360 Secure SkillHub
360 安全技能中心

Essay Humanizer

基于 Qwen3-8B 微调的 LoRA 适配器与语料统计权重，把 AI 起草的议论文改写为更接近人类学术写作的连续纯文本，并保留作者立场与引用；需要 Apple Silicon macOS 与 MLX，首次运行下载基础模型，可选自建 HTTP 服务供远程调用。

kevin0818-lxd
kevin0818-lxd
热度 50.32
最近更新 2026-05-11 17:34:09
分类：效率依赖：代码执行 / 读写文件 / 内部调用大模型 / 第三方 API应用场景：校对润色 / 工作流

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add essay-humanizer
Metadata
name
essay-humanizer
description
Rewrite AI-drafted essays into more human-like academic prose. Fine-tuned LoRA over Qwen3-8B guided by 24 Wikipedia-style AI-writing pattern weights plus MDD/ADD syntactic targets from CAWSE/LOCNESS vs DeepSeek baselines. Includes trained LoRA adapter and inference script. Requires Apple Silicon macOS with MLX. Optional FastAPI host for MCP/tool linking. Orchestrator: output plain text only (no LaTeX dollar delimiters).

Essay Humanizer (corpus-informed)

Rewrites AI-generated argumentative/academic essays toward human baseline style informed by CAWSE (M/D bands) LOCNESS, and contrast with DeepSeek-generated counterparts. Ships with a fine-tuned LoRA adapter (9.3 MB) and inference script.

Skill contract

Component Path Notes
Inference script scripts/inference.py Entry point — humanize() function or CLI
LoRA adapters assets/adapters/adapters.safetensors.json 12.3 MB base64 JSON; auto-decoded to binary on first run
Pattern weights data/analysis/weights.json Corpus-derived, loaded by inference at runtime
Decoder scripts/decode_adapters.py Reconstructs .safetensors binary from JSON (auto or manual)
Installer scripts/install_deps.sh One-time: pip install mlx mlx-lm transformers + decode
Base model Qwen/Qwen3-8B-MLX-4bit Downloaded from HuggingFace on first run (~4.5 GB, cached)

Requirements: Apple Silicon macOS with Python 3.9+.

Quick Start

bash scripts/install_deps.sh          # one-time: installs deps + decodes adapter
python scripts/inference.py --file draft.txt   # adapter auto-decodes if not already done

Or from Python:

from scripts.inference import humanize
print(humanize("Your AI-drafted essay text here..."))

Weighted pattern table (descending priority)

When humanizing, address higher-weight rows first. Weights are data-driven from corpus analysis (Mann-Whitney); zero-weight rows were not statistically significant.

ID Weight Category Pattern
P06_CLICHE_METAPHORS 0.1358 vocabulary Cliche metaphors
P15_EM_DASH_OVERKILL 0.1358 punctuation Em dash overkill
P21_MARKDOWN_ARTIFACTS 0.1358 formatting Markdown artifacts
P23_TEXTBOOK_BOLDING 0.1358 formatting Textbook bolding
P12_PRESENT_PARTICIPLE_TAIL 0.1133 rhetorical Present participle tailing
P10_RULE_OF_THREES 0.0806 rhetorical Rule of threes
P04_AI_VOCABULARY 0.0621 vocabulary AI vocabulary
P14_COMPULSIVE_SUMMARIES 0.0598 rhetorical Compulsive summaries
P05_EXCESSIVE_ADVERBS 0.0540 vocabulary Excessive adverbs
P13_OVER_ATTRIBUTION 0.0529 rhetorical Over-attribution
P11_FALSE_RANGES 0.0341 rhetorical False ranges
P17_TRANSITION_OVERUSE 0.0001 punctuation Overuse of transition words
P01_UNDUE_EMPHASIS 0.0000 content Undue emphasis
P02_SUPERFICIAL_ANALYSIS 0.0000 content Superficial analysis
P03_REGRESSION_TO_MEAN 0.0000 content Regression to the mean
P07_REDUNDANT_MODIFIERS 0.0000 vocabulary Redundant modifiers
P08_FILLER_HEDGING 0.0000 vocabulary Filler hedging
P09_NEGATIVE_PARALLELISM 0.0000 rhetorical Negative parallelisms
P16_EN_DASH_AVOIDANCE 0.0000 punctuation En dash / hyphen misuse for ranges
P18_COLLABORATIVE_REGISTER 0.0000 register Collaborative register
P19_LETTER_FORMALITY 0.0000 register Letter-style formality
P20_INSTRUCTIONAL_CONDESCENSION 0.0000 register Instructional condescension
P22_EXCESSIVE_LISTS 0.0000 formatting Excessive bulleted/numbered lists
P24_EMOJI_SYMBOL 0.0000 formatting Emoji/symbol injection

Syntactic complexity (MDD / ADD advisory)

Human Merit / Distinction-range writing in CAWSE often shows variable mean dependency distance (MDD); AI prose may cluster more tightly. When humanizing:

  • Reference MDD means from analysis: human ~2.333775514332394, AI ~2.4553791855163483.
  • Variance ratio (human/AI) ~1.7153931408079544: prefer natural mix of shorter and longer dependency links, not uniformly smoothed sentences.
  • Avoid flattening every sentence to minimal dependency length; that can read as a different kind of machine polish.

Mandatory rule (orchestrator)

  1. Output continuous prose suitable for submission (no chat-signoffs, no "hope this helps").
  2. Plain text only for math if any — no raw $$ LaTeX unless user explicitly requests LaTeX.
  3. Preserve author stance and citations if present; do not fabricate references.

Hosted HTTP API (optional, for non-Mac or remote use)

For non-Apple-Silicon machines or multi-user deployments, run the optional FastAPI server on a Mac host and connect via HTTP/OpenAPI:

  1. Install: pip install fastapi uvicorn[standard]
  2. Run: uvicorn api.main:app --host 0.0.0.0 --port 8765 (set HUMANIZE_API_KEY env var for auth)
  3. Point MCP / OpenAPI tools at https://<your-host>/openapi.json
  4. Call POST /v1/humanize with JSON {"text":"..."} (+ Authorization: Bearer …)

See references/hosted_api.md for details.

References