Back to skill

Security audit

deepseek-harness-windows-deploy

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed Windows troubleshooting guide for DeepSeek Harness, with sensitive steps scoped to local setup tasks and requiring user confirmation.

Install this only if you are using an agent to deploy or troubleshoot DeepSeek Harness on Windows. Before allowing it to run sensitive commands, confirm the exact target: clearing NODE_OPTIONS should apply only to the single dsh web launch, process termination should target only the port-3080 Harness process, and deletion should be limited to the documented stale symlink under ~/.dsh.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 33)May include surrounding context.

md
本技能遵循开放的 **Agent Skills 标准**(`SKILL.md` + YAML frontmatter),任何支持该标准的 agent 运行时都能加载。按确认程度分类:

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 36)May include surrounding context.

md
本技能遵循开放的 **Agent Skills 标准**(`SKILL.md` + YAML frontmatter),任何支持该标准的 agent 运行时都能加载。按确认程度分类:

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 37)May include surrounding context.

md
本技能遵循开放的 **Agent Skills 标准**(`SKILL.md` + YAML frontmatter),任何支持该标准的 agent 运行时都能加载。按确认程度分类:

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file presents the skill description entirely in Chinese from the outset, which can impose a language constraint on users and agents without explicit opt-in. Under the policy rule, forcing a specific language is a natural-language policy violation unless the skill offers a language choice or clearly justifies the locale restriction.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The file’s natural-language instructions and description are entirely in Chinese, and there is no indication that users may choose another language or that the skill is limited to a Chinese-speaking audience for compliance or regional reasons. Under the policy, imposing a language/locale without opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The entire skill reference is written in Chinese and does not indicate that language selection is optional or that the document is intentionally limited to a Chinese-speaking audience for a specific compliance or regional reason. Under the policy, forcing a specific language without user opt-in is a natural-language locale violation.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
86% confidence
Finding

This duplicate finding describes the same behavior: clearing NODE_OPTIONS disables an injected deletion-safety mechanism for the launched Node process tree. In a managed sandbox, advising users to suppress such controls can create a precedent for bypassing other enforced protections and may expose files to unintended destructive operations if the process or dependencies misbehave.

Content

Scanner excerpt · references/deploy-pitfalls.md (reported line 59)May include surrounding context.

md
`NODE_OPTIONS=--require=.../genie-safe-delete.cjs`,把 `fs.rm/unlink` 劫持去调用
  `genie-trash.exe`(Windows 上超时,并以 fail-closed 方式报错)。
- **影响点**:Harness 设置持久化 `FileSettingsProvider.persistSection` 用
  `withFileLock`(mkdir → writeFileAtomic → `rm(lock)`),`finally` 里的 `rm(lock)` 被拦截抛错
  → `mutate` 失败 → 前端只显示通用文案 **"暂时无法保存确认状态,请重试"**(即内测声明的确认步骤)。
  **实际上数据(`~/.dsh/settings.yaml` 的 `ui-onboarding.welcomeNoticeVersion`)已写成功,只是内存未标记为已确认。**
- **解法**:启动 dsh 时前缀 **`NODE_OPTIONS=""`**(node 不加载该钩子,锁清理恢复正常)。此变量仅对这条启动命令的子进程生效,不影响其他进程;**执行前请向用户说明并确认**。

Session Persistence

Medium
Category
Rogue Agent
Confidence
86% confidence
Finding

This duplicate finding describes the same behavior: clearing NODE_OPTIONS disables an injected deletion-safety mechanism for the launched Node process tree. In a managed sandbox, advising users to suppress such controls can create a precedent for bypassing other enforced protections and may expose files to unintended destructive operations if the process or dependencies misbehave.

Content

Scanner excerpt · references/deploy-pitfalls.md (reported line 59)May include surrounding context.

md
`NODE_OPTIONS=--require=.../genie-safe-delete.cjs`,把 `fs.rm/unlink` 劫持去调用
  `genie-trash.exe`(Windows 上超时,并以 fail-closed 方式报错)。
- **影响点**:Harness 设置持久化 `FileSettingsProvider.persistSection` 用
  `withFileLock`(mkdir → writeFileAtomic → `rm(lock)`),`finally` 里的 `rm(lock)` 被拦截抛错
  → `mutate` 失败 → 前端只显示通用文案 **"暂时无法保存确认状态,请重试"**(即内测声明的确认步骤)。
  **实际上数据(`~/.dsh/settings.yaml` 的 `ui-onboarding.welcomeNoticeVersion`)已写成功,只是内存未标记为已确认。**
- **解法**:启动 dsh 时前缀 **`NODE_OPTIONS=""`**(node 不加载该钩子,锁清理恢复正常)。此变量仅对这条启动命令的子进程生效,不影响其他进程;**执行前请向用户说明并确认**。

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/deploy-pitfalls.md (reported line 119)May include surrounding context.

~/.dsh/sessions/<sanitized-path>/ 是否生成。若都有,说明添加 + 开会话后端成功。 2. 用 RPC 复现"开会话"验证:

bash
curl -X POST http://127.0.0.1:3080/api/session.create \
  -H 'Content-Type: application/json' \
  -d '{"type":"client-request","rpcId":"x","method":"session.create","payload":{"workspaceId":"<工作区id>"}}'