1
25
近 7 天 6
21.8
生态多维模型
5 个月前
2026-02-27
快速安装与配置
opencode.json写入当前项目的 opencode.json,只对这个仓库生效。
opencode.json
{
"$schema": "https://opencode.ai/config.json",
"plugin": ["opencarly@1.0.3"]
}写入 ~/.config/opencode/opencode.json,对所有项目生效。
~/.config/opencode/opencode.json
{
"$schema": "https://opencode.ai/config.json",
"plugin": ["opencarly@1.0.3"]
}若你要在本地改造这个插件,先装到项目里再从本地路径引用。
shell
pnpm add -D opencarlyopencode 启动时会通过内嵌运行时自动加载 npm 依赖并缓存至本地目录,无需手动在全局环境执行安装。
Context Augmentation & Reinforcement Layer for OpenCode
OpenCarly is an intelligent plugin for OpenCode that dynamically manages your AI's context window. Instead of dumping all your rules, API guidelines, and project instructions into a single massive prompt, OpenCarly loads rules only when they are relevant and seamlessly trims them from the chat history when they aren't.
This saves massive amounts of tokens, dramatically reduces your API costs, and keeps your AI laser-focused on the task at hand without being distracted by irrelevant guidelines. Heavily inspired by Claude Code - CARL.
🚀 Features
- Dynamic Rule Injection: Automatically injects specific instructions based on the files currently loaded in your context (e.g., injects
Reactrules only when a.tsxfile is open). - Keyword Triggers: Trigger rule injection simply by typing a keyword in your prompt (e.g., typing "*api" injects your backend API guidelines).
- History Trimming: Aggressively removes injected rules from previous messages in the chat history, ensuring you only pay for the context once.
- Cost Estimation & Stats: Run
*statsat any time to see exactly how many tokens (and estimated dollars!) OpenCarly has saved you.
📦 Installation
To install OpenCarly globally, use npm:
npm install -g opencarly
Then, initialize OpenCarly in your project directory:
cd your-project-dir
npx opencarly init
This will create an .opencarly configuration directory in your project containing a config.json file and a rules/ folder where you can place your dynamic guidelines.
⚙️ Configuration
Open your newly created .opencarly/config.json to start adding rules.
A rule consists of:
name: A descriptive name for the rule.files: (Optional) An array of file globs. The rule will automatically inject if any file matching these globs is loaded in OpenCode.keywords: (Optional) An array of keywords. The rule will inject if any of these words (prefixed with a*, like*sql) are typed in your prompt.content: The path to the markdown file containing your instructions (relative to the.opencarly/rules/directory).
Example Configuration
{
"rules": [
{
"name": "React Guidelines",
"files": ["**/*.tsx", "**/*.jsx", "components/**/*"],
"content": "react.md"
},
{
"name": "Database Schema",
"keywords": ["db", "sql", "database"],
"content": "schema.md"
}
]
}
With this setup:
- Editing a
Button.tsxfile will automatically inject the rules fromreact.md. - Asking the AI "Please write a *sql query" will automatically inject the rules from
schema.md.
📊 Viewing Token Savings
You can see how many tokens OpenCarly has saved you by using the built-in stats command inside OpenCode:
user: *stats
OpenCarly will output a detailed report showing total tokens trimmed, prompts processed, and an estimated dollar amount saved based on your current AI model's input token pricing.
📝 License
MIT