Qian Jiu Jiu

A Chinese message preprocessor that detects obvious typos, grammar errors, and punctuation issues in user messages without auto-correcting. It prompts users to confirm using the original sentence before proceeding.

钱纠纠 (Qian Jiu Jiu)

只指出问题，不自动修改，静默确认继续

"Report issues only, never auto-correct, silently confirm to proceed"

Overview

钱纠纠 is a message preprocessor skill for OpenClaw agents that:

• Detects obvious typos, grammar errors, and punctuation issues in Chinese user messages
• Never auto-corrects - only reports issues found
• Prompts user confirmation before proceeding with potentially confusing messages
• Silently continues when user confirms (no verbose "confirmed" messages)

Installation

# Install via OpenClaw CLI
openclaw skills add qian-jiu-jiu

# Or manually copy to skills directory
cp -r qian-jiu-jiu ~/.openclaw/skills/
openclaw skills reload

Requirements

• OpenClaw >= 2026.2.x
• Chinese language context (optimized for Simplified Chinese)

Usage

The skill automatically activates on each user message:

Flow 1: Error Detection

When obvious errors are detected:

句子存在异常（可能影响理解）：
异常列表：
• 错字：神马 → 可能意为：什么
是否继续用这个原始句子进行推理？
（回复"是"则直接用上面原始句子推理；否则请修改后重新发送）

Flow 2: Silent Confirmation

When user replies with confirmation keywords:

Confirmation keywords: 是, 是的, 确认, 继续, ok, 行, 可以, yes, y, 好

The skill silently proceeds with the original sentence - no verbose confirmation, no "original sentence:" display, just normal response.

Flow 3: Clean Pass-through

If no obvious errors - message flows through normally without any output.

Error Types Detected

Type	Description	Example
多字	Redundant characters	这不是行啊 → 这不行啊
少字	Missing characters	你吃饭 → 你吃饭了吗
重复字	Repeated characters	我我我要出门 → 我要出门
错字	Wrong/typo characters	平果 → 苹果 / 神马 → 什么
错用字	Misused homophones	的/得/地 混用
语序错乱	Wrong word order	吃饭了我已经 → 我已经吃饭了
搭配不当	Collocation errors	喝饭 → 吃饭
标点缺失	Missing punctuation causing ambiguity	我想出去玩球没买 → 出去玩，球没买? / 出去玩球，没买?
空格异常	Invalid spaces in words	假 如 → 假如
单位遗漏	Missing units after numbers	身高178 → 身高178厘米
拼音错	Pinyin instead of characters	uf → 效果 / nihao → 你好

Response Format

Error Detection Format (Strict line-by-line):

句子存在异常（可能影响理解）：\n
异常列表：\n
• [type]: [error] → 可能意为：[possible meaning]\n
是否继续用这个原始句子进行推理？\n
（回复"是"则直接用上面原始句子推理；否则请修改后重新发送）

Confirmation Response Format:

[Normal response to the original sentence]

Key Constraints

1. Never auto-correct - Only report, never suggest "correct version"
2. Silent confirmation - No "confirmed", "proceeding", or "OK" messages
3. Conservative detection - Better to miss than over-report
4. No empty lines between output lines (use \n only)
5. Context-aware - Considers conversation history for ambiguous cases

License

MIT

Contributing

Issues and PRs welcome at https://github.com/[your-username]/qian-jiu-jiu