DSH Plugins Marketplace

DSH Plugins

Plugins

/

Tools & Capabilities

/

dsh-doom-loop-guard

d

dsh-doom-loop-guard

Identified

Detects repetitive loops in an agent's reasoning output and records, nudges, or cuts them, with a history report. Defaults to a nudge (a reminder injected into the next request); cutting and auto-resu

dsh-doom-loop-guard

检测推理中的重复循环(doom loop),命中时提醒、截断或响铃,并生成历史报告。

能力

  • 实时:包装 llm/stream,检测破折号填充、括号嵌套、短句重复
  • 处置:observe 只记录 / nudge 注入提醒 / cut 截断并自动续跑
  • 门槛:上下文占用达到设定百分比后才生效
  • 音效:命中响铃,内置 assets/loop-alert.mp3
  • 报告:扫描会话存档,给出会话 doom 分与 14 天趋势

防护等级

等级破折号括号短句重复间隔占比
宽松12003002068%
普通8002001585%
敏感50012010123%

普通档为校准值:1.4 万条真实推理上召回 100%、误报 0。

配置

写在 profile 的 cordis.patch.yml(id dsh-doom-loop-guard),或在设置页改;设置页写入 ~/.dsh/dsh-doom-loop-guard.json。

字段默认说明
presetstandardrelaxed / standard / sensitive
interventionnudgeobserve / nudge / cut
warnAbovePercent0生效门槛,0 为全程
autoResumetruecut 后自动续跑
maxConsecutiveCuts3连续截断上限
soundEnabledtrue音效
soundPath''自定义音频,留空用内置
reportDays / reportLimit7 / 20历史扫描范围

阈值由等级决定,也可在 patch 里直接写 maxDashRun / maxParenRun / shortSentenceRepeat / maxRepeatGap / shortRepeatRatio 覆盖。

界面语言

设置页文案有中文与英文两套,跟随 DSH 界面语言(设置 → 常规 → 语言)即时切换,不需重启。插件清单里的名称与描述另由 locale/zh.json / locale/en.json 提供。

宿主若缺 locale 服务(精简组合),界面退回中文并在控制台留一条 warn。

安装

dsh plugin --profile <profile> add dsh-doom-loop-guard

开发时用 install_bundle 指向本目录。

要求 DSH >= 0.2.1、Node >= 22.15(用到内置 zstd 解压)。

与 dsh-loop-rescue

同现象、不同检测器。1.4 万条样本上,本插件召回 100%、loop-rescue 60.5%,后者漏的主要是纯符号填充。

两者都包装 llm/stream:cut 与 loop-rescue 同时开会重复截断,二选一。

边界

  • 只覆盖符号崩坏与低信息循环,不含语义打转
  • nudge 的提醒落到下一步请求;cut 才能当场中断

Comments

Loading…

Similar plugins

dsh-output-loop-guard

by ADistanceThereIs

DSH guard against self-repetition in streamed model output: detects a degeneration loop in reasoning or visible text, aborts the runaway turn and wakes the agent with a corrective notice.

Tools & CapabilitiesManifest valid

★ 0

MIT

JavaScript

Oct 16, 2026

dsh plugin --profile web add dsh-output-loop-guard

by himi-li

Detects verbatim reasoning loops in DSH LLM streams and cuts them early. Zero-dependency Cordis plugin with a fire journal and a reasoning_loop_log tool.

Development & InfrastructureManifest valid

★ 2

MIT

JavaScript

Oct 18, 2026

dsh plugin --profile web add dsh-reasoning-loop-guard

by lyc0110

Guardrail for DSH llm/stream: detects when model reasoning or text degenerates into a periodic repetition loop and aborts the request instead of streaming until you give up.

Manifest valid

★ 0

MIT

JavaScript

Oct 21, 2026

dsh plugin --profile web add dsh-loop-guard

by erdholion

Result-aware stuck-loop guard for DeepSeek Harness: advisory nudges plus a monotonic hard stop. Only repeats with identical results count, so progressing polls and post-edit re-reads never trigger.

Tools & CapabilitiesDevelopment & InfrastructureManifest valid

★ 0

MIT

JavaScript

Aug 31, 2026

dsh plugin --profile web add dsh-loop-guard

by MrWeiCodes

为 DeepSeek Harness(DSH)提供的思考循环守护插件:模型陷入「只想不做」的退化循环时自动打断,不必再手动中止

Workflow & AutomationManifest valid

★ 36

↓ 79/wk

MIT

JavaScript

Nov 5, 2026

dsh plugin --profile agent add dsh-loop-guard

by Jessicaisleep

Strips DSML markers that leak into the assistant text stream and resumes a turn only when it genuinely produced nothing — an invoke-block leak with no tool call emitted, or a closing-tag leak whose pr

Development & InfrastructureSessions & MessagesManifest valid

★ 2

MIT

JavaScript

Oct 28, 2026

dsh plugin --profile web add dsh-dsml-guard