Back to skill

Security audit

Meeting Notes Generator

Security checks for vulnerabilities and agentic risk

Overview

The skill is a meeting-minutes helper, but it directs full meeting documents to DeepSeek for exhaustive processing without a clear consent or redaction gate, and it ships auxiliary packaging code with unsafe archive handling.

Review before installing or using with confidential material. Do not submit regulated, personal, legal, financial, or internal meeting transcripts unless DeepSeek processing is approved for your organization. Treat the included scripts as developer utilities only, and avoid verifying untrusted .skill archives or packaging directories that may contain secrets until the archive extraction and file inclusion logic is fixed.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (3)

other

Warning
Location
SKILL.md:34
Finding

Uncontrolled Disclosure of Meeting Content to an External AI Service

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/package_skill.py:62
Finding

Recursive Packaging Can Include Secrets and Unintended Files

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/package_skill.py:132
Finding

Unsafe Tar Archive Extraction During Package Verification

Content
View full analysis
Optional[str]: """解压技能包(用于验证)""" package_path = Path(package_path) if not package_path.exists(): print_error(f"包文件不存在: {package_path}") return None # 创建临时目录 temp_dir = Path(tempfile.mkdtemp(prefix="skill_extract_")) extract_path = temp_dir / package_path.stem.replace('.skill', '') extract_path.mkdir(exist_ok=True) try: with tarfile.open(package_path, "r:gz") as tar: tar.extractall(extract_path) ``` ### Technical Analysis The verification functionality accepts a package selected by the user and extracts all archive members with `tar.extractall()` without explicitly validating member paths or types. On Python runtimes where a safe extraction filter is not enabled by default, a malicious tar archive can contain: - Absolute paths. - Paths containing `../` traversal components. - Symbolic links or hard links targeting locations outside the extraction directory. - Special file entries. A traversal entry can escape the newly created temporary directory and overwrite another file writable by the current user. Link-based entries can potentially produce a similar result when later archive members are written through an extracted link. Creating a unique temporary directory does not mitigate this issue because the archive controls member paths interpreted during extraction. ### Attack Path 1. An attacker creates a `.skill` tar archive containing a member such as `../../target-file`. 2. The attacker sends the archive to a user for verification. 3. The user runs: ```text python package_skill.py --verify attacker.skill ``` 4. `verify_package()` calls `extract_package()`. 5. `tar.extractall(extract_path)` processe ...[truncated 1022 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (23)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个“会议纪要整理专家”,核心能力应是处理会议文本/录音转写并输出详细纪要。但实际代码是 package_skill.py,其主要功能是技能发布相关的工程脚本:读取 _meta.json,遍历技能目录,排除部分文件后创建 tar.gz 格式的 .skill 包,并支持解压验证包内容。代码没有任何自然语言处理、摘要、会议要点提取、字数控制或文档整理逻辑。另有文件系统访问和压缩/解压能力,这些都未在声明用途中体现,且与会议纪要生成这一主要用途无关。因此属于明显的描述与实际行为不一致。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个面向最终用户的会议纪要生成/整理技能,核心能力应是理解会议文本并输出详细纪要。代码却完全没有任何自然语言处理、摘要生成、文档解析、会议内容提取或字数控制逻辑。相反,它是一个开发/打包阶段使用的验证工具,检查技能文件是否存在、frontmatter字段是否正确、版本号格式是否合规、目录/命名是否规范,并扫描潜在敏感文件和符号链接。其主要用途与触发场景都与“整理会议记录”明显不一致,因此属于明显的描述与行为不匹配。

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

该描述将“帮我整理会议记录”“会议总结”“这个会议说了什么”等常见自然语言表述列为触发短语,并使用“包括但不限于”进一步放宽范围,缺少清晰边界。这类表达与普通对话高度重叠,容易在非预期场景下触发技能。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

文件整体以中文规定技能行为与输出预期,但未说明用户可选择其他语言或根据输入语言自适应输出。根据规则,未经用户选择而隐含强制特定语言/locale 属于自然语言政策问题。

