Back to skill

Security audit

Boss

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent as a development automation workflow, but it can automatically modify projects, run/install tools, start services, and deploy with weak scoping and easy trigger phrases.

Review this skill carefully before installing. Use explicit /boss invocation, avoid --quick, prefer --skip-deploy until you have reviewed the generated plan, and run it only in a disposable or version-controlled workspace. Require approval before package installs, global npm installs, docker-compose, service startup, or any deployment/external URL exposure.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
agents/boss-qa.md:425
Finding

Unpinned Third-Party Package Retrieval and Execution

Content
View full analysis

Vulnerability Details

File Location:

  • agents/boss-qa.md:425
  • agents/boss-qa.md:459-463
  • agents/boss-qa.md:479-490
  • agents/boss-qa.md:553-558
  • agents/boss-devops.md:45

Vulnerability Type: Supply-chain exposure through unpinned npm dependencies and runtime package execution
Risk Level: Medium

Vulnerable Code

From agents/boss-qa.md:

bash
pnpm test --coverage 2>&1 || npm test --coverage 2>&1 || npx vitest run --coverage 2>&1
bash
# Frontend Playwright
npx playwright test 2>&1
# Frontend Cypress
npx cypress run 2>&1
# Backend API tests
pnpm test:e2e 2>&1 || npm run test:e2e 2>&1
bash
# Install Playwright
npm install -D @playwright/test
npx playwright install

# Or install Cypress
npm install -D cypress

# Install agent-browser (optional)
npm install -g agent-browser
agent-browser install
bash
# Check browserslist configuration
npx browserslist

# Run multi-browser tests
npx playwright test --project=chromium --project=firefox --project=webkit 2>&1

From agents/boss-devops.md:

text
| Static HTML | npx serve | 3000 |

Technical Analysis

The QA and deployment instructions direct the Agent to install or execute npm packages without exact version constraints, integrity verification, or a requirement that packages already exist in the project's lockfile.

When a referenced package is unavailable locally, npx may download and immediately execute it from the configured npm registry. Commands such as npm install -D @playwright/test, npm install -D cypress, and npm install -g agent-browser similarly resolve mutable package versions and execute package lifecycle scripts. The global agent-browser installation additionally modifies the user's shared environment rather than remaining isolated to the audited project.

These operations are releva ...[truncated 2116 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin every testing and browser automation dependency to an reviewed exact version.
  2. Commit and enforce a package lockfile, then install dependencies with npm ci or an equivalent frozen-lockfile command.
  3. Replace runtime downloads with project-local execution:
    • Use package scripts backed by committed dependencies.
    • Use npx --no-install where appropriate.
    • Fail safely when the required local binary is unavailable.
  4. Remove npm install -g agent-browser. Install a pinned version as a local development dependency or run it in an isolated, pinned container.
  5. Require explicit user approval before:
    • Adding or updating dependencies.
    • Downloading browser binaries.
    • Installing global tooling.
    • Running package lifecycle scripts.
  6. Disable lifecycle scripts during initial installation when feasible, review the package contents, and enable only the scripts that are required.
  7. Use an approved registry, package allowlist, integrity verification, and available npm provenance controls.
  8. Prefer existing project test runners rather than falling back automatically to remotely resolved npx packages.
  9. Replace npx serve with a pinned project-local server dependency or a trusted runtime component already present in the environment.
  10. Run downloaded development tools in a sandbox with restricted filesystem and network access.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (46)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The --quick option explicitly skips confirmation nodes and enables fully automatic execution, but the README gives no strong warning about the resulting loss of human review. In a skill that orchestrates multiple agents through development and deployment stages, this materially increases the risk of unintended code changes, unsafe command execution, or accidental deployment.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

该技能宣称会自动编排从需求到部署的完整研发流程,但文档内容主要是流程说明与产物落盘要求,缺少可验证的实际编排约束与安全边界。这种能力描述与真实行为不一致会误导用户授予过高信任,且还弱化了对写文件、可能部署、以及潜在覆盖现有产物等系统影响操作的警觉。

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

触发词包含“帮我做一个”“build this”“new feature”等高频日常表达,极易在普通对话中误触发该高权限编排技能。一旦误触发,技能会进入自动化研发/写产物/潜在部署流程,使用户在未充分知情或未明确授权的情况下发生系统级操作。

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · agents/boss-architect.md (reported line 485)May include surrounding context.

md
| 环境 | 用途 | 数据库 | 配置 |
|------|------|--------|------|
| local | 本地开发 | SQLite/Docker | .env.local |
| dev | 开发测试 | 测试数据库 | .env.development |
| prod | 生产环境 | 生产数据库 | .env.production |

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · agents/boss-architect.md (reported line 486)May include surrounding context.

md
| 环境 | 用途 | 数据库 | 配置 |
|------|------|--------|------|
| local | 本地开发 | SQLite/Docker | .env.local |
| dev | 开发测试 | 测试数据库 | .env.development |
| prod | 生产环境 | 生产数据库 | .env.production |

### 7.3 监控告警

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · agents/boss-architect.md (reported line 487)May include surrounding context.

md
|------|------|--------|------|
| local | 本地开发 | SQLite/Docker | .env.local |
| dev | 开发测试 | 测试数据库 | .env.development |
| prod | 生产环境 | 生产数据库 | .env.production |

