Security audit

Auto Agent 4974

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a Bilibili learning/knowledge-entry helper, but it relies on an unreviewed external Python module path and can persist caller-supplied text into a knowledge base.

Review before installing. Use it only in an environment where D:\\coze-local\\db and learn.py are trusted and protected from modification, and expect the skill to write entries into a local knowledge base when invoked. Avoid broad automatic activation on the word "agent" unless you are comfortable with accidental knowledge writes.

A.I.G

Vulnerability Patterns

Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
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

Findings (2)

T07 · Tool Hijacking and Spoofing

Error

Location: auto_agent_4974.py:6
Finding: Untrusted Import Path Enables Local Module Hijacking
Content: View full analysis
Remediation: View remediation

T02 · Agent Memory Poisoning

Warning

Location: auto_agent_4974.py:12
Finding: Unvalidated User Input Is Written to Persistent Agent Knowledge
Content: View full analysis
Remediation: View remediation

SkillSpector

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)

Vague Triggers

High

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The trigger phrase 'agent' is extremely broad and is likely to activate during normal conversation unrelated to this skill. That can cause unintended invocation, context hijacking, or accidental execution of skill behavior when the user did not explicitly request it, which is a real security and safety issue for agent systems.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 86% confidence
Finding: The skill's visible title, functionality description, and trigger text are entirely in Chinese, with no indication that users may interact in other languages or opt into this locale. Under the policy, language-specific behavior should either offer user choice or clearly justify the locale restriction.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The module docstring and run() docstring describe educational/tutorial content about building an AI agent, which implies explanatory or instructional behavior. In reality, the implementation imports a KnowledgeBase and persists supplied input and metadata via kb.store_concept(), which is a materially different operational intent.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 83% confidence
Finding: The file’s user-facing title and runtime output are entirely in Chinese, and there is no indication that another language can be selected or that the locale restriction is intentional and documented. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.