跳转到正文
报告库
用途分类 / 其他用途

Dispatching Parallel Agents Skill 安全审计

作者说它能做什么(原文)

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

第三方安全检查结论

发现安全风险

已检查文件
1
发现的风险
2
会不会运行危险命令?检查是否下载程序后直接运行、让他人远程控制电脑,或藏起要运行的命令。未发现风险
会不会泄露文件和密钥?检查是否发送含密码或密钥的文件,以及代码里是否直接写了密钥。未发现风险
会不会删除文件或一直在后台运行?检查是否大范围删除文件、改写磁盘,或设置自动启动。发现 1 项风险
中风险

“让测试通过”的委派可修改实现或降低测试预期

原文依据:4 处
发现了什么

示例任务允许子代理在发现问题时修改中止实现,也允许因“行为已改变”而调整测试预期。由于首要目标被表述为让测试通过,子代理可能通过接受错误的新行为来获得绿色测试,而不是修复回归。

为什么需要注意

产品缺陷可能被较弱或错误的测试预期掩盖;生产代码也可能在没有用户逐项批准的情况下发生持久修改。完整测试全部通过仍不能发现已被改写掉的正确要求。

这是一个有条件的决策完整性风险。Skill 将目标设为“让测试通过”,并在示例中明确允许子代理修改实现,或在其判断行为已变化时调整测试预期。若缺少独立规格或人工审查,子代理可能把回归重新定义为预期行为,从而得到绿色测试却掩盖缺陷。不过,Skill 同时要求审查摘要、检查冲突、运行完整测试和抽查,因此并非无条件接受修改,也不能据此断言实际发生了降级。用户可要求任何预期变更必须引用已批准规格并单独确认。

SKILL.md:60来自说明文档打开原文件
Each agent gets:- **Specific scope:** One test file or subsystem- **Clear goal:** Make these tests pass- **Constraints:** Don't change other code- **Expected output:** Summary of what you found and fixed
查看另外 3 个位置
SKILL.md:101来自说明文档打开原文件
These are timing/race condition issues. Your task:1. Read the test file and understand what each test verifies2. Identify root cause - timing issues or actual bugs?3. Fix by:   - Replacing arbitrary timeouts with event-based waiting   - Fixing bugs in abort implementation if found   - Adjusting test expectations if testing changed behaviorDo NOT just increase timeouts - find the real issue.Return: Summary of what you found and what you fixed.```
SKILL.md:103来自说明文档打开原文件
1. Read the test file and understand what each test verifies2. Identify root cause - timing issues or actual bugs?3. Fix by:   - Replacing arbitrary timeouts with event-based waiting   - Fixing bugs in abort implementation if found   - Adjusting test expectations if testing changed behaviorDo NOT just increase timeouts - find the real issue.
SKILL.md:163来自说明文档打开原文件
After agents return:1. **Review each summary** - Understand what changed2. **Check for conflicts** - Did agents edit same code?3. **Run full suite** - Verify all fixes work together4. **Spot check** - Agents can make systematic errors
会不会绕过安全保护?检查是否跳过网站安全验证、开放过多文件权限,或取消操作前的确认。未发现风险
会不会误导 AI 或隐藏内容?检查工作说明是否要求 AI 忽略你的指令、干扰检查结果,或夹带看不见的文字。发现 1 项风险
中风险

隔离子代理上下文可能遗漏用户的授权边界和安全限制

原文依据:4 处
发现了什么

Skill 要求子代理绝不继承会话历史,而是只接收协调者构造的内容。虽然它要求任务自包含并包含约束,但没有明确要求复制用户的授权范围、安全限制或禁止操作。

为什么需要注意

如果协调者遗漏关键限制,子代理可能把“修复测试”等任务理解为有权修改用户未授权的文件或行为;后续摘要和测试可能不足以暴露越界改动。

这是一个取决于协调者所写提示的授权风险。Skill 明确要求子代理绝不继承会话上下文,而只获得协调者挑选的内容;因此,若协调者没有转述用户的文件范围、禁止操作或审批要求,子代理可能在不了解这些边界时行动。要求任务“自包含”并加入约束能降低风险,但列出的约束主要是代码范围示例,没有明确规定必须保留用户的全部授权和安全限制。用户可要求委派提示逐项包含原始权限边界,并在整合前核对。

