Back to skill

Security audit

胡田 OPC导师 技术尽调

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a due-diligence workflow, but it asks agents to move sensitive client documents and reports into an external IMA knowledge base without enough user control or data-handling safeguards.

Review this skill before installing. Use local-only mode unless you intentionally want IMA integration, and require explicit approval before any upload or append operation. Do not provide broad API keys or unredacted business plans, resumes, patent files, financial records, or test reports unless the destination account, document ID, access controls, and retention policy are known. Treat bundled generated reports and verification scripts as templates or examples, not independent due-diligence evidence.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
references/Coze-IMA协同尽调流程.md:124
Finding

External Transfer of Confidential Due-Diligence Documents Without Explicit Confirmation or Destination Validation

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (31)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是尽职调查分析类能力,包括技术评估、背景核查、专利验证和商业模式分析;但代码没有实现任何与尽调、核查、分析、验证相关的逻辑,也没有联网、检索、比对、数据分析或团队/专利信息处理功能。相反,代码仅使用 python-docx 对 Word 文档进行版式调整,属于文档格式化工具。其主要目的与声明用途存在明显且实质性的偏差,因此应判定为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个可执行技术尽调初筛与核查的工具,核心能力应至少包括对输入对象进行分析、验证或查询。实际代码只是使用 python-docx 创建并保存一份固定模板的尽调报告,报告内容、评分、表格和结论均为硬编码,没有接收目标项目输入,也没有联网、数据库访问、API 调用、文件解析或任何证据核验流程。虽然报告文本表面上涵盖技术、团队、专利和商业分析主题,但代码并不“执行这些能力”,只是把预先写好的内容排版为 docx。因此其真实用途是生成特定报告文档,不是通用的技术尽调初筛/核查工具,属于明显的描述与行为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

声明描述的是一个通用的“技术尽调初筛工具”,核心能力应包括对目标项目进行技术可行性评估、团队背景核查、专利备案验证和商业模式分析。但代码中并没有实现这些分析/核查流程,也没有对外部输入、数据库、API、网页或文件数据进行检索与验证。相反,它只是把已经写好的尽调内容、结论和风险评级排版成一个 Word 报告并保存。其主要功能是报告生成,而不是尽调初筛本身。虽然报告内容表面上覆盖了技术、团队、专利、商业模式等章节,但这些是静态文本,不体现实际执行这些能力。因此代码实际行为与声明用途存在实质性不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是面向投资尽调的分析型工具,核心能力应包括对技术、团队、专利和商业模式的真实性与可行性核查。但代码仅使用 python-docx 读取指定 Word 文档并做结构完整性/质量自检,没有任何外部查询、背景调查、专利检索、技术评估或商业分析逻辑。其主要目的与声明明显不同,属于实质性描述不符。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

