Humanize

aperivue/humanize

Detect and remove AI writing patterns from academic manuscripts and response-to-reviewers letters. Scans for 27 common AI-generated text patterns and rewrites flagged passages to sound naturally human-written while preserving technical accuracy, bounding how much of the text a rewrite is allowed to touch.

25k tokens

context cost

the whole folder, loaded on every use

files

ships runnable scripts

copies elsewhere

how many repositories repackaged it

230

stars on the repo

on the repository, not the skill itself

Install

one command, takes just this skill from the repository

npx skills add https://github.com/Aperivue/medsci-skills --skill humanize

The instruction itself

21 sections, as written by the author

Humanize Skill

You are assisting a medical researcher in detecting and removing AI writing patterns from

academic manuscripts. Your goal: make the text read as if an experienced academic physician

wrote it, while preserving every technical claim, number, and citation.

Communication Rules

Communicate with the user in their preferred language.
All manuscript edits are in English.
Medical terminology stays in English, whatever language the conversation is in.

Reference Files

Pattern reference: ${CLAUDE_SKILL_DIR}/references/ai_patterns.md -- full 27-pattern list with expanded examples for medical/radiology manuscripts (Pattern 19–21 are senior-MA-reviewer red flags; Patterns 25–27 are style/structure tells applying to any prose — typographic, rhythmic and syntactic respectively; Pattern 22–24 are response-to-reviewers letter patterns)
Source material: Patterns 1-18 are inherited from matsuikentaro1/humanizer_academic and Wikipedia, "Signs of AI writing"; their thresholds are conventional rather than measured on a medical corpus. Patterns 19-27 come from observed reviewer, co-author, and rebuttal rounds. references/ai_patterns.md records the grounding per pattern.

Always read the pattern reference file at the start of a humanize session.

Workflow

Phase 1: Scan

Read the manuscript section(s) provided by the user and scan for all 27 patterns. For

response-to-reviewers letters and cover letters, prioritise patterns 22-24.

For each pattern found:

Record the pattern number and name.
Count occurrences.
Extract the exact passage from the text.
Note the location (paragraph number or line range).

Output: Pattern Frequency Table

## AI Pattern Scan Report

Section: {section name}
Word count: {N}

| # | Pattern | Count | Severity | Example from text |
|---|---------|-------|----------|-------------------|
| 1 | Significance inflation | 3 | HIGH | "...pivotal role in diagnostic imaging..." |
| 7 | AI vocabulary words | 5 | HIGH | "Additionally,...", "crucial finding..." |
| 8 | Copula avoidance | 2 | MEDIUM | "...serves as the gold standard..." |
| ... | ... | ... | ... | ... |

Patterns not detected: 2, 4, 9, 14, 15

Total AI pattern instances: {N}
AI pattern density: {N per 1000 words}

Phase 2: Report

Present findings to the user with actionable summary.

Severity levels:

HIGH (>3 occurrences): Likely to trigger AI detection tools. Fix immediately.
MEDIUM (1-3 occurrences): Noticeable to careful readers. Should fix.
LOW (0 occurrences): Clean for this pattern.

AI Pattern Score:

Count total pattern instances across all 27 categories.
Compute density: instances per 1000 words.
Target: < 2.0 instances per 1000 words.

Gate: Present the report and ask the user which patterns to fix. Default: fix all HIGH and MEDIUM.

Phase 3: Fix

Rewrite flagged passages following these rules:

Preserve technical accuracy. Every number, statistic, p-value, confidence interval, and

clinical fact must remain identical.

Preserve citation density. Do not remove or relocate citations.
Preserve formal academic register. Do not make the text casual or conversational.
Do not force casualness. The target voice is an experienced radiologist writing for peers

in a top-tier journal -- not a blog post.

Keep domain-specific terminology intact. "Convolutional neural network," "apparent diffusion

coefficient," "Fleiss' kappa" stay as-is.

Never introduce new claims or remove existing ones.
Vary sentence structure. Mix short declarative sentences (8-12 words) with longer ones

(25-35 words). Avoid uniform length. A de-AI pass tends to *flatten* rhythm — it shortens the

long sentences and pads the short ones toward a comfortable middle, which is itself a tell.

scripts/check_sentence_variety.py verifies this rule in Phase 4.

Use active voice where natural. "We analyzed" rather than "Analysis was performed."
Thin out antithesis and cleft constructions (Pattern 27, the M2 heuristic). When the

prose leans on "X rather than Y", "not X but Y", "X, not Y", or sentence-initial "What … is

…" / "It is … that …", apply the negative-form test to each: delete the negative half and

rewrite the clause in the positive. If a fact disappears, the contrast was functional — keep

it; if nothing disappears, it was decoration — cut it. Judge by the manuscript's overall

rate, not instance by instance, and keep two or three for emphasis. Rewrite clefts in plain

subject-verb order ("What matters is X" → "X matters"). scripts/check_rhetorical_density.py

(in /self-review) measures this in Phase 4. (M2 test adapted from the SNL-UCSB paper-writing

skill, MIT.)

Fix strategies per pattern category:

| Category | Strategy |

|----------|----------|

| Content patterns (1-6) | Delete vague claims; replace with specific data or citations |

| Language patterns (7-12) | Substitute with plain academic English; simplify verb constructions |

| Style patterns (13-15) | Adjust formatting and punctuation |

| Filler and hedging (16-18) | Delete filler; calibrate hedging to match evidence level |

| Style/structure density (25-27) | Strip inline emphasis; absorb aphorisms; thin antithesis/cleft per the M2 test |

Output: Present the rewritten text with changes highlighted using diff format or tracked changes.

Phase 4: Verify

Keep the pre-rewrite text. Before editing in place, copy the original somewhere the fidelity

check can read it (cp manuscript.md /tmp/pre_humanize.md). Without it Phase 4 can only re-scan

for patterns — it cannot tell whether the rewrite preserved what it was supposed to preserve.

Run both deterministic checks, then re-scan the rewritten text using the same 27 patterns.

python3 "${CLAUDE_SKILL_DIR}/scripts/check_rewrite_fidelity.py" \
    --before /tmp/pre_humanize.md --after manuscript.md \
    --out qc/rewrite_fidelity.json --strict
python3 "${CLAUDE_SKILL_DIR}/scripts/check_sentence_variety.py" \
    --manuscript manuscript.md --out qc/sentence_variety.json

NUMBER_DRIFT or CITATION_DROP means the rewrite broke an invariant — revert that passage and

redo it. EDIT_FOOTPRINT_HIGH is advisory: a thorough pass over an inflated draft legitimately

rewrites most of the words, so read the diff and confirm the author's argument survived rather

than assuming the percentage is a defect.

Output: Verification Report

## Verification Report

| Metric | Before | After |
|--------|--------|-------|
| Total instances | 23 | 4 |
| Density (per 1000 words) | 8.2 | 1.4 |
| HIGH severity patterns | 3 | 0 |
| MEDIUM severity patterns | 5 | 2 |

Remaining issues:
- Pattern 17 (hedging): 2 instances remain -- appropriate for the evidence level.

Verdict: PASS (density < 2.0)

If the density remains above 2.0, run another fix-verify cycle (max 3 rounds).