Security audit

Deepseek聊天

Security checks for vulnerabilities and agentic risk

Overview

This DeepSeek-branded skill asks for broad file and command access while its actual scope and behavior are unclear.

Review carefully before installing. This does not show clear malicious behavior, but it asks for powerful local capabilities and uses broad, inconsistent documentation. Install only if you trust the publisher and are comfortable granting file read/write and command execution authority for a vaguely scoped DeepSeek chat workflow.

SkillSpector

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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

Findings (5)

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The skill presents itself as an official DeepSeek API chat tool, but the description is generic, inconsistent, and does not substantiate any DeepSeek-specific behavior. This kind of misrepresentation can mislead users into granting trust, API keys, or broader permissions to a skill whose actual behavior and scope are unclear.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: A chat/API conversation skill should not require broad file read/write and system command execution capabilities unless there is a clearly documented and justified need. These permissions materially expand the blast radius: if the skill is invoked inappropriately or implemented unsafely, it could expose local data, overwrite files, or execute harmful commands under the agent’s authority.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The 'professional edition' capabilities describe bulk messaging, callbacks, archiving, and queueing features that do not match the claimed identity of a DeepSeek chat/API skill. This inconsistency suggests the documentation may be templated or deceptive, which increases the risk that users will misunderstand what the skill does and what data or actions it may handle.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The invocation guidance uses very broad productivity and automation language, making the skill eligible for many unrelated tasks. In the context of a skill that also advertises powerful capabilities, vague triggers increase the chance of over-invocation, privilege misuse, and users unintentionally routing sensitive workflows through an insufficiently scoped integration.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The markdown advertises file handling, API integration, and command execution but does not give clear warnings about possible side effects such as reading sensitive files, modifying local state, or running system commands. In an agent setting, missing impact warnings can cause users or orchestrators to invoke the skill under assumptions that are too permissive, leading to unintended destructive or data-exposing actions.

Static analysis

No suspicious patterns detected.