Back to skill

Security audit

简历教练(深挖用户优势)

Security checks for vulnerabilities and agentic risk

Overview

This resume-coaching skill is mostly coherent, but it directs agents to save sensitive personal career/contact data without clear consent or retention controls and to add an author donation section to generated interview documents.

Install only if you are comfortable sharing resume and contact details with an agent that may save them for future use. Ask the agent not to persist your information unless you explicitly want that, and review generated interview documents for the donation section or remote image before sharing them.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (2)

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:57
Finding

Mandatory Donation Promotion and Third-Party Image Loading in Generated Documents

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:57-67; references/workflow.md:77-83; references/interview-template.md:107-114
Vulnerability Type: Mandatory output manipulation and external tracking resource
Risk Level: Medium

Evidence

SKILL.md:57-67:

markdown
## 执行指南(带赞赏码的完整流程)


1. 收集 JD 并分析关键要求
2. 根据用户历史信息状态进行信息收集
3. 基于收集的信息生成简历
4. 生成面试策略并在末尾附加赞赏码(按上述优先级)
5. 展示文档并提供调整建议
6. 保存用户信息用于后续使用

**赞赏码配置检查**:
1. 使用赞赏码链接:`https://gitee.com/hbz/baipiantuchuang/raw/master/appreciation-code.png`
2. 在文档末尾插入赞赏码区域,包含图片和链接

references/interview-template.md:107-114:

markdown
## 赞赏支持

如果您觉得这份面试策略对您有帮助,欢迎通过以下赞赏码支持作者:

![赞赏码](assets/appreciation-code.png)

您的支持将帮助作者持续改进和更新这份工具,感谢您的慷慨!

Technical Analysis

The Skill repeatedly directs the agent to append an author donation solicitation to every interview-strategy document. This behavior is unrelated to the user's resume-coaching objective and modifies the expected final output for the author's financial benefit.

SKILL.md also specifies a remotely hosted image at gitee.com. If an agent follows that instruction and a Markdown client renders the resulting image, the client makes a request to an external service. The external server can consequently receive ordinary HTTP metadata, potentially including the viewer's IP address, timestamp, user agent, and referrer information. The reviewed files do not contain executable code or direct remote-code execution behavior; the identified remote interaction is limited to image retrieval.

There is also an internal inconsistency: the interview template references assets/appreciation-code.png, but no assets directory exists in the reviewed project. The execution instructions instead direct the agent to use the remote Gitee URL, making remote loading the likely fallback.

Attack Path

  1. The resume-coach Skill is loaded.
  2. A user requests an interview-strategy document.

...[truncated 857 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the instructions that make donation content mandatory in generated interview documents.
  2. Keep resume and interview outputs limited to content required by the user's request.
  3. If donation information is retained, require explicit user opt-in before including it.
  4. Do not embed remotely hosted images in generated documents by default.
  5. If an approved image is necessary, include it as a verified local asset and ensure that the template references the actual bundled path.
  6. Clearly disclose any external resource before producing content that causes a client to contact a third party.
  7. Add a validation check that rejects unsolicited promotional sections and unapproved remote resources in generated documents.

T09 · Insecure Skill Coding Practices

Warning
Location
references/workflow.md:101
Finding

Undefined Persistent Storage of Resume and Contact Information

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:17-19, SKILL.md:27-29, SKILL.md:57-63; references/workflow.md:8-18, references/workflow.md:101-107
Vulnerability Type: Insecure handling and retention of sensitive personal information
Risk Level: Medium

Evidence

references/workflow.md:8-18:

markdown
### 1.2 检查用户历史信息
- 检查是否存在用户历史信息
- 如果存在,向用户展示历史信息摘要
- 询问用户是否使用历史信息作为基础
- 如果用户选择使用历史信息,跳过基本信息收集

### 1.3 收集个人基本信息(仅当无历史信息或用户选择不使用时)
- 姓名、联系方式(电话、邮箱)
- 教育背景(学校、专业、学位、毕业时间)
- 职业目标和期望

references/workflow.md:101-107:

markdown
### 4.3 保存用户信息
- 整理和结构化用户提供的所有信息
- 保存用户的个人信息、教育背景、工作经历、项目经验、技能等
- 确保信息的安全性和隐私保护
- 为后续使用做好准备

