Back to skill

Security audit

Feishu Agent Team

Security checks for vulnerabilities and agentic risk

Overview

This looks like an incomplete Feishu multi-agent routing prototype rather than malware, but its documentation asks users to set up broad external chat automation and extra dependencies that the shipped code does not actually implement.

Install only after treating this as prototype-quality: pin and review dependencies, remove unused packages, avoid placing real Feishu app secrets or sensitive group content into it until access controls and privacy handling are added, and verify the missing Feishu/MQ/LangGraph pieces before running background listeners.

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
SKILL.md:50
Finding

Unpinned Third-Party Dependencies Create a Mutable Software Supply Chain

Content
View full analysis

Vulnerability Details

File Locations:

  • SKILL.md:50
  • README.md:24

Vulnerability Type: Unpinned and unverifiable third-party dependencies
Risk Level: Medium

Affected code snippets:

SKILL.md:50

bash
pip install langgraph langchain-openai feishu-oapi

README.md:24

bash
pip install pyyaml

Technical Analysis

The installation instructions retrieve mutable package versions from the configured Python package index without version constraints, integrity hashes, or a reviewed lockfile. As a result, the code installed by these commands can differ from the code available when the Skill was audited.

Python package installation can execute package build and installation logic. A compromised upstream release, dependency takeover, malicious transitive dependency, or unsafe package-index configuration could therefore result in arbitrary code execution during installation or later import.

The risk is increased by the discrepancy between the supplied implementation and the dependencies requested in SKILL.md. The audited code directly uses PyYAML, but the documented langgraph, langchain-openai, and feishu-oapi integrations are not implemented in the supplied project. These unnecessary dependencies expand the supply-chain attack surface without supporting current functionality.

No evidence was found that the currently named packages are malicious. The vulnerability is the absence of reproducible version and integrity controls.

Attack Path

  1. A user follows the documented installation instructions.
  2. pip resolves the latest compatible versions from the user's configured package index.
  3. An upstream package, transitive dependency, or configured package source supplies a compromised release.
  4. Malicious build or installation logic executes during pip install, or malicious package code executes when imported.
  5. The payload runs with t ...[truncated 695 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove langgraph, langchain-openai, and feishu-oapi from the installation instructions unless the corresponding integrations are implemented and required.
  2. Pin every direct dependency to an explicitly reviewed version.
  3. Resolve and lock transitive dependencies using a reproducible dependency-management process.
  4. Generate and verify cryptographic hashes, such as by installing from a hash-locked requirements file with pip install --require-hashes.
  5. Use a trusted, explicitly configured package index and prevent unintended fallback to untrusted indexes.
  6. Install dependencies in an isolated virtual environment under a non-privileged account.
  7. Keep README.md, SKILL.md, Skill metadata, and the actual implementation synchronized so users install only dependencies required by the shipped code.
  8. Add automated dependency scanning and periodic review of pinned package versions.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

声明描述的是一个完整的多 Agent 协作/调度系统,核心在于 Coordinator、飞书群聊 @mention 触发、自动路由和团队编排。而提供的代码片段仅是一个通用监听器骨架,面向单个 agent 轮询消息并返回简单字符串结果,再通过打印模拟发送到群。虽然导入了 TeamRouter,但并未实际使用任何路由逻辑;飞书接入和 MQ 获取也都是 TODO 占位。代码没有表现出与声明相符的主要能力与触发方式,因此描述与实际行为存在实质性不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

声明描述的是一个较完整的多 Agent 协作/调度系统入口,包含消息触发、Coordinator 调度、可用于自动化工作流团队的能力。实际代码块只覆盖了其中很小一部分:从配置文件读取 specialist 列表,并根据任务文本中是否包含关键词返回某个专家名称。它没有实现 Feishu 触发、没有调用或编排多个 Agent、没有展示 OpenClaw 相关接口,也没有任何团队协作执行逻辑。因此该代码的实际行为只是“配置驱动的简单路由组件”,与声明的整体技能用途相比存在实质性不匹配。

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
72% confidence
Finding

The skill declares executable capabilities and installation/use of Python-based components, but it does not define an explicit tool scope such as permissions or allowed-tools. In an agent ecosystem, missing scope boundaries can allow the runtime or users to assume broader file access than intended, increasing the chance of unintended local file reads or over-privileged execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill encourages sending group-chat task content through Feishu and across multiple agents, but it does not warn users that potentially sensitive data will leave the immediate local context and be propagated to external services and additional processing components. This can lead to accidental disclosure of confidential business, personal, or regulated information when users assume the workflow is self-contained.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The coordinator is triggered by a simple group-chat @mention name with no visible scope restrictions, sender validation, channel allowlist, or exclusion conditions. In a multi-agent routing skill, this can let any participant invoke orchestration behavior, increasing the risk of prompt abuse, unwanted task execution, spam-triggered workflows, or accidental delegation to specialist agents.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This file uses Chinese-only natural-language descriptions and runtime messages, including the module docstring and command-line help text, with no indication that the skill is intended only for Chinese-speaking users or that language selection is optional. That creates a language/locale policy issue because the skill implicitly enforces a specific language without opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The argparse description and help text are presented only in Chinese, and the runtime status messages elsewhere in the file follow the same pattern. For a general-purpose skill, this is a natural-language policy violation unless the locale restriction is explicitly documented and justified or users can choose their language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The module docstring includes the product name followed by a Chinese-only description, which indicates a fixed language presentation. For an all-file-types policy check, this can be a locale/language policy violation when no user choice or opt-in is offered.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This code uses Chinese-only natural-language descriptions and output strings such as the module description, docstrings, exception text, and printed test output. Under the policy, forcing a specific language without user opt-in or documented justification is a natural-language policy violation.

Content

No source excerpt is available for this finding.