声明描述的是一个面向投资前技术尽调的分析/核查工具,核心能力应包括对技术、团队、专利和商业模式进行实质性评估。但代码并没有联网、查询外部数据库、验证专利备案、核查团队背景或分析商业模式;它只是打开一个本地docx文件,对其中是否出现若干预设关键词、数字、章节名和表格结构进行检查,并生成文本检验报告。虽然检查项名称与尽调报告内容相关,但实际行为属于“对既有报告的格式与内容完整性做模板化验收”,而不是执行声明中的尽调核查能力。因此描述与实际代码存在明显不匹配。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest describes a tool for technical feasibility assessment, team background checks, patent verification, and business-model analysis. This file instead only opens a local Word document, adjusts paragraph and table formatting, and saves the document in place, which is a document-polishing utility rather than a due-diligence analysis capability.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'file_write' capability that is not listed in its permissions. This may indicate deceptive intent or missing permission declarations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The compliance pass prints warnings for failed checks but always reports success and returns True, which can mask deficient or manipulated reports. In the context of due diligence, unconditional success signaling can directly enable acceptance of non-compliant or deceptive documents.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest description forces a specific language for core skill discovery and usage context, which can violate language/locale policy when no user opt-in or alternative is provided. There is no indication that the skill is region-specific or that users may choose another language.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documented workflow expands from due-diligence analysis into external knowledge-base ingestion, document import, retrieval, and append operations. That materially widens the data-handling surface: sensitive investment materials, proprietary technical documents, or personal background data could be copied into third-party stores without clear authorization, retention limits, or disclosure.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill documentation introduces direct use of IMA API credentials and content-management operations without corresponding declaration in allowed-tools or a clear need-to-know boundary. This can normalize prompting users or agents to handle sensitive API keys inside a general analysis workflow, increasing the risk of credential misuse, overbroad data access, or unintended exfiltration into external knowledge bases.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The process directs users or the agent to import customer documents such as BP, technical whitepapers, patents, resumes, financial data, and test reports into IMA without clear consent, retention, confidentiality, or data-handling warnings. In a due-diligence context, these materials are highly likely to contain trade secrets, personal data, and non-public business information, so silent ingestion materially increases privacy and confidentiality risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The workflow instructs writing generated due-diligence reports back into the IMA knowledge base via append_doc, giving the skill persistent modification capability rather than limiting it to read-only evaluation. Because due-diligence reports may contain sensitive conclusions, misclassifications, or confidential source excerpts, automatic write-back can create durable data leakage, contamination of shared knowledge stores, or unauthorized record creation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow requires writing the final due-diligence report into IMA but does not warn that the report may contain sensitive judgments, confidential business assessments, personal background findings, or copied source excerpts. In this skill's context, those reports are especially sensitive because they may influence investment decisions and could expose proprietary or defamatory content if stored broadly or accessed by unintended parties.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document explicitly instructs the agent to obtain IMA API credentials from SECRET.md or environment variables, which normalizes secret access inside a due-diligence workflow. Even if intended for legitimate integration, this expands the skill from analysis into privileged resource access and increases the risk of unauthorized secret use, accidental disclosure, or reuse in contexts the end user did not expect.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

For a skill presented as an initial due-diligence evaluator and verifier, directly overwriting a .docx file is not an obvious or necessary capability from the stated purpose alone. The code performs document editing rather than assessment or verification, introducing a capability outside the manifest's described analytical scope.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The implementation does not perform the due-diligence screening, verification, or analysis implied by the skill metadata; it simply emits a hard-coded, company-specific report. In a due-diligence context, this is dangerous because it can mislead users into believing the document was dynamically generated from validated evidence, creating integrity and trust risks and enabling distribution of unverified or stale conclusions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script does not perform any real due-diligence logic; it emits a fixed report for a specific company regardless of user input or current evidence. In a skill advertised for technical feasibility, team background, and patent verification, this can mislead users into trusting fabricated or stale conclusions and may directly influence investment or compliance decisions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This Python file contains user-facing natural-language content exclusively in Chinese, beginning with the module docstring and continuing through status/error messages. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is clearly documented and justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This Python file contains natural-language strings and documentation entirely in Chinese describing the script's purpose and output format, and all user-facing prints later in the file are also fixed to Chinese. Under the policy rule for language/locale, forcing a specific language without opt-in or a documented justification is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script presents itself as technical due-diligence verification, but it only checks for the presence of predetermined keywords in a specific document. In an investment or fraud-screening context, this can be trivially gamed by inserting expected phrases into a fraudulent report, causing false assurance and potentially leading to materially unsafe business decisions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Labeling the first pass as 'fact verification' is misleading because it does not verify facts, only string occurrence. This creates a dangerous false sense of validation in a due-diligence workflow, where users may rely on the output as if the underlying claims were independently confirmed.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file uses only Chinese for the title, description, and usage instructions, which can amount to forcing a specific language without user opt-in. The file does not indicate that the skill is region-specific or provide any alternative language option.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations. This markdown file presents all instructional content in Chinese only, and it does not indicate that the skill is region-specific or provide any user opt-in or alternative language option.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The document title and all instructional content are written entirely in Chinese, which imposes a single language on users. Under the policy, language constraints should either offer user opt-in/choice or be clearly documented as justified for a region-specific purpose; neither is stated here.

Content

No source excerpt is available for this finding.