold long-term goal up front, with this proposal as one principled step
  203. toward it.
  204. - **Run-in lead-ins for scannability**: bold/italic **Goal:**, **Motivation:**, **Innovation:**,
  205. *Thrust/Aim N (one-line mission):*. Reviewers skim; visible structure earns time.
  206. - **A concrete running example or protagonist** to make an abstract method vivid and consistent across aims.
  207. - **Sharp challenge/aim statements posed as questions**: a crisp open question reads as a well-posed
  208. problem (a boxed or set-off question per aim works well).
  209. - **Anchoring novel work in deep, named classical results** to signal rigor and lineage: a known
  210. inequality, capacity notion, or test that the new method generalizes.
  211. - **Foreground the team's standing as feasibility evidence**: prior funded work, preliminary results,
  212. publications, collaborators, and demonstration partners belong *early*, as proof the plan is executable.
  213. A real track record is evidence, not boasting; place it where it de-risks the aims. *(Use only the PI's
  214. own real, supplied record; never invent funding, results, partners, or letters.)*
  215. ### 6.5 Claim <-> feasibility (the proposal analog of Layer 4)
  216. For every aim and promised outcome, check: is the ambition matched by a credible means, such as
  217. preliminary data, a prior method, a classical foundation, a collaborator, or staged de-risking? If yes,
  218. keep the ambitious verb. If no, attach the missing evidence or scale the claim to what the plan supports.
  219. Never invent preliminary results, prior funding, partners, or letters; if the support does not exist, flag
  220. the gap for the author rather than papering over it.
  221. ---
  222. ## Output
  223. Return the cleaned text plus a short change report: patterns removed (by type), claims softened or given
  224. evidence pointers, and any voice/venue notes. Confirm that no number, equation, or citation was altered.
README.md · README · 121 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. <div align="center">
  2. <img src="assets/banner.svg" alt="Academic Humanizer: personalized editing for AI-assisted academic drafts, keeping your voice and every claim, number, and citation intact" width="860">
  3. [![license](https://img.shields.io/badge/license-MIT-2f8f57?style=flat-square)](LICENSE)
  4. &nbsp;![version](https://img.shields.io/badge/version-0.3.2-2f8f57?style=flat-square)
  5. &nbsp;![skill](https://img.shields.io/badge/skill-papers_and_grant_proposals-1c1a15?style=flat-square)
  6. &nbsp;![built by](https://img.shields.io/badge/built_by-NSF,_CAREER,_NIH_R01-555?style=flat-square)
  7. </div>
  8. ## Why we built this
  9. Some of us write a lot of papers and grant proposals, and our team started using AI to help with
  10. drafts. The problem is that AI-assisted drafts come out generic and verbose, with "In recent years..."
  11. openers, inflated phrasing, and over-long sentences. They also drift from the author's own voice and
  12. lose the precision scholarship depends on.
  13. There are tools called "humanizers," but they are built for blogs and marketing. Run one on a paper or
  14. an NSF proposal and it flattens the precision along with everything else. The careful wording academic
  15. writing depends on is the first thing to go.
  16. So we put together our own. To calibrate it, we had the AI compare its own drafts with our team's
  17. accepted papers and funded proposals, and we went through the differences by hand. It is nothing fancy,
  18. and it is not about gaming review, defeating detectors, or adding fake novelty. We wanted AI-assisted
  19. drafts to read clearly and in the author's own voice, with the numbers, citations, and claims left
  20. exactly as written.
  21. ## Ethics and disclosure
  22. This is an editing aid for clarity and voice, calibrated to an author's own prior accepted work. It does
  23. not generate findings, invent data, or change citations, and it is not designed to evade AI-use
  24. detection. Using it does not remove your obligation to disclose AI assistance: always follow the
  25. disclosure policy of the venue you submit to.
  26. ## See it work
  27. > [!CAUTION]
  28. > **Before** (a generic AI draft):
  29. >
  30. > In recent years, continual learning has attracted increasing attention and achieved remarkable
  31. > success. However, existing methods still face crucial challenges. In this proposal, we propose a novel
  32. > framework that leverages cutting-edge techniques to delve into these intricate problems, paving the way
  33. > for a transformative paradigm that will revolutionize the field.
