Security audit

AI KEY·今天拍什么

Security checks for vulnerabilities and agentic risk

Overview

This topic-planning skill needs review because ordinary feedback can automatically change shared rules that guide future writing.

Before use, require approval of proposed shared-rule changes and keep recoverable copies. Restrict access to the intended content vault. Review the missing scripts and shared instructions before enabling their execution; this package alone cannot establish their safety. Supply only recordings you may reuse, and remove identifying or confidential remarks before turning them into public content.

A.I.G

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

SkillSpector

Vulnerability Patterns

Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
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 (3)

The topic-selection skill automatically rewrites a shared writing framework based on any user feedback.

Medium

Category: Excessive Agency
Confidence: 90% confidence
Finding: The skill directs the agent to automatically incorporate any user feedback into a shared framework that governs future work. This can promote a temporary preference, misunderstood correction, or adversarial feedback into persistent instructions affecting other writing skills; no validation or approval boundary is specified in the supplied source.
Content: No source excerpt is available for this finding.

Automatically persists user feedback into the governing framework without requiring user notification.

Medium

Category: Output Handling
Confidence: 90% confidence
Finding: This is the same automatic-write mechanism viewed as a persistence and user-control issue: ordinary feedback can trigger lasting changes to the shared framework without a separate request. The text does not explicitly prohibit notification, but it also does not require confirmation or disclosure, making unintended cross-session behavior changes possible.
Content: No source excerpt is available for this finding.

The workflow uses offline and one-to-one recordings as source material without warning about participant privacy or consent.

Medium

Category: Output Handling
Confidence: 90% confidence
Finding: The workflow explicitly promotes offline classes and one-to-one recordings as first-hand material for a table containing participants' original statements and problems, without checking permission for reuse or requiring anonymization. In a public-content workflow, this creates a path for confidential or identifiable remarks to enter candidate files and later scripts, although actual disclosure is not demonstrated.
Content: No source excerpt is available for this finding.