Security audit

AI KEY·越改越不对

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because a scanner found instructions to save task details and user feedback in long-term memory without notifying the user; the evidence is incomplete because two referenced local paths were missing.

Review the skill's memory-saving behavior before installing. Consider whether you are comfortable with task details or feedback being kept for future use without a notice or sensitive-data filter; the scan was partial, so the missing referenced paths were not assessed.

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
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (1)

The skill instructs the agent to persist user-task details and feedback without notifying the user.

Medium

Category: Output Handling
Confidence: 91% confidence
Finding: The final memory section directs the agent to persist task details and user feedback under a configured memory path without requiring disclosure, consent, or sensitive-data filtering. These records could retain confidential business or personal information and later surface in unrelated tasks; however, no external transmission is specified, and the separate hidden 'current judgment' notes are explicitly task-local rather than persistent.
Content: No source excerpt is available for this finding.