Andy8647/dsh-auto-approval

Name: Andy8647/dsh-auto-approval
Author: Andy8647

Plugin插件 Native原生 ⭐ 3 BSD-3-Clause Approval & Security审批与安全 Prompts & Skills提示词与技能

A fourth permission preset for DeepSeek Harness: pick Automode in the composer's permission dropdown and the session runs on full access with an LLM classifier as the only gate before every tool call. No approval prompts — the classifier answers for you.

catalog descriptioncatalog 简介 / catalog description：Automode for DeepSeek Harness: a fourth permission preset — full access with an LLM classifier as the only gate before every tool call.

Project Overview项目介绍

Edited by Zhiwei Agent (zhiwei)由知微Agent(zhiwei)编辑

This repository is a native permission preset plugin built exclusively for DeepSeek Harness, that adds a fourth "Automode" auto-approval tier to DSH’s existing three official sandbox permission levels. It eliminates the constant interruptions from manual approval prompts that break flow for long running LLM tasks. When you select the Automode preset from the permission dropdown, every tool call is evaluated by a two-stage guard system that automatically approves or denies calls without any manual input from you. The plugin only activates when Automode is selected, and switching to any other built-in DSH preset automatically disables it with no extra toggle required.

The two-stage guard system first runs deterministic L0 rules that hard-deny high-risk commands like rm -rf / or curl piped to sh, based on user-configured regex patterns and allowlists. The classification logic never exposes tool output to the LLM classifier, so injected content can’t trick the classifier into allowing a bad request. Any call that passes L0 moves to L1, where a configured LLM evaluates the latest user message and tool call to decide whether to allow or deny the request. This setup is ideal for users that want to run long, unattended multi-step LLM workflows without stopping to manually approve calls.

You can install the plugin directly via the DSH CLI with the command dsh plugin --profile web add dsh-auto-approval, which pulls the latest version from the DSH plugin registry. After installation, just select Automode from the composer’s permission dropdown to start using it; first-time users in a browser will see a one-time notice explaining the security tradeoffs of automatic approval. The plugin is open source under the permissive BSD-3-Clause license, and is currently verified to work with the latest DSH release 0.1.2-rc.1, with older DSH versions not officially supported.

dsh-auto-approval是专为DeepSeek Harness开发的原生插件，它为DSH新增了一个名为Automode（自动批准）的权限预设选项。选中该预设后，所有工具调用都会由内置的两层规则+LLM分类器自动批准或拒绝，不需要用户手动点击确认审批，在官方三个沙箱权限等级之外新增了自动审批层级。

该插件采用两级审核逻辑：先执行确定性L0硬拒绝规则，拦截高危命令，再将通过规则的调用交给L1层大语言模型分类器判断允许或拒绝。它适合需要运行无人值守的多步骤大语言模型任务，又不想完全放开所有权限的用户使用，只要选中其他预设插件就会自动禁用，不需要额外开关。

可以通过DSH官方命令行直接安装，命令是dsh plugin --profile web add dsh-auto-approval，安装完成后在编辑器权限下拉菜单选中Automode即可启用。项目遵循BSD-3-Clause开源许可，当前适配最新版DSH 0.1.2-rc.1，旧版DSH不支持，首次启用浏览器会显示一次性提示说明安全取舍。

View on GitHub在 GitHub 查看

Send to your DSH发给你的 DSH 安装

请帮我安装这个 DSH 插件。安装前先完成【兼容性检查 + 安全性检查】，检查通过再动手。

插件：dsh-auto-approval（Andy8647/dsh-auto-approval）
仓库：https://github.com/Andy8647/dsh-auto-approval
本站详情页：https://www.yhbd.top/plugins/andy8647-dsh-auto-approval/
本站登记：类型 plugin · 归类 原生 DSH 插件 · 许可证 BSD-3-Clause · ⭐ 3 · 最近提交 2026-09-20 · 主语言 TypeScript

按下面顺序执行，每步先把结论告诉我，再进入下一步：

【1 兼容性检查】
① 我这边：DSH 版本、Node 版本、操作系统、当前 profile（web / desktop）。
② 读它的 README、package.json、插件 manifest，列出它要求的 DSH 版本 / Node 版本 / 操作系统 / 外部依赖 / 需要另外先装的运行时。
③ 逐条比对，结论只写「满足 / 不满足 / 未知」三种；不满足的给出可行替代方案。
④ 检查是否和我已装的插件冲突：命令名重复、skill / tool 重名、端口占用、重复注册的 MCP server。

