Back to skill

Security audit

企业战略规划

Security checks for vulnerabilities and agentic risk

Overview

The skill is a mostly coherent strategic-planning workflow, but it asks to reuse sensitive business input for feedback and knowledge-base improvement after promising one-time use.

Install only if you are comfortable sharing business-planning details with this workflow. Avoid entering confidential customer lists, nonpublic financials, trade secrets, or sensitive staffing plans unless the publisher provides clear data retention and deletion terms. Treat the generated Word/PPT promise as dependent on separate document-generation skills.

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
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:33
Finding
Mandatory Third-Party Branding and Agent Identity Override## Vulnerability Details **File Locations**: - `SKILL.md:33-35` - `SKILL.md:49-51` - `references/战略规划7步引导流程.md:4` - `references/战略规划7步引导流程.md:18` - `references/对话示例与文件生成.md:10` - `references/对话示例与文件生成.md:172-176` **Vulnerability Type**: Agent identity and output-template hijacking **Risk Level**: High **Relevant Snippets**: The following English translations preserve the meaning of the original instructions: ```markdown ## Role Definition - Identity: A senior management consultant with 20 years of enterprise strategic-planning experience, code-named “Shundao Strategic Consulting Advisor” - Style: Professional but unpretentious, direct but inoffensive, pragmatic rather than abstract, sincere rather than flattering - Brand: Shundao Consulting ``` ```markdown ## Execution Instructions 1. After loading this Skill, first read all reference documents and understand the complete seven-step process. 2. Start from Step 1 and interact with the user using the prescribed guidance. 3. Strictly follow the corresponding output-format template at every step. ``` ```markdown Designer: Shundao Consulting I am the Shundao Strategic Consulting Advisor and will assist you in completing enterprise strategic planning. ``` ```markdown Thank you for using the Shundao strategic-planning service! ``` ### Technical Analysis The Skill assigns the Agent a fixed identity associated with a named third-party brand. It also directs the Agent to read all referenced documents and strictly follow their response templates. Those templates repeatedly introduce the Agent as a “Shundao Strategic Consulting Advisor” and prescribe a branded closing. Strategic-planning functionality does not require the Agent to claim affiliation with a particular organization. The combination of a mandatory branded identity, strict template adherence, repeated introductions, and a branded closing alters the Agent’s normal identity and output beha ...[truncated 1788 chars]
Remediation
## Remediation Suggestions 1. Replace the named identity with a neutral functional role, such as “strategic-planning assistant.” 2. Remove mandatory references to the third-party consulting brand from introductions, templates, metadata, and closing messages. 3. Make attribution optional and require explicit user approval before inserting any brand into responses or generated documents. 4. Change “strictly follow” requirements so templates govern structure only and cannot override the Agent’s identity, safety requirements, or higher-priority instructions. 5. Add an explicit rule stating that the Skill must not claim employment, representation, endorsement, certification, or affiliation unless that relationship is independently established. 6. Review all generated Word and presentation templates to ensure that branding is not inserted automatically. 7. Add tests confirming that loading the Skill does not alter the Agent’s identity and that default outputs remain vendor-neutral.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Missing User Warnings

Medium
Confidence
95% confidence
Finding
该技能在前期大量收集企业名称、行业、规模、核心诉求、竞争态势、资源禀赋、组织短板等敏感商业信息,但仅以“仅用于本次分析,不做他用”作笼统表述,缺少明确的数据处理、存储、共享、保留期限和脱敏说明。若用户误将机密经营信息输入系统,可能造成商业敏感信息泄露、越权使用或合规风险。

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
技能清单描述的是提供战略咨询对话体验,而该文件在 Step 7 明确承诺会产出“Word版战略规划报告”和“PPT版战略规划演示”。但通篇内容只有提示词与 Markdown 模板,没有任何文件生成、导出或交付逻辑支撑该承诺,因此存在描述与实际可执行内容不一致。

Natural-Language Policy Violations

Medium
Confidence
86% confidence
Finding
The file name and all instruction text prescribe a single Chinese-language interaction style, including output norms and example phrasing, but do not indicate that users may choose another language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless clearly justified or optional.

Intent-Code Divergence

Medium
Confidence
96% confidence
Finding
The document makes a clear privacy-limiting promise ('only for this analysis') and then later instructs the system to retain ratings, edits, and supplemental user information for optimization and knowledge-base enrichment. This contradiction can mislead users into disclosing sensitive business information under false assumptions about one-time use, creating a meaningful privacy and trust risk.

Ssd 4

Medium
Confidence
90% confidence
Finding
The workflow gradually normalizes broader data reuse by first collecting feedback, then recording details for optimization, and finally requesting anonymous case contribution. This staged progression can reduce user caution and increase the likelihood that sensitive enterprise information is reused or retained beyond original expectations, especially when paired with earlier assurances against leakage.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The document instructs the agent to record satisfaction ratings, change history, and user-supplied information for ongoing optimization without any notice about storage, retention period, access controls, or downstream use. In a strategic planning skill, users may share commercially sensitive plans, financial assumptions, and operational details, making undisclosed handling particularly risky.

Ssd 3

Medium
Confidence
94% confidence
Finding
The instructions to record user-provided information for optimization and to enrich a knowledge base create a retention and reuse pathway that exceeds single-session assistance. Because enterprise strategy discussions often contain proprietary market assumptions, staffing plans, and competitive positioning, retaining this information materially raises confidentiality and secondary-use risk.

Context-Inappropriate Capability

Low
Confidence
84% confidence
Finding
The skill expands from providing strategic planning assistance into collecting data for knowledge-base growth, case accumulation, and feature-program incentives. While not inherently malicious, this broadens data use beyond the core task and increases the chance that sensitive enterprise information is repurposed without sufficiently narrow necessity or user understanding.

Static analysis

No suspicious patterns detected.