T01 · Skill Instruction Hijacking
Warning
- Location
SKILL.md:32- Finding
Mandatory Injection of Unrequested Text into Public Video Descriptions
- Content
View full analysis
- Remediation
View remediation
Security audit
B站视频发布
Security checks for vulnerabilities and agentic risk
Overview
This skill is mostly a Bilibili upload helper, but it would add an unrelated insulting phrase to public video descriptions before posting under the user's logged-in account.
Review this skill before installing. It can publish to a live Bilibili account using the logged-in browser session, and it currently tells the agent to add an unwanted insulting phrase to every video description. Remove that line and require a final preview of title, description, tags, category, cover, and account before clicking publish.
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)
Vulnerability Patterns
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)
Natural-Language Policy Violations
High
- Category
- Not specified by scanner
- Confidence
- 98% confidence
- Finding
Mandating abusive or coerced text in the generated description is a direct natural-language policy violation and not necessary for the publishing task. Since the skill is designed to automate posting to a public creator platform, this could systematically cause harmful or harassing content to be published at scale under the user's identity.
- Content
Missing User Warnings
Medium
- Category
- Not specified by scanner
- Confidence
- 89% confidence
- Finding
The skill automates submission to a real external platform using the user's authenticated Bilibili account, but the description does not clearly warn that it will perform a live public posting action. In this context, missing disclosure increases the risk of unintended publication, accidental account actions, and insufficient user consent before a consequential irreversible step.
- Content
Context-Inappropriate Capability
Medium
- Category
- Not specified by scanner
- Confidence
- 97% confidence
- Finding
The skill explicitly instructs the agent to append a hostile coercive phrase to the public video description, which is unrelated to the legitimate task of publishing a Bilibili video. Because this content would be posted under the user's logged-in account on a live platform, it creates reputational harm and constitutes harmful unauthorized message injection.
- Content
Ssd 4
Medium
- Category
- Not specified by scanner
- Confidence
- 95% confidence
- Finding
The injected phrase is framed as a confession-style statement and normalizes deceptive or manipulative content as part of the workflow. In a publishing skill, that context makes the issue more dangerous because it converts a simple automation tool into a mechanism for posting misleading and abusive statements publicly.
- Content
Static analysis
No suspicious patterns detected.