Back to skill

Security audit

发布(智能体一键部署与上架)

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to perform its stated Coze bot deployment workflow, with expected use of a Coze API token and document upload, but users should review privacy and credential handling before running it.

Install only if you intend to let the skill use your Coze API token to create and publish bot resources and upload the chosen knowledge-base files to Coze. Use a minimally scoped token, review the kb directory before running, avoid including secrets or unnecessary personal data, and add privacy/consent wording to any customer-facing appointment or lead-capture templates.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (22)

Tainted flow: 'headers' from os.environ.get (line 44, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · run_all_tests.py (reported line 49)May include surrounding context.

python
"Content-Type": "application/json",
        "Agw-Js-Conv": "str",
    }
    resp = requests.post(api_url, headers=headers, data=json.dumps(payload).encode("utf-8"), timeout=timeout)
    raw = resp.text
    answers = []
    lines = raw.split("\n")

Tp4

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

描述的大方向与代码主功能基本一致:确实是在调用 api.coze.cn 相关接口执行知识库创建、文档上传、Bot 创建、知识库绑定和发布,属于“扣子智能体一键部署与发布”。但声明将能力扩展到了完整上架链路,包括自动回归测试、上架文案生成、提审材料生成、发布到商店等,而当前代码块并未实现这些功能。代码中仅有一条提示说明“随后可跑 run_all_tests.py 回归”,这不能算已经实现测试能力;也没有任何生成文案/材料或商店上架接口调用。因此这是描述夸大能力的情况,属于不准确描述。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个覆盖“创建-配置-绑定-发布-测试-上架材料生成”的端到端自动部署工具;但提供的代码片段只是在已有 bot_id 前提下执行 12 个聊天回归测试。它调用的唯一接口是 /v3/chat,没有任何创建知识库、上传文件、创建或发布 bot 的 API 调用,也没有生成上架文案或提审材料的逻辑。测试能力本身与声明中的“12 用例自动回归测试”部分一致,但整体主用途与大部分关键能力均未在代码中体现,因此属于明显不匹配。

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The restaurant template states that reservation information is pushed directly to the owner's WeChat, but gives no user-facing warning, consent mechanism, or access-control guidance. This is more dangerous than generic data collection because it implies transmission of personal reservation details into a potentially informal messaging channel, increasing the risk of unauthorized access, oversharing, and uncontrolled retention.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill describes behaviors that require environment access, local file reads, and outbound network calls, but it does not declare any explicit tool scope or permissions boundary. This creates a least-privilege and transparency problem: a caller may invoke the skill without understanding it can access credentials, read documents from disk, and make authenticated API requests to external services.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

该描述把触发条件写成“当用户需要……等需求时使用此技能”,其中包含“搭建 Bot”“把方案做成可上架的智能体”等较宽泛表述,且使用“等需求时”扩展了范围,没有给出排除条件或负例。这类触发说明在 manifest/markdown 中容易与常见对话意图重叠,难以判断何时应激活、何时不应激活。

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs users to provide an API token but does not clearly warn about its sensitivity, storage handling, rotation, or limits on exposure. In a workflow that performs authenticated external requests and reads tokens from environment variables, poor credential guidance materially raises the risk of token leakage, misuse, and unauthorized bot or knowledge-base changes.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

文档将特定场景下的回复固定为中文句子“我只能为您生成 AI 客服机器人的功能设计方案”,属于语言/locale 约束。文件中未说明该技能仅面向中文场景,也未给用户提供语言选择或显式同意机制,因此构成自然语言层面的语言策略风险。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

第 6 条把注入防御时的统一回复写死为中文固定文案,等同于在所有场景下强制特定语言输出。由于文档没有给出语言选择、地区限制或中文场景限定,这与语言/locale 需经用户选择或合理限定的政策要求不符。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The example opening script is written as a fixed Chinese greeting and interaction pattern, with no indication that the user can choose another language or locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The medical-aesthetics template explicitly collects name, phone number, project, and appointment time, but does not instruct the bot to present any privacy notice, consent step, or data-minimization rule. In this context, the bot is designed for customer intake, so silent collection of contact and scheduling data can expose users to privacy harm, improper retention, and compliance violations, especially in a sensitive-service industry.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The education template requests name, contact details, and booking time for a trial lesson without any notice about data handling or consent. Because this is a lead-capture workflow and may involve minors indirectly through parents/students, the lack of privacy messaging and collection safeguards increases the risk of misuse, over-collection, or noncompliance.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The SME template includes lead capture for name and business need, but does not tell users how their information will be handled. In a deployment-and-publishing skill whose purpose is to create real customer-facing bots, omission of privacy guidance is operationally significant because the template is likely to be reused directly in production.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

This plain-text template includes the opener “产品、价格、合作方式都可以问我”, which is a very broad activation-style description and does not define boundaries for when the skill should or should not engage. Because the file provides reusable industry templates and no negative examples or scope limits, the trigger surface is overly open-ended.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill reads arbitrary .md/.txt files from a local directory, base64-encodes their full contents, and uploads them to a remote Coze API endpoint without any explicit user confirmation, warning, or scoping check at the point of transfer. In this skill's context, knowledge-base documents are likely to contain business documents, customer-service playbooks, or potentially sensitive operational data, so silent transmission to a third-party service creates a real data exposure and privacy/compliance risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The onboarding prologue and suggested questions are fixed in Chinese, which forces a specific language for user-facing interaction. Under the policy, language constraints should either be user-selectable or clearly justified as region-specific; this file provides neither.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The prompt contains detailed behavior and output rules, but inside the skill file it does not clearly state strict invocation boundaries, allowed inputs, or explicit out-of-scope conditions. In an agentic environment, this can cause the skill to activate on loosely related requests and produce business-facing artifacts in the wrong context, increasing the risk of unintended actions, misrouting, or policy bypass through incorrect skill selection.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction '全程简体中文' imposes a fixed language/locale policy on all interactions. Under the stated policy criteria, forcing a specific language without user opt-in is a natural-language policy violation unless the restriction is clearly justified or optional.

Content

No source excerpt is available for this finding.

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 · run_all_tests.py (reported line 16)May include surrounding context.

python
if not token:
    raise SystemExit("缺少扣子开放 API Token 凭证")

api_url = "https://api.coze.cn/v3/chat"
DEFAULT_OUT = "./test_output"

CASES = [

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

该文件的上架文案把技能定位为根据行业与需求生成《功能设计清单》和报价建议的效率工具,核心输出是方案文档,而不是执行 bot 的创建、部署、发布和测试流程。这与 manifest 中强调的“通过扣子开放 API 一键部署并发布 AI 客服智能体”的完整链路存在明显语义偏差。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The sample opening message begins in English and presents a fixed bilingual script rather than offering the user a choice of language. Under the policy rule, language or locale constraints should be optional or explicitly justified, and this line does not indicate user opt-in for the language used.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The file provides user-facing opening and sales scripts entirely in Chinese, and does not indicate that language should be selected based on user preference. Under the stated policy, forcing a specific language without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.