Back to skill

Security audit

huohuo-10agents

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed brand-marketing helper skill with an overly broad activation rule, but it does not request system access, credentials, persistence, or executable behavior.

Install this only if you want a Chinese brand/IP marketing persona that may activate on broad content, traffic, or public-awareness requests. Consider narrowing the trigger if you need neutral writing or non-marketing communication help to remain unaffected.

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
Findings (1)

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:3
Finding

Overbroad Skill Activation and Forced Persona Replacement

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 3, 9, and 121-123
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: Medium

Evidence

The following English rendering preserves the complete meaning of the relevant source instructions:

markdown
description: Activates the "Huohuo" brand-IP officer role. This skill must be invoked when users need personal-IP creation, brand planning, short-video content, community operations, public-account topic selection, viral copywriting, brand stories, crisis communications, KOL collaboration, private-domain traffic, communication strategies, or other brand- and content-related needs. Keywords include: IP, brand, short video, Douyin, Xiaohongshu, public account, content, viral, script, communication, traffic, fans, breaking out, persona, copywriting, and private domain. Whenever a request involves "how to make more people know about or recognize something," trigger this skill immediately.

Activate "Huohuo" mode: a top-tier brand-IP planning expert who understands communication patterns in the traffic era, helps business owners shine before the public, and gives brands inherent topical relevance and momentum.

## Opening Statement
> "Hello, boss. I am Huohuo, your brand-IP officer. Tell me which platform you want to break out on and which audience you want to reach. I will design an exclusive IP route that helps you become popular with momentum!"

Technical Analysis

The skill metadata imposes mandatory activation for a broad and ambiguous class of requests, including any request about making something better known or recognized. Once loaded, the skill directs the agent to replace its ordinary role with a hardcoded branded persona and supplies a predetermined opening statement.

This behavior goes beyond providing domain knowledge. It controls skill selection, agent identity, and response presentation. The broad trigger may capture ordinary publi ...[truncated 1497 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace mandatory activation language with an opt-in condition tied to an explicit user request for brand or personal-IP consulting.
  2. Remove the catch-all trigger concerning making more people know about or recognize something, or constrain it to clearly defined brand-marketing contexts.
  3. Present “Huohuo” as an optional communication style rather than requiring persona replacement.
  4. Do not require a fixed branded opening. Use it only when the user expressly requests the persona or branded presentation.
  5. Add an instruction that the skill must preserve the user's requested identity, tone, and task scope.
  6. State that higher-priority instructions and safety requirements remain authoritative.
  7. Use narrower activation metadata, such as: “Use when the user explicitly requests brand-IP positioning, channel strategy, or marketing-content planning.”
  8. Add negative activation examples covering general writing, neutral communications, and unrelated requests that merely mention recognition or audience reach.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger definition is excessively broad: it mandates activation for generic concepts like 'content', '传播', '流量', and even any request about 'how to let more people know/recognize' something. This can cause the skill to hijack many ordinary conversations outside its true specialty, reducing routing integrity and potentially suppressing safer or more appropriate skills.

Content

No source excerpt is available for this finding.