deaiify

该插件在助手回复交付前拦截消息，检测代码块和行内代码之外的 U+2013 与 U+2014，调用嵌入式 LLM 改写为不含这些字符的文本，并通过词数和长度校验。它使用会话默认模型，仅提供重写超时配置。

shawnpetros

热度 82.97

最近更新 2026-05-11 17:43:29

分类：效率依赖：内部调用大模型应用场景：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 deaiify

Metadata

name: deaiify
description: Ban LLM em-dashes and en-dashes before delivery, rewrite via embedded LLM.
homepage: https://github.com/shawnpetros/deaiify
metadata: {"openclaw":{"emoji":"🚫","plugin":true}}

deAIify Plugin

Enforces the "no em dashes" rule across all output by intercepting messages, detecting U+2013 (en dash) and U+2014 (em dash), and forcing a rewrite turn via embedded LLM correction.

What it does

Detects U+2013 and U+2014 in outbound messages
Calls embedded LLM to rewrite without dashes (preserves meaning, tone, style)
Returns corrected output to user
Never touches U+002D (hyphen-minus in code)
Excludes content inside code blocks and inline code

How it works

before_agent_reply intercepts the completed assistant text before delivery
Detects banned dash characters outside of code blocks
Calls runEmbeddedPiAgent with a restructuring prompt
Verifies the rewrite is sane (word count and length checks)
Returns the rewritten reply

Installation

openclaw plugin install deaiify

Configuration

Option	Type	Default	Description
`rewriteTimeoutMs`	integer	`15000`	Timeout in ms for the embedded rewrite call

The plugin uses the session's default model for rewrites. No model config needed.