⚔️ DeepSeek Harness cybersecurity red-team tool (dsh-purge)
Version 1.1.81
DeepSeek Harness jailbreak: every model, swap prompts per model. Default prompt for 小码酱. Please star ⭐ If you build on this, see Attribution required.
English | 中文 | 日本語
A note from the author
People have been selling this open-source work in the open. After that, the author was asked to come in and talk. No need to spell out the rest. Read between the lines.
From now on the public internet is closed. The open-source build only allows local lab IPs: loopback and RFC1918. Domains are resolved first (including hosts); every answer must be private or loopback. A hosts trick that points .lab at a public IP, a failed lookup, or a mixed public answer is blocked. Game cheats are forbidden (aimbot, wallhack, memory edits against a game, anti-cheat bypass); the prompt and the session gate both stop them. This build does not ship the skill library or the knowledge base.
The original build may not be resold. Charging for it, skinning it, or hawking it in groups is not allowed. Forks are fine, but you must attribute and name the source: YuJunZhiXue/dsh-purge. Hide the source or pass it off as your own, and you are on the other side.
The author's personal channel: t.me/LLM_PoJia. Notes and updates show up there.
Contents
What this is
dsh-purge is a cybersecurity red-team plugin for the official DeepSeek Harness already installed on your machine. 1.1.81 targets DSH 0.2.1-rc.2 (0.2.0 still works). DSH 0.2.1-rc.3 and newer RCs are not supported. Official Web (dsh web) and the official desktop EXE are both supported. Install and Apply them separately.
Harness does more than forward a prompt. It rewrites prompts, approval gates, sandbox policy, and session migration at runtime. Asking only “will the model answer” treats a host-layer block as a model refusal. This plugin separates those two layers on your own install, for controlled evaluation and for seeing which default policy is too tight.
| You get |
What it does |
| dsh-purge dock |
A button beside the session title opens the right dock. Two pages: Clean and Drill |
| Clean |
Grouped patch status, Apply / Restore / Uninstall, prompt editor, multiple rule sets |
| Drill |
Built into the stable release. After authorization: assets and the local environment. This build does not ship skills or the knowledge base. Local lab only |
| Host policy |
Default copy, permission policy, and tool limits. Official capabilities stay. No second invented identity |
| On start |
Checks and reapplies. After npm overwrites node_modules, you do not hand-edit files |
No hardcoded drive letters. It looks at $DSH_HOME, .dsh next to the dsh launcher, then ~/.dsh. It does not patch the Harness source tree. Apply on the Clean page of dsh-purge is what writes the changes. Identity comes from the encrypted prompt shipped in the plugin, the text the prompt box shows by default. The host prompt file is not read.
It only touches the official @deepseek-ai packages and local config on the user's machine. It is not a public scanner and not an attack kit for third-party sites. The repo does not ship malware, unauthorized-exploit scripts, or payloads aimed at the public internet.
The original may not be resold. Forks must attribute.
Attribution required
The original build may not be resold. No paid resale, no skinned store listing, no hawking it in groups.
Forks must attribute and name the source repository: YuJunZhiXue/dsh-purge. Same if you borrow the name, ideas, code, or prompts.
Hiding the source, passing it off as your own, or reselling the original will be pursued.
License
This repo is Apache License 2.0. Forks and redistributions must keep the copyright notice and NOTICE, and name YuJunZhiXue/dsh-purge. The original may not be resold.
### ⚠️ Strict legal and compliance disclaimer
**Disclaimer:** This is a non-profit open-source project. It follows applicable laws and the rules of the platforms it uses, and is intended only for learning and research. Do not use it for any illegal or non-compliant purpose; the user bears any resulting consequences.
**Zero-tolerance statement:** This project opposes and forbids any illegal activity. The authors do not support, encourage, or assist unauthorized network attacks, exploit use, data theft, unlawful intrusion into computer information systems, or generation of illegal or prohibited content. **Anyone who uses this project for crime is solely responsible under the law. The authors have no liability.**
1. **This repository contains no illegal material.** The published code, docs, patches, and default prompt are **not** malware, backdoors, unauthorized pentest kits, ransomware, credential-stuffing scripts, or attack payloads aimed at the public internet or third-party systems. The project does not supply illegal content and does not incite, organize, or assist crime.
2. **Local official Harness only.** Security-eval patches and prompt injection run only on the **official DeepSeek Harness already installed on the user's machine** (`@deepseek-ai` packages, local profile / `$DSH_HOME`). The target is the user's own official local software, **not** someone else's website, server, account, or information system.
3. **Eval patches do not attack the public internet.** Apply, inject, revert, and uninstall stay in local files and local processes. They **do not scan, probe, intrude, or send attack traffic to any public host or unauthorized system**. Do not use this project as a jump host against the public internet. If "check for updates" is on, the plugin may only contact this plugin's own GitHub repository to compare versions. That is **not** an attack on third-party systems and is **not** authorization to penetrate anything.
4. **Lawful, controlled scope.** This project is an aid for red-team research and robustness evaluation on an **official Harness the user is entitled to administer**. **Do not run it against targets without the owner's lawful written authorization, public online systems, or production workloads.** Testing must stay on the **authorized local official Harness, offline local synthetic fixtures, authorized cybersecurity ranges, and compliant lab environments**.
5. **The open-source build is local-lab only.** Attack tasks only allow loopback and RFC1918 addresses. Domains are resolved, then judged by IP; suffixes are not an allow list. A public IP, a name that resolves public, a failed lookup, or a hosts entry that points a lab name at the public internet is not an exception. A key, a password, a network range, a wildcard, or a claim made in the conversation does not add an exception. Game cheats are forbidden. This build does not ship the skill library or the knowledge base.
6. **Forbidden uses.** Users must not use this project, directly or indirectly, to violate the following (each must be followed; no excuse to evade):
- Criminal Law of the People's Republic of China
- Cybersecurity Law of the People's Republic of China
- Data Security Law of the People's Republic of China
- Personal Information Protection Law of the People's Republic of China
- and other applicable laws, regulations, and supervisory rules;
- Also forbidden:
- Unauthorized intrusion or attacks on public or private computer information systems; extortion, sabotage, credential stuffing, or spreading malicious payloads;
- Inducing, generating, or spreading any content the law forbids, including threats to national security, terrorism, violence, pornography, gambling, fraud, and privacy or intellectual-property infringement;
- Violating the model provider's terms of service and acceptable-use policy.
7. **The user bears all responsibility.** The project is provided under Apache-2.0 as-is. The authors make no warranty of completeness, security, or fitness. **Users independently bear all civil, administrative, and criminal liability** for download, deploy, run, modify, distribute, and all resulting inputs and outputs. Authors and contributors bear no direct, indirect, or joint liability for abuse.
8. **The license ends on breach.** Anyone who uses this project