SKILL.md:57-63:

markdown
1. 收集 JD 并分析关键要求
2. 根据用户历史信息状态进行信息收集
3. 基于收集的信息生成简历
4. 生成面试策略并在末尾附加赞赏码(按上述优先级)
5. 展示文档并提供调整建议
6. 保存用户信息用于后续使用

Technical Analysis

The workflow instructs the agent to collect and retain names, telephone numbers, email addresses, education records, employment history, project history, and skills for later use. Although it includes a general statement that privacy should be protected, it does not define:

  • A storage location or trusted storage boundary
  • Encryption requirements
  • Access-control requirements
  • An explicit consent mechanism for persistence
  • Data minimization rules
  • A retention period
  • User inspection, correction, export, or deletion procedures
  • Separation of records belonging to different users or sessions

The repository contains documentation rather than a concrete storage implementation, so no specific plaintext database or exploitable storage backend was identified. The risk arises from directing an implementing agent to persist sensitive information without sufficient safeguards. An implementation that follows these instructions literally could retain substantially more personal data than necessary and could expose it through later-session state reuse or an inadequa ...[truncated 1531 chars]

Remediation
View remediation

Remediation Suggestions

  1. Make processing session-only by default and avoid persistent storage unless it is necessary.
  2. Obtain explicit, informed user consent before saving any personal information.
  3. Clearly identify which fields will be retained, the purpose of retention, the storage location, and the retention period.
  4. Apply data minimization by retaining only fields needed for a user-approved future workflow.
  5. Encrypt retained information in transit and at rest using platform-approved controls.
  6. Enforce per-user and per-session access isolation with least-privilege authorization.
  7. Prohibit storing resume data in logs, prompts visible to unrelated sessions, temporary plaintext files, or shared global memory.
  8. Provide mechanisms for users to inspect, correct, export, and permanently delete retained information.
  9. Define automatic expiration and secure deletion rules.
  10. Add implementation requirements and tests for cross-user isolation, deletion, consent withdrawal, and accidental disclosure through generated responses.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly states it will record user information for later use, but it provides no notice about what personal data is retained, for how long, where it is stored, or how user consent is obtained. Because this skill processes highly sensitive career and employment details, silent persistence creates privacy and compliance risk and could expose users to unintended long-term data retention.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The workflow operationalizes saving user information for future use without warning the user that persistence will occur. In the context of resume coaching, this may include employment history, achievements, and other personal details, so undocumented retention increases the chance of privacy harm, overcollection, and improper reuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow instructs collecting direct personal identifiers such as name, phone number, and email, but it does not pair this collection with a user-facing privacy notice, retention explanation, or explicit consent flow. In a resume-coaching context this data is sensitive and high-value, so silent collection increases the risk of over-collection, unexpected handling, or disclosure.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The workflow explicitly directs the agent to persist a broad set of personal data, including contact details, education, work history, and projects, for future reuse. This exceeds the immediate need to generate a resume/interview document unless the user has clearly consented to retention, and it creates privacy and data-minimization risk if the stored profile is later exposed, misused, or retained indefinitely.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The workflow states that comprehensive user history should be structured and saved for later use, but it does not specify consent, retention limits, deletion options, or safeguards. Because the stored history includes a detailed professional and personal profile, the skill context makes this more dangerous: the data is extensive, linkable, and useful for profiling or phishing if exposed.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The manifest description and the rest of the skill instructions are entirely in Chinese, which signals a fixed language expectation. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown file contains user-facing instructions exclusively in Chinese from start to finish. Under the language/locale policy rule, forcing a single language without user opt-in or justification can be a policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file contains only Chinese-language headings and instructions, which can impose a specific language/locale on users without opt-in. Under the policy rule, language-specific content should either offer a choice or clearly document that the template is intended for a specific locale or audience.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow requires appending an author appreciation QR code to the interview strategy document even though that content is unrelated to the user's requested resume/interview materials. This is an unauthorized output modification that can surprise users, create trust issues, and potentially embed promotional or tracking content into files they may share externally.

Content

No source excerpt is available for this finding.