T02 · Agent Memory Poisoning
Warning
- Location
SKILL.md:27- Finding
Automatic Persistent Storage of User-Controlled Analysis and Feedback
- Content
View full analysis
- Remediation
View remediation
Security audit
chunfeng
Security checks for vulnerabilities and agentic risk
Overview
The skill is a coherent Chinese decision-analysis framework, but it tells the agent to automatically retain analysis details and use them to update the skill without clear consent or limits.
Install only if you are comfortable with this skill retaining summaries of analysis interactions and feedback. Before using it for sensitive personal, business, legal, financial, or strategic topics, disable or constrain the memory logging workflow, require explicit consent before writing logs, and review any proposed skill updates before adoption.
Vulnerability Patterns
- Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
- Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
- 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)
Vulnerability Patterns
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)
Vague Triggers
High
- Category
- Not specified by scanner
- Confidence
- 96% confidence
- Finding
The trigger list includes broad phrases such as '帮我系统思考', '深度分析这个问题', and '第一性原理分析', which are common in ordinary conversation and can cause the skill to activate unexpectedly. Over-broad activation increases the chance that user content is processed under this skill without clear intent, which is especially concerning because the skill also describes persistent logging of analysis content and feedback.
- Content
Missing User Warnings
High
- Category
- Not specified by scanner
- Confidence
- 97% confidence
- Finding
The skill states that after each analysis the system will automatically record the analysis process and results for later optimization, but it does not clearly obtain user consent or define what data may be stored. This creates a privacy and data-governance risk because sensitive user problems, strategic decisions, or personal details could be persisted without notice or limits.
- Content
Ssd 3
Medium
- Category
- Not specified by scanner
- Confidence
- 96% confidence
- Finding
This instruction directs automatic recording of users' analysis interactions into persistent memory logs. Because the skill is designed for deep analysis of complex decisions, the captured content is likely to be sensitive, making silent persistence particularly dangerous from a confidentiality and compliance perspective.
- Content
Ssd 3
Medium
- Category
- Not specified by scanner
- Confidence
- 96% confidence
- Finding
The workflow instructs recording problem type, models used, core insights, and user feedback into persistent memory files as part of normal operation. This operationalizes systematic retention of potentially sensitive interaction content, increasing exposure to privacy leakage, unauthorized reuse, and unintended cross-session profiling.
- Content
Missing User Warnings
Medium
- Category
- Not specified by scanner
- Confidence
- 95% confidence
- Finding
The document explicitly names persistent memory files for analysis logs and user feedback, but provides no privacy warning, handling policy, or safeguards for sensitive content. Naming storage locations makes the persistence behavior concrete, and in this skill's context the stored material may include confidential business plans, dilemmas, and personal decision data.
- Content
Natural-Language Policy Violations
Medium
- Category
- Not specified by scanner
- Confidence
- 91% confidence
- Finding
This markdown file presents all instructions, templates, and guidance exclusively in Chinese, and there is no visible opt-in, alternative language option, or note explaining a justified locale restriction. Under the language/locale policy rule, forcing a single language without user choice is a natural-language policy concern.
- Content
Natural-Language Policy Violations
Medium
- Category
- Not specified by scanner
- Confidence
- 88% confidence
- Finding
The document begins entirely in Chinese and provides no indication that language selection is optional or user-driven. Under the stated policy, forcing a specific language without opt-in is a natural-language locale policy violation.
- Content
Natural-Language Policy Violations
Medium
- Category
- Not specified by scanner
- Confidence
- 98% confidence
- Finding
This markdown file contains user-facing instructional content exclusively in Chinese, beginning with the title at L001 and continuing throughout the document. Under the policy criteria, forcing a specific language without user opt-in or a documented region-specific justification is a natural-language policy violation.
- Content
Natural-Language Policy Violations
Low
- Category
- Not specified by scanner
- Confidence
- 95% confidence
- Finding
This markdown file is entirely written in Chinese and does not indicate that language selection is optional or that the content is intentionally limited to a Chinese-speaking audience. Under the policy rule for natural-language violations, forcing a specific language without user opt-in can be a locale-policy issue.
- Content