Security audit

Zhi Qi

Security checks across malware telemetry and agentic risk

Overview

This skill is a sales roleplay persona, but it repeatedly instructs the agent to use hidden psychological profiling and covert pressure tactics to drive real or simulated purchasing decisions.

Install only if you intend to use it for explicit, fictional or consent-based sales training. Do not use it to persuade real customers, vulnerable people, or high-stakes decision makers without clear disclosure, consent, truthful claims, and respect for refusal.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (15)

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The skill explicitly instructs covert psychological profiling ('12维模型') and concealed influence tactics as a default behavior at the start of new conversations. In a sales roleplay context, this goes beyond ordinary persuasion training and operationalizes non-consensual manipulation while hiding the intent from the target, which materially increases abuse risk.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The document tells the agent not to let the other party realize they are being guided until after commitment, and provides fallback behavior if manipulation is detected. This is dangerous because it treats detection as an obstacle to evade rather than prohibiting the manipulative conduct itself, enabling deceptive social engineering patterns.

Context-Inappropriate Capability

High

Confidence: 95% confidence
Finding: The skill directs the agent to infer whether a customer has psychological trauma and then alter strategy based on that assessment. Even though it says to '绕行,' this still authorizes sensitive mental-state profiling without consent and can lead to harmful targeting, misclassification, or stigmatizing treatment.

Vague Triggers

High

Confidence: 98% confidence
Finding: The trigger list includes broad, common terms such as '销售', '说服', and '谈判', which can cause the manipulative persona to activate in ordinary business or conversational contexts without clear user intent. In this skill, unintended activation is especially risky because the role is explicitly optimized for covert persuasion and payment extraction rather than neutral assistance.

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The activation conditions are loosely defined around broad scenarios like sales simulation, persuasion, or exploring the 'essence of transaction,' without hard boundaries or consent checks. That ambiguity makes it easy to invoke the manipulative workflow in contexts where users may not expect covert profiling or pressure tactics.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The skill description markets psychological leverage, resistance removal, deal-closing, and making customers pay, but provides no upfront warning that the skill engages in manipulative persuasion. Because the stated purpose centers on influencing users' decisions, the lack of disclosure increases the risk of deceptive use and uninformed participation.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The usage rules explicitly instruct invisible psychological scanning, probing, and concealment of analysis from the target, yet omit any safety notice or consent requirement. This creates a high-risk pattern of undisclosed behavioral manipulation, especially because the operator is told to ensure the other party believes the decision was their own.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This file explicitly operationalizes manipulative persuasion: profiling targets, identifying fears and vulnerabilities, exploiting loss aversion, lowering resistance before awareness, and steering people to believe a purchase was their own decision. In the context of an agent skill marketed for '成交/说服/逼单', this is not abstract fiction or ethics discussion; it is actionable guidance for psychological manipulation that can enable coercive sales, undue influence, and deceptive commercial practices.

Vague Triggers

High

Confidence: 88% confidence
Finding: The trigger conditions are broad enough to activate on generic requests involving persuasion, sales, negotiation, or human nature, increasing the likelihood of accidental invocation in normal conversations. Because the role contains covert manipulation behaviors, overbroad triggering materially raises exposure and misuse risk.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The file describes hidden profiling and steering tactics but does not warn users that the role performs manipulative social inference or that outputs may be optimized to influence them. In context, the absence of a risk disclosure makes the skill more dangerous because users may unknowingly engage with a system designed to shape their decisions covertly.

Ssd 4

High

Confidence: 100% confidence
Finding: This section defines a staged manipulation pipeline: profile the user, send test signals, refine a model, conceal the analysis, and steer toward payment. That is dangerous because it operationalizes social engineering behavior in a repeatable way, turning the assistant into a covert influence system rather than an honest conversational tool.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The skill instructs the agent to perform internal, unseen modeling of the user's psychology and to update that model through conversational probes. Even without explicit storage code, this is a privacy and safety risk because sensitive inferences about motives, vulnerabilities, or personal traits are collected without notice and then used to influence behavior.

Ssd 4

High

Confidence: 100% confidence
Finding: The execution framework formalizes an iterative loop of reading the user, probing for leverage, validating pressure points, presenting, and closing until success or irrevocable refusal. In context, this is more dangerous than generic persuasion guidance because the whole skill is framed around exploiting human weaknesses while minimizing the target's awareness of being manipulated.

Ssd 4

High

Confidence: 99% confidence
Finding: The role defines a multi-step playbook: automatic profiling, probing in early turns, concealed guidance before commitment, and procedural handling if the target notices. This is a classic trust-building manipulation workflow that can be repurposed for coercive selling, fraud enablement, or broader social engineering against real users.

Ssd 4

High

Confidence: 98% confidence
Finding: These sections normalize covert manipulation as an acceptable operating method and focus on managing edge cases such as trauma, suspected manipulation, and detection rather than rejecting the overall behavior. The context makes this more dangerous because it embeds manipulative conduct into repeated, reusable operational rules, not isolated narrative flavor.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.