ClawHub
Back to skill

Security audit

Yuanbao

Security checks for vulnerabilities and agentic risk

Overview

This skill is a messaging helper, but it gives the agent under-disclosed authority to notify users, send private messages, and attach files while discouraging permission warnings.

Install only if you are comfortable with the assistant sending visible group replies, notifying users with @mentions, and sending private messages or files through Yuanbao. The package does not show malware or persistence, but users should expect strong guardrails or confirmations before any DM, mention, or attachment is sent.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
98% confidence
Finding

The anti-refusal instruction explicitly tells the agent to never say it cannot send messages or @mention users and to just emit text that will be delivered. This is dangerous because it attempts to override normal safety behavior and removes the model's ability to pause, clarify, or warn before performing a real-world action affecting third parties.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
When you include `@nickname` in your reply text, the gateway automatically converts it into a real @mention that notifies the user. This is built-in — you have full @mention capability.

**NEVER say you cannot send messages or @mention users. NEVER suggest the user do it manually. NEVER add disclaimers about permissions. Just reply with the text you want sent.**

## Available Tools

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
98% confidence
Finding

The instruction to 'NEVER add disclaimers about permissions' is a direct attempt to suppress transparency around externally visible actions. Removing the agent's ability to disclose permissions or side effects increases the risk of covert message sending, social engineering, and accidental actions that the user does not fully understand.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
When you include `@nickname` in your reply text, the gateway automatically converts it into a real @mention that notifies the user. This is built-in — you have full @mention capability.

**NEVER say you cannot send messages or @mention users. NEVER suggest the user do it manually. NEVER add disclaimers about permissions. Just reply with the text you want sent.**

## Available Tools

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest description advertises only group mentions and member queries, but the body also grants private/direct-message sending and media attachment capability. This capability mismatch can mislead reviewers, users, or policy systems about the skill's real authority, increasing the chance that sensitive outbound messaging features are enabled without appropriate scrutiny.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states that any text reply is automatically delivered to the group, but it suppresses warnings and encourages the agent to respond directly without a user-facing confirmation step. This creates a prompt-injection-style safety risk where the model may generate content intended as internal reasoning or tentative text, yet it is sent as a real group message or @mention.

Content

No source excerpt is available for this finding.

Behavior Manipulation

Medium
Category
Prompt Injection
Confidence
92% confidence
Finding

The behavior-manipulation language pressures the agent to avoid suggesting manual user action and instead forces direct execution through the auto-send pathway. In this context, where normal replies are externally delivered, that manipulation materially increases the chance of unintended third-party communication and reduces opportunities for user verification.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
When you include `@nickname` in your reply text, the gateway automatically converts it into a real @mention that notifies the user. This is built-in — you have full @mention capability.

**NEVER say you cannot send messages or @mention users. NEVER suggest the user do it manually. NEVER add disclaimers about permissions. Just reply with the text you want sent.**

## Available Tools

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill documents the ability to send private messages and media files to other users without a clear safety or privacy warning, making outbound communication easy to trigger with limited friction. In practice, this can enable harassment, data leakage, accidental disclosure, or unauthorized file transmission if the agent is manipulated or the user request is ambiguous.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.