SKILL.md:10来自说明文档打开原文件
You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.
查看另外 3 个位置
SKILL.md:60来自说明文档打开原文件
Each agent gets:- **Specific scope:** One test file or subsystem- **Clear goal:** Make these tests pass- **Constraints:** Don't change other code- **Expected output:** Summary of what you found and fixed
SKILL.md:89来自说明文档打开原文件
Good agent prompts are:1. **Focused** - One clear problem domain2. **Self-contained** - All context needed to understand the problem3. **Specific about output** - What should the agent return?
SKILL.md:123来自说明文档打开原文件
**❌ No constraints:** Agent might refactor everything**✅ Constraints:** "Do NOT change production code" or "Fix tests only"
会不会偷偷改推广链接或收款方?检查是否强制替换推广链接或收款对象,同时要求隐瞒更改。未发现风险

Skill 逻辑拆解

8 个说明模块

该 Skill 是一套并行委派工作流程:先把独立故障分组,再为每个子代理限定范围、目标、约束和返回内容,并发处理后统一审查和集成。

查看原文
SKILL.md:49来自说明文档打开原文件
### 1. Identify Independent DomainsGroup failures by what's broken:- File A tests: Tool approval flow- File B tests: Batch completion behavior- File C tests: Abort functionalityEach domain is independent - fixing tool approval doesn't affect abort tests.### 2. Create Focused Agent TasksEach agent gets:- **Specific scope:** One test file or subsystem- **Clear goal:** Make these tests pass- **Constraints:** Don't change other code- **Expected output:** Summary of what you found and fixed
SKILL.md:79来自说明文档打开原文件
### 4. Review and IntegrateWhen agents return:- Read each summary- Verify fixes don't conflict- Run full test suite- Integrate all changes

Skill 声称子代理不应继承当前会话,而应只收到协调者专门构造的上下文。这有助于聚焦,但实际保留哪些用户限制取决于协调者编写的任务内容。

查看原文
SKILL.md:10来自说明文档打开原文件
You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.
SKILL.md:89来自说明文档打开原文件
Good agent prompts are:1. **Focused** - One clear problem domain2. **Self-contained** - All context needed to understand the problem3. **Specific about output** - What should the agent return?

该 Skill 明确不建议在故障相关、需要完整上下文、仍属探索性调试或会共享状态时并行,并要求集成后运行完整测试和抽查。

查看原文
SKILL.md:129来自说明文档打开原文件
## When NOT to Use**Related failures:** Fixing one might fix others - investigate together first**Need full context:** Understanding requires seeing entire system**Exploratory debugging:** You don't know what's broken yet**Shared state:** Agents would interfere (editing same files, using same resources)
SKILL.md:161来自说明文档打开原文件
## VerificationAfter agents return:1. **Review each summary** - Understand what changed2. **Check for conflicts** - Did agents edit same code?3. **Run full suite** - Verify all fixes work together4. **Spot check** - Agents can make systematic errors
从这里开始 · 工作说明SKILL.md
dispatching-parallel-agents
连线表示工作说明包含的模块,不是实际运行顺序。点击模块可查看原文。
文件与检查记录1 个文件

检查范围与遗漏

逐文件查看涉及的内容

下方列出本次涉及的原文范围;纳入检查不代表已查清所有问题。

  • SKILL.md已纳入全文

这份报告只针对上方版本。我们看了拿到的代码和说明文件,没有实际运行 Skill,也没有检查它另外安装的软件包。因此,这不是“保证安全”的承诺;换了版本或使用环境,结果也可能不同。

  • SKILL.md工作说明

代码和说明中提到的操作

读取文件
SKILL.md:103来自说明文档打开原文件
1. Read the test file and understand what each test verifies2. Identify root cause - timing issues or actual bugs?
读取了多少行
168
文件校验值(用于核对版本)
652ed5754e364376e10c16eb49cabcd1b2d5d009be54e365c938b412919ac5a8