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instruction to ensure complete sentence-by-sentence coverage and avoid omitting any information encourages wholesale reproduction of source material, including confidential names, decisions, credentials, financial figures, or other sensitive meeting content. In a meeting-minutes skill, this context makes the issue more dangerous because uploaded transcripts often contain internal or regulated information that should be minimized, redacted, or selectively summarized.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The mandated 100% coverage check, mapping every paragraph, and zero-omission validation explicitly instruct the model to surface all original content and verify nothing has been left out. That creates a data-exposure risk because any sensitive content that might otherwise be prudently suppressed will instead be forced into the output and validated for inclusion.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The forced supplementation step requires re-inserting every omitted item and correcting any missing detail, which defeats safe summarization and redaction. In the context of meeting records, this can reintroduce confidential, personal, or legally sensitive statements that were initially excluded for safety or relevance reasons.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description defines a very broad activation scope around meeting-note generation and document processing, with many trigger phrases and strong guarantees like '100% no omission' and minimum length requirements. This can cause the skill to activate in unintended contexts, increasing the chance of processing sensitive uploaded documents or overriding normal user intent with overly expansive summarization behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The metadata presents the skill entirely in Chinese and implies a fixed output style and locale without indicating that users may choose another language. While not directly exploitable like code execution, forced locale behavior can cause user confusion, mishandling of multilingual documents, or incorrect assumptions about how sensitive meeting content will be transformed and presented.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file title and all prompt templates require outputs in Chinese and prescribe exact Chinese formatting, but do not provide any user opt-in or language selection mechanism. This is a natural-language policy concern because it imposes a specific language/locale on users across the skill content without an explicit documented justification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt library includes a company-name verification and business-research template that is unrelated to the declared meeting-minutes function. This creates unjustified scope expansion and can cause the agent to perform external-knowledge synthesis or simulated research outside the user-approved task boundary, increasing the risk of data leakage, hallucinated verification, and misuse.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The template explicitly instructs the model to validate company names using broad knowledge, simulated searches, news, financial reports, and public business information, which is not necessary for meeting-minute generation. That unnecessary research behavior widens the attack surface by encouraging unsupported claims and unintended handling of external or sensitive business context beyond the skill’s intended purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The document is entirely written as a prescriptive format specification for meeting minutes in Chinese, including Chinese title templates, date formats, and naming examples, but it does not indicate that this locale restriction is optional or limited to a China-specific use case. Under the policy, forcing a specific language or locale without user 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
81% confidence
Finding

The script's natural-language strings and CLI messaging are entirely in Chinese, which imposes a specific language on users without any opt-in or fallback. The provided policy says locale/language constraints should not be forced unless choice is offered or the restriction is clearly justified.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The exclusion logic does not implement the documented glob-style patterns correctly: checks like startswith(p.replace('','')) will not match entries such as '.gitignore', and patterns like '.pyc'/'*.pyo' are ineffective. As a result, hidden files or build artifacts that operators believe are excluded may be silently included in the generated package, potentially leaking metadata or sensitive content during distribution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script extracts a tar archive using tar.extractall on a package path without validating member paths. A crafted archive can exploit path traversal or link entries to write outside the intended extraction directory, making this a real filesystem-write vulnerability rather than merely a missing warning/confirmation issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This Python file contains natural-language strings and documentation entirely in Chinese, including the module docstring and later CLI output, which indicates the skill forces a specific language for users. The policy allows locale constraints only when explicitly justified or when users can opt in, neither of which is present here.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

该文件实现的是对技能目录进行 frontmatter、_meta.json、目录结构、命名和敏感文件检查的本地验证流程,而不是整理会议记录、生成会议纪要或提取会议要点。结合清晰的模块/函数命名与执行流程可见,这属于技能打包校验工具,与清单中声明的会议纪要生成功能存在明显语义不一致。

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

In the processing rules, L060 says not to generate meeting-background descriptions such as time information and to keep only structured content, but the documented output template at L177-L181 explicitly requires meeting time, location, participants, and host. These instructions conflict and create intent ambiguity about what the skill is actually supposed to output.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The output format hard-codes Chinese section headers and evaluation labels such as '通过/不通过', requiring a single locale for the response. Because no alternative language path or opt-in is offered, this creates a language-policy violation under the stated rule.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

模块文档字符串首行写明“会议纪要生成器 - 技能验证脚本”,其中“会议纪要生成器”暗示其承担会议纪要生成能力;但后续代码仅检查 SKILL.md、_meta.json、目录结构、命名与敏感文件。该文档表述会使读者误解脚本用途,属于文档意图与实现行为不一致。

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

控制台输出标题重复使用“会议纪要生成器”这一表述,但 run_validation 实际仅执行校验步骤并输出通过/失败结果。该运行时文案会主动传达与真实功能不符的用途说明。

Content

No source excerpt is available for this finding.