Auto Continue

@fazaboa/opencode-auto-continue·v0.1.0·其他

OpenCode plugin: automatically continues responses that were cut off by the model's output limit.

#opencode #opencode-plugin #ai #auto-continue

GitHub 仓库 npm 页面官方主页

GitHub 星标

近 30 天 +1

月装机量

近 7 天 4

综合评分

36.1

生态多维模型

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": ["@fazaboa/opencode-auto-continue@0.1.0"]
}

写入 ~/.config/opencode/opencode.json，对所有项目生效。

~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@fazaboa/opencode-auto-continue@0.1.0"]
}

OpenCode 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。

README.md查看原始文件→

English · Português (BR)

OpenCode plugins for smarter AI sessions — thinking-guard (anti loop) and truncation detection.

Features

🛡️ Thinking Guard

Detects and interrupts reasoning loops using two strategies:

Token count — aborts when output crosses the hard limit.
Repetition detection — Jaccard similarity on bigrams catches repetitive reasoning.

When a loop is detected: aborts the step + injects "stop thinking and answer".

→ Docs

🔄 Truncation Detection

Detects responses cut off by the model's output limit and auto-injects a continuation prompt telling the model to pick up exactly where it left off.

→ Docs

Install

npm install @fazaboa/opencode-plugins

Then register it in your opencode.json:

{
  "plugin": ["@fazaboa/opencode-plugins"]
}

Or drop the source file directly into .opencode/plugins/auto-continue.ts in your project:

export { default } from "../../opencode-plugins/src/index.ts"

Usage

Both features (default)

export { default } from "../../opencode-plugins/src/index.ts"

Individual features

import { createThinkingGuardPlugin } from "../../opencode-plugins/src/index.ts"

export default createThinkingGuardPlugin({
  softLimit: 10_000,
  hardLimit: 20_000,
})

import { createTruncationPlugin } from "../../opencode-plugins/src/index.ts"

export default createTruncationPlugin({
  softLimit: 115_000,
  maxAutoContinues: 3,
})

Configuration

All config lives in src/config.ts. Defaults:

thinking: {
  softLimit: 10_000,           // alert
  hardLimit: 20_000,           // abort
  repetitionWindow: 4,         // chunks in buffer
  repetitionSimilarity: 0.6,   // Jaccard threshold
  repetitionConsecutive: 3,    // high comparisons = loop
}
truncation: {
  softLimit: 115_000,          // ~88% of ~131k cap
  maxAutoContinues: 3,
}
global: {
  maxInterventionsPerTurn: 2,
  cooldownMs: 5_000,
}

Project Structure

opencode-plugins/
├── src/
│   ├── index.ts                # Entry point — registers all features
│   ├── config.ts               # Centralized configuration
│   ├── features/
│   │   ├── thinking-guard.ts   # Loop detection + abort
│   │   └── truncation.ts       # Truncated response detection
│   └── utils/
│       ├── repetition.ts       # Jaccard similarity on bigrams
│       └── tokens.ts           # Per-session token counter
├── tests/                      # Vitest unit tests
├── docs/                       # Per-feature documentation
├── CHANGELOG.md
├── package.json
├── README.md
└── tsconfig.json

Guardrails

Guardrail	Default
Max auto-continues per user turn (anti infinite-loop)	3
Counter resets when you send a new message	✓
Soft limit well below the cap (catches "finished but glued to the cap")	115k tokens
Silent failure (never breaks your session)	✓
Thinking guard hard limit (abort reasoning loops)	20k tokens

License

MIT