### 7.3 监控告警

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The design describes a fully automated pipeline that can proceed through deployment and produce a reachable URL, but the skill description does not clearly warn users about that capability. This reduces informed consent and increases the chance that a user invokes the skill without realizing it may perform irreversible or externally exposing actions, which is more dangerous in an automation-heavy orchestration skill than in a read-only planning assistant.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger phrases are broad natural-language prompts such as generic development requests, which can cause the skill to activate when the user did not explicitly intend to launch a full autonomous pipeline. In this skill’s context, accidental activation is especially risky because the documented workflow includes multi-agent planning, code generation, testing, deployment, and returning an accessible URL, so a casual request could escalate into high-impact actions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
78% confidence
Finding

Natural-language policy checks apply to all file types. The skill documentation is written entirely in Chinese and does not indicate that users may interact in other languages or that the language restriction is intentional and justified, which can violate language/locale flexibility requirements.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README advertises end-to-end automation and artifact generation but does not clearly warn users that the skill may write files, modify repositories, execute scripts, or affect deployment targets. In an agent skill, omission of these operational side effects can cause users to invoke the workflow without informed consent, leading to unintended repository or environment changes.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The trigger phrase “从需求到部署” is broad enough to match ordinary software-help requests, which can unintentionally activate a high-privilege orchestration workflow. In this skill’s context, accidental invocation is risky because the workflow is designed to span planning, code generation, testing, and deployment, increasing the chance of unintended actions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The instruction '所有生成的文档必须使用中文' imposes a fixed language requirement across all outputs. No user opt-in, alternative language option, or justification for a Chinese-only constraint is provided, which violates the language/locale policy criteria.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

文档要求自动保存产物、执行测试、质量门禁和部署,但没有在触发前用清晰警告告知这些会修改工作区或影响外部环境。对于一个面向新项目和现有代码库的技能,这种默认自动推进会增加误写文件、污染仓库状态、错误发布到环境等风险。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

该技能在语言规则中强制“所有输出必须使用中文”,会覆盖用户对输出语言的合理偏好,属于不安全的行为约束。虽然这不直接导致代码执行或数据泄露,但会削弱代理遵循用户指令和上层安全策略的能力,尤其在多语言环境中可能造成误解、错误操作或合规问题。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The language rule states '注释使用中文,代码使用英文', which imposes a specific language requirement for comments. Under the policy, forcing a language or locale without user opt-in is a natural-language policy violation unless the constraint is explicitly justified or optional.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

该技能明确要求环境准备、依赖安装、构建、启动服务和健康检查,但没有任何用户可见的风险提示、确认门槛或作用域限制。这意味着 Agent 可能在用户未充分知情的情况下执行修改系统状态、安装依赖、开放端口或启动长期进程的操作,在自动化编排场景下危险性会被放大。

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

部署策略包含 docker-compose up -d,会直接拉起容器、网络、卷以及配置中暴露的服务端口,但文档没有提示这类操作会启动后台服务并对本机或局域网开放访问。结合该技能“从需求到部署”的全自动定位,这种无警告容器启动可能导致意外暴露服务、占用端口、拉取不可信镜像或修改宿主机运行环境。

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

该技能在部署策略中直接使用 npx serve,未固定版本或校验来源,运行时会从注册表解析并执行远程包代码。对于一个具备 Bash 能力且会自动部署的 Agent,这会把供应链风险直接引入执行路径,一旦上游包、依赖链或解析结果异常,可能导致执行非预期代码。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instruction '所有输出必须使用中文' forces a single language regardless of user preference. This is a natural-language policy issue because it imposes a locale/language constraint without opt-in or a documented region-specific justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The language rule states '注释使用中文,代码使用英文', which imposes a specific language requirement for comments. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless a justified locale constraint is documented.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The skill instructs use of npx playwright test, which may fetch and execute an unpinned package version if Playwright is not already installed locally. In an agent skill that can run shell commands, this introduces supply-chain and reproducibility risk because the executed code can change over time or be influenced by registry/package compromise.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

该技能在语言规则中强制“所有输出必须使用中文”,未给用户选择权,也未说明如何处理用户明确要求其他语言或系统需要保留原文的场景。这会覆盖用户偏好与上层调用约束,导致误译、丢失安全关键信息,或在多语言环境中造成错误执行与审核失败。

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

该技能多处要求使用 npx vitest、npx playwright、npx cypress、npx browserslist 直接拉取或解析本地/远程包而未固定版本,增加了供应链漂移和意外执行恶意包代码的风险。由于该 Agent 还被设计为“必须真正执行测试”,这些命令更可能在自动化环境中被直接运行,从而把文档中的不安全执行模式转化为真实攻击面。

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

该处使用未固定版本的 npx playwright,会依赖当时解析到的包版本或触发下载安装流程,存在供应链投毒、版本漂移和执行非预期代码的风险。结合该技能要求 Agent 自动执行浏览器测试,这种风险在实际运行时更高。

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

该处推荐未固定版本的 npx cypress run,可能触发从注册表获取最新或被污染版本的 Cypress/相关依赖并立即执行。对具备 Bash 工具且鼓励真实执行测试的 Agent 来说,这会把普通文档示例升级为可利用的执行链入口。

Content

No source excerpt is available for this finding.

Static analysis

Detected: suspicious.exposed_secret_literal

File appears to expose a hardcoded API secret or token.

Critical
Code
suspicious.exposed_secret_literal
Location
agents/boss-backend.md:286

File appears to expose a hardcoded API secret or token.

Critical
Code
suspicious.exposed_secret_literal
Location
agents/boss-qa.md:265