Jonah-Wu23/dsh-gungnir

Plugin插件 ⭐ 6 Apache-2.0 Productivity & Tasks效率与任务

首个面向 DeepSeek Harness、根据运行时证据在单次任务执行过程中动态切换 Agent Loop Strategy 的自适应 Loop 插件。基于环境证据链裁决任务完成度,在保证正常执行流畅的同时,精准拦截虚假完成与逻辑缺陷。

catalog descriptioncatalog 简介 / catalog description:Lock the goal. Adapt the loop. Prove the hit.

Project Overview项目介绍

Gungnir is a native plugin built exclusively for DeepSeek Harness (DSH), designed to provide evidence-guided adaptive loop control for AI agents during single-task execution. It dynamically adjusts the agent's loop strategy based on real-time runtime environmental evidence, and judges task completion status via a structured evidence chain. This allows it to accurately catch instances of false task completion and logical flaws in large language model outputs, without disrupting the normal smooth flow of task execution. Its three core capabilities work together to improve overall task completion quality for DSH users.

The plugin follows a layered, decoupled plugin architecture that is fully aligned with the DSH ecosystem. It stays completely silent during normal execution paths, and only intervenes when it detects definite conflicts between task claims and collected environmental evidence. When intervention is required, it injects clear, objective feedback based on real environmental facts to guide the large language model to fix identified issues. Independent testing confirms it only adds a 7.8% increase in token usage, with zero extra model interaction rounds, so it has minimal performance impact.

To use Gungnir, you must first be running DeepSeek Harness v0.1.2-alpha.1 or newer, because the plugin is not compatible with older versions like v0.1.1-rc2. After opening your DSH workspace in your terminal, run the pnpm add dsh-gungnir command to install the package from NPM. Next, add the plugin registration entry to your DSH configuration YAML file to enable it. Once you start DSH, Gungnir will automatically listen for execution events and run validation when tasks reach completion. The project is open-source under Apache 2.0, and welcomes community contributions.

Gungnir是专门为DeepSeek Harness(DSH)开发的原生证据导引型控制面插件,可在单次任务执行过程中根据运行时环境证据,动态调整智能体的执行回路策略。它基于环境证据链裁决任务完成度,在保证任务正常执行流畅度的前提下,能精准拦截大语言模型虚假完成任务和输出逻辑缺陷的情况,核心能力分为目标锁定、回路适配、证据裁决三类。

它采用分层解耦的插件化架构,从定义任务交付物和验证规则的目标契约,到调度执行回路策略的自适应运行时,再到收集环境事实进行静默验证的证据裁决层,全面适配DSH生态。正常执行路径下它保持静默,只有检测到确定性证据冲突时,才会注入明确的客观任务反馈,引导模型修正问题,额外Token开销仅增加约7.8%,不会明显影响性能。

该插件需在DSH v0.1.2-alpha.1及以上版本使用,不兼容旧版本。安装可通过pnpm add dsh-gungnir命令完成,之后在DSH配置文件中注册插件即可启用,支持被动监听和主动校验配置。项目采用Apache 2.0开源许可,欢迎开发者提交Issue和Pull Request参与社区贡献。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 2 warnings2 项注意
  • Only 6 stars - very few users, little community feedback星标只有 6,几乎没人在用,遇到问题缺少社区反馈
  • No DSH plugin manifest detected - it may only carry the dsh-plugin topic, so the install method must be confirmed on the spot未检测到 DSH 插件清单:可能只是打了 dsh-plugin 话题,安装方式要现场确认
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin --profile web add github:Jonah-Wu23/dsh-gungnir

把 Jonah-Wu23/dsh-gungnir 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

Gungnir (冈格尼尔)

Lock the goal. Adapt the loop. Prove the hit.

言出必行:DeepSeek Harness 的证据导引型控制面插件

npm package Version 0.1.1 Platform Apache License 2.0 Control Plane

30 秒了解 · 核心特性 · 架构设计 · 实验评测 · 快速上手 · 设计哲学 · 许可协议

30 秒了解

首个面向 DeepSeek Harness、根据运行时证据在单次任务执行过程中动态切换 Agent Loop Strategy 的自适应 Loop 插件。基于环境证据链裁决任务完成度,在保证正常执行流畅的同时,精准拦截虚假完成与逻辑缺陷。

Gungnir 为大语言模型智能体提供面向任务结果的控制面能力:

  • Lock the goal(目标锁定):建立版本化目标契约,维护明确的预期交付物和检验标准。
  • Adapt the loop(回路适配):作为 DSH 官方扩展,提供对智能体执行回路的平滑替换与策略调度。
  • Prove the hit(证据裁决):将模型输出视为待检验的主张,通过测试退出码与文件状态等环境证据裁决成败。
  • 静默守护与最小介入:正常执行路径下保持静默,只有在观测到确定性证据冲突时才注入面向任务的明确反馈。

核心特性

维度 原生智能体执行 Gungnir 控制面
完成判定 依赖模型自我宣称 依据环境证据与退出码客观裁决
假完成拦截 容易放行未完成的任务 运行时拦截虚假完成并要求修正
正常任务开销 基础开销 保持静默,Token 额外开销仅 +7.8%
控制面额外交互 无 正常路径零额外模型往返

架构设计

Gungnir 采用分层解耦的插件化架构,全面适配 DeepSeek Harness 生态:

┌─────────────────────────────────────────────┐
│ Gungnir Goal Contract                       │  目标锁定:定义任务交付物与验证规则
├─────────────────────────────────────────────┤
│ Gungnir Adaptive Loop Runtime               │  回路适配:调度执行回路与策略
├─────────────────────────────────────────────┤
│ Gungnir Evidence / Verifier / Reconciler    │  证据裁决:收集环境事实,静默验证并按需介入
├─────────────────────────────────────────────┤
│ DSH Agent Contract / Session Log / Services │  基础平台:提供会话与工具交互能力
└─────────────────────────────────────────────┘

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-fs-encoding 下一个 Next busabase-dsh-plugin →