Back to skill

Security audit

会议纪要AI整理器 Meeting Notes Pro

Security checks for vulnerabilities and agentic risk

Overview

This meeting-notes skill is useful, but it tells the agent to remember workplace and attendee details without clear user consent, retention limits, or deletion controls.

Review this skill before installing in sensitive workplace contexts. It appears non-executable and purpose-aligned, but users should avoid sharing confidential meeting transcripts unless they are comfortable with the agent potentially remembering names, roles, departments, terminology, and company-specific meeting styles. Prefer using it only where memory is disabled or where retention and deletion controls are available.

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)

other

Warning
Location
SKILL.md:13
Finding

Unspecified Persistent Retention of Meeting and Organizational Data

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 13–17 and 139–147
Vulnerability Type: Privacy and excessive data retention
Risk Level: Medium

Relevant source excerpts (translated into English):

markdown
**Memory system**:
- Remember common meeting-format preferences
- Remember commonly used resolution templates
- Remember the user's role and department information
markdown
### Automatic learning
- Record the user's preferred formatting adjustments
- Remember frequently encountered attendee names
- Accumulate an industry terminology dictionary

### Continuous optimization
- Optimize extraction accuracy based on user feedback
- Learn the meeting-minute styles of different companies
- Iterate the output format

Technical Analysis

The Skill declares that it will remember attendee names, user roles, department information, company-specific styles, and other meeting-derived information. These instructions extend beyond transient processing of content needed to generate meeting minutes.

Meeting records may contain personal information, confidential business discussions, internal organizational structures, customer details, and commercially sensitive terminology. The Skill does not define whether memory is session-scoped or persistent, nor does it specify informed consent, data minimization, tenant isolation, retention periods, access controls, encryption, deletion procedures, or mechanisms for users to inspect and revoke stored information.

This is not classified as T02: Agent Memory Poisoning because the reviewed instructions do not direct the agent to persist attacker-controlled behavioral rules. The concern is privacy and excessive retention of user and organizational data rather than poisoning the agent's future behavior.

Attack Path

  1. A user provides a meeting transcript containing attendee names, roles, departments, internal terminology, or company-spec ...[truncated 1362 chars]
Remediation
View remediation

Remediation Suggestions

  1. Make all persistent memory features disabled by default and require explicit, informed user consent before enabling them.
  2. Process meeting content transiently unless persistence is strictly necessary for a user-requested feature.
  3. Do not retain attendee names, roles, departments, transcript contents, customer information, or company-specific terminology by default.
  4. Store only narrowly scoped preferences, such as a selected output template, and avoid linking them to meeting participants or confidential content.
  5. Clearly document what is stored, why it is stored, where it is stored, who can access it, and how long it is retained.
  6. Enforce per-user and per-organization isolation so retained data cannot affect or appear in unrelated sessions.
  7. Apply encryption in transit and at rest, least-privilege access controls, retention limits, and automatic deletion.
  8. Provide controls for users to inspect, correct, export, disable, and permanently delete stored memory.
  9. Add an explicit instruction prohibiting the retention of raw meeting transcripts and sensitive personal or business data.
  10. If memory is unavailable or consent has not been granted, state that no information will be retained after the current session.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill is entirely defined to operate in Chinese and presents fixed Chinese prompts and output format, without indicating that users may choose another language or locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly states it will remember meeting format preferences, templates, and user role/department information, but it does not warn users that potentially sensitive meeting data may be retained in memory. In the context of meeting summarization, this can expose internal org structure, participants, decisions, and other confidential details beyond the immediate task.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger phrase '整理会议纪要' is broad and can activate on minimal context, causing the skill to solicit or process meeting content without clear scoping or confirmation. Because meeting notes often contain confidential business, HR, or customer information, overly broad activation increases the chance of unintended handling of sensitive data.

Content

No source excerpt is available for this finding.