【2 安全性检查】
① 仓库可信度：和上面「本站登记」是否一致；star / fork 数、创建时间、最近提交，是否归档或长期停更。
② 安装脚本：逐行看 package.json 的 preinstall / install / postinstall，以及 install.sh、setup.ps1 之类脚本。出现 curl|bash、下载后直接执行、混淆代码、访问与插件功能无关的域名，立刻停下来告诉我，不要继续装。
③ 依赖：列出新增依赖，标出无人维护、或与知名包拼写近似的可疑包（typosquatting）。
④ 权限与副作用：它会读写哪些目录、访问哪些域名、需要哪些 DSH 权限（filesystem / network / shell / clipboard 等），以及怎么卸载和回滚。
⑤ 如果它要求 sudo / 管理员权限，或权限明显超出功能所需，先停下来问我。

【3 安装】
上面两步没有「不满足」和「高危项」时才执行；用官方推荐方式安装，不要自行提权。

【4 汇报】
用表格输出：检查项 / 结论 / 依据 / 是否需要我决策。拿不准的一律写「未知」并说明要我怎么确认——不要猜，也不要替我决定。

Send this message to DSH in your current session: it verifies compatibility and security first (answering met / not met / unknown item by item) and only installs once everything checks out — it will stop and ask you if it finds a high-risk item. The box scrolls; the copy is the full prompt. CLI install commands may not be accurate across systems, so DSH is the safer route.把上面这条消息直接发给当前会话里的 DSH：它会先核对兼容性与安全性（逐条给「满足 / 不满足 / 未知」），确认没问题再安装，有高危项会停下来问你。框内可滚动，复制到的是完整提示词；安装命令不一定准确，发给 DSH 更稳。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意

Only 3 stars - very few users, little community feedback星标只有 3，几乎没人在用，遇到问题缺少社区反馈

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 dsh-auto-approval

把 Andy8647/dsh-auto-approval 加入你的 DSH 配置（web profile）即可启用。

READMEREADME

9.7 KB · 缓存于 2026-09-20 19:40:57 · 9.7 KB

dsh-auto-approval

English | 中文

A fourth permission preset for DeepSeek Harness: pick Automode in the composer's permission dropdown and the session runs on full access with an LLM classifier as the only gate before every tool call. No approval prompts — the classifier answers for you.

The permission model stays 3 + 1: the three official sandbox levels, plus one auto tier. Selecting any other preset turns the plugin off; there is no second switch.

0.2.0 is a rewrite of the 0.1.x line. The on/off switch became a permission preset: install, pick Automode, done. Upgrading? See Upgrading from 0.1.x.

Demo

Automode: a multi-step task runs unattended, then a denied call shows up in the decision table

The recording: pick Automode, send one prompt (create a directory, write a file, read it back, then run echo danger_test). The agent runs the file steps unattended — the classifier allows them (L1-deep / whitelist) — and the final command is blocked by a hard rule (L0-deny). The chip's dialog shows both verdicts and the cumulative counts.

How it works

model wants a tool call
        │
        ├─ preset ≠ automode ──────────────► untouched (official behavior)
        │
        └─ preset = automode
                 │
                 ├─ L0 rules (hard deny) ───► deny   (rm -rf /, curl | sh, self-kill …)
                 ├─ trusted tools / bash prefixes ──► allow
                 └─ L1 classifier ──────────► allow | deny   (fail-closed: timeout/parse/no-model → deny)

L0 — regex deny rules, self-kill guard, trusted-tool and bash-prefix allowlists. Deterministic, no model call.
L1 — the latest real user message plus the bare tool call go to a classifier. Two backends: llm (default) runs a two-stage prompt on a chat model (fast single-token filter → deep CoT check when flagged); jev sends one HTTP request to TypeSafe's System One and gets back typed probabilities — the verdict is a threshold comparison, no text parsing. Tool output is never shown to the classifier on either backend, so injected content cannot talk it into an allow.
Fail-closed — a timeout, a parse failure, or a missing model denies the call.

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 →

渲染基于仓库 Andy8647/dsh-auto-approval 的 README.md。在 GitHub 查看原文 →