  34. > [!TIP]
  35. > **After** (clear, in the author's voice, with claims tied to evidence):
  36. >
  37. > Continual learning matters, but today's methods stay empirical and their principles are unclear. That
  38. > limits reliability and progress. This proposal builds a principled framework on three fronts:
  39. > adaptation, soft supervision, and cross-domain knowledge. We demonstrate it on autonomous driving and
  40. > network management.
  41. **More before/after passes** are in [`examples/before-after.md`](examples/before-after.md): a general
  42. example, an NIH Specific Aims page, and a funded NSF CAREER summary.
  43. ---
  44. ## What it does
  45. - **Sharpens clarity and voice:** trims generic AI phrasing ("paves the way", "extensive experiments", "to the best of our knowledge", "In recent years...", delve/underscore/tapestry, rule-of-three, very long sentences, em-dashes) and brings the draft closer to the author's own style.
  46. - **Keeps claims tied to evidence:** no verb stronger than the data (`prove` → `show empirically`), and
  47. vague magnitudes become attributed ranges.
  48. - **Leaves real scholarship alone:** evidence-tied hedging, passive voice where it fits, `we`,
  49. definitions, symbols, and every citation. It doesn't change a number or a reference.
  50. - **Has a separate mode for grant proposals (NSF, NIH):** it keeps the vision a paper would trim, and
  51. spends most of the effort on the first pages, since that's what reviewers score.
  52. ## Install
  53. ```bash
  54. git clone https://github.com/AIScientists-Dev/academic-humanizer ~/.claude/skills/academic-humanizer
  55. ```
  56. It is a plain `SKILL.md` plus examples, so it also runs as a skill or system prompt for **Codex** and
  57. **MorphMind**. Point your agent at `SKILL.md`.
  58. ## Use
  59. ```
  60. /academic-humanizer
  61. [paste a section, or point at main.tex]
  62. # optionally: "match my voice from prior_paper.pdf; target venue: ICLR"
  63. ```
  64. ## Make it yours
  65. The rules here reflect one group's voice. Fork the repo and adapt them to your own: point it at a few of
  66. your past papers, keep the checks that fit your field, and adjust the rest. It is meant to be
  67. personalized, not a one-size-fits-all filter.
  68. ## How it works
  69. Six layers: general AI-tell catalog → academic-specific tells → preserve scholarly conventions →
  70. claim↔evidence matching → voice/venue calibration → funding-proposal mode (NSF/NIH structure,
  71. first-page primacy, claim↔feasibility). The audit→rewrite loop is defined in [`SKILL.md`](SKILL.md).
  72. ## References
  73. Layer 6 distills the *stable* structure of NSF and NIH proposals. For current, binding requirements
  74. (page limits, formatting, deadlines), consult the source:
  75. - NSF: [Proposal & Award Policies & Procedures Guide (PAPPG)](https://www.nsf.gov/policies/pappg)
  76. - NSF: [CAREER program](https://new.nsf.gov/funding/opportunities/career-faculty-early-career-development-program)
  77. - NIH: [Write Your Application](https://grants.nih.gov/grants/how-to-apply-application-guide/format-and-write/write-your-application.htm) (Specific Aims, Significance, Innovation, Approach)
  78. ## Acknowledgments
  79. - **[blader/humanizer](https://github.com/blader/humanizer)** (MIT). *Focus:* removing general
  80. AI-writing patterns for blog, casual, and encyclopedic text. This skill reuses its general AI-tell
  81. catalog (Layer 1) and extends it for academic prose.
  82. - **[koaeraser/ARMS](https://github.com/koaeraser/ARMS)**. *Focus:* an autonomous pipeline for
  83. statistics/methodology research papers (idea → validated, revised manuscript). A complementary,
  84. broader-scope project that informed the claim-evidence and numerical-precision emphasis here.
  85. This skill is the narrower piece: a single-purpose **editing pass** that sharpens clarity and matches
  86. claims to evidence while preserving the author's scholarly voice.
  87. ## License
  88. MIT.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.