Security audit

Elicitation - how to talk with humans and ask them questions?

Security checks across malware telemetry and agentic risk

Overview

This is a markdown-only skill, not malware, but it teaches agents to elicit and infer sensitive psychological information without strong consent, privacy, or emotional-safety boundaries.

Install only for clearly consent-based research, coaching, interviewing, or personal-reflection uses. Do not use it to covertly profile people, pressure them into sensitive disclosures, make diagnostic claims, or store or share psychological profiles without explicit permission. Add guardrails before use: explain the purpose, make sensitive topics optional, stop when someone declines or shows discomfort, avoid mental-health conclusions, and minimize retention of personal details.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (9)

Missing User Warnings

High

Confidence: 95% confidence
Finding: The skill explicitly promotes psychological profiling, eliciting formative memories, and detecting belief patterns, but it provides no upfront warning about privacy sensitivity, emotional risk, consent boundaries, or limits on use. In practice this can lead agents to solicit highly sensitive personal data and vulnerable disclosures without informed consent or safeguards, creating privacy, manipulation, and emotional-harm risks.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The file explicitly frames Motivational Interviewing as 'powerful for elicitation' because it can create conditions for 'authentic disclosure,' but it does not include safeguards, consent language, privacy warnings, or boundaries for handling sensitive emotional material. In the context of a skill whose stated purpose is psychological profiling and gradual self-disclosure, this omission increases the risk of manipulative collection of intimate personal data and emotional harm to users.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill explicitly asks for low points involving pain, sadness, fear, or despair and encourages the user to be 'as honest and detailed as possible' without any emotional-safety framing, opt-out, or crisis guidance. In a conversational agent context, this can pressure users into disclosing traumatic material the system is not equipped to safely handle, increasing risks of distress, retraumatization, and inappropriate collection of sensitive mental-health information.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The probing questions request granular autobiographical details about time, place, people present, thoughts, feelings, and identity significance, which can elicit highly sensitive personal data and third-party information. Because the skill is designed for psychological profiling and gradual disclosure, the absence of privacy warnings, data-minimization guidance, or boundaries makes the collection more dangerous than ordinary interviewing content.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This file provides extensive prompts designed to elicit highly sensitive autobiographical, relational, sexual, health, grief, and identity information without any embedded warning, consent guidance, privacy notice, or emotional-safety boundaries. In the context of a skill explicitly intended for psychological profiling and gradual self-disclosure, this increases the risk of manipulative over-collection of personal data, emotional distress, and collection of special-category sensitive information without informed user awareness.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This section gives concrete prompts for eliciting childhood origins, emotional triggers, and schema-level beliefs without any warning that the content is not mental-health care, may surface trauma, or should avoid diagnosing users. In the broader skill context—explicitly focused on psychological profiling and deep self-disclosure—these probing techniques increase the chance of unconsented emotional manipulation, distress escalation, and collection of highly sensitive mental-health data.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This content explicitly guides the elicitation of self-defining, emotionally intense, and identity-linked autobiographical memories, which are highly sensitive personal data and may surface distressing material. The file includes techniques for deep disclosure and emotional probing but does not provide a clear warning about emotional risk, privacy sensitivity, optionality, or boundaries, making misuse or inadvertent harm more likely.

Ssd 4

Medium

Confidence: 93% confidence
Finding: The metadata and description frame the skill as a way to build psychological profiles through gradual disclosure and to elicit authentic self-disclosure, which normalizes extracting sensitive personal information through rapport rather than transparent purpose limitation. Even if presented as research or conversation design, the stated objective is profiling, making the technique more dangerous than ordinary interviewing guidance.

Ssd 4

Medium

Confidence: 89% confidence
Finding: The anti-pattern section advises earning disclosure gradually and starting with easier territory before moving deeper, which in this skill's broader context functions as operational guidance for staged extraction of intimate information. Because the overall goal is elicitation and profiling, trust-building here is not neutral rapport advice; it can be used to lower defenses before soliciting more sensitive disclosures.

VirusTotal

55/55 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.