通过实证方法检测模型会重复使用的默认值，然后将这些发现转化为可重用的指令文件，使未来的输出不那么泛泛。

把这个发给你的经纪人

复制并发送这句话：

代理可读的指令在线：skills/unslop/SKILL.md

它的作用

每个模型都有其默认的方向。在写作中，这意味着相同的开场、过渡和修辞动作。在视觉作品中，这意味着相同的落地页结构、相同的渐变、相同的卡片网格、相同的CTA区块。

unslop 测量这些默认值，而不是猜测。

1. 你给它一个域名。
2. 它生成提示集，并用 Claude 代码生成大量样本。
3. 对于视觉工作，它会生成截图，以便分析页面实际呈现的样子。
4. 它分析整个集合中的重复模式。
5. 它写入 skill.md]，一个专注于下次应避免内容的文件。
6. 它运行前后对比，这样你可以看到配置文件是否真的改变了结果。

主要输出是一个可以重复使用的 Claude Code、CLAUDE.md、Codex、光标规则或任何系统提示符的 markdown 文件。

快速开始

unslop

Empirically detect the repetitive defaults a model falls back on, then turn those findings into a reusable instruction file that makes future outputs less generic.

Send This To Your Agent

Copy and send this sentence:

Clone https://github.com/mshumer/unslop, read https://github.com/mshumer/unslop/blob/main/skills/unslop/SKILL.md, run unslop for DOMAIN_HERE, review unslop-output/analysis.md and unslop-output/skill.md for specificity, and return the finished skill file plus the main repeated patterns you found.

Agent-readable instructions live here: skills/unslop/SKILL.md

What It Does

Every model has defaults it collapses toward. In writing, that means the same openings, transitions, and rhetorical moves. In visual work, that means the same landing-page structure, the same gradients, the same card grids, the same CTA blocks.

unslop measures those defaults instead of guessing at them.

1. You give it a domain.
2. It generates a prompt set and produces many samples with Claude Code.
3. For visual work, it renders screenshots so the analysis can inspect what the pages actually look like.
4. It analyzes the whole set for repeated patterns.
5. It writes skill.md, a focused file of what to avoid next time.
6. It runs a before/after comparison so you can see whether the profile actually changed the result.

The main output is a single markdown file you can reuse in Claude Code, CLAUDE.md, Codex, Cursor rules, or any system prompt.

Quickstart

You need Claude Code installed.

git clone https://github.com/mshumer/unslop.git
cd unslop

python3 -m venv .venv
source .venv/bin/activate

# Text / writing domains
python3 unslop.py --domain "blog writing"

# Visual / website domains
pip install playwright && playwright install chromium
python3 unslop.py --domain "startup SaaS landing pages" --type visual --count 20 --concurrency 3

Output

Results land in ./unslop-output/:

unslop-output/
  skill.md
  analysis.md
  prompts.json
  samples/
  screenshots/
  before-after/
    before.md / before.html
    after.md / after.html
    before.png / after.png

skill.md is the file you usually want to keep and reuse.

Common Commands

python3 unslop.py --domain "marketing emails" --count 50 --concurrency 5
python3 unslop.py --domain "startup SaaS landing pages" --type visual --count 20 --concurrency 3
python3 unslop.py --domain "Python tutorials" --count 100 --concurrency 10
python3 unslop.py --domain "React landing pages" --type visual --model sonnet --effort low

Options