Security audit

Soul Decision Engine Free

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent decision-support tool, but it persistently profiles user decision habits and can update or reuse that profile from broad conversational cues without strong consent boundaries.

Review this before installing if you use shared machines or discuss sensitive work or life decisions. It stores decision history and inferred preferences under ~/soul-decision, and you should only use it if you are comfortable with that local profile being reused across sessions; consider requiring explicit confirmation before any memory write.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (8)

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The example triggers are broad enough to activate on ordinary requests like 'help me decide X', which can cause the skill to engage decision-memory and profiling behavior without a clearly bounded invocation condition. In this skill, that matters because activation is coupled to reading and potentially writing persistent local memory, so over-triggering increases privacy and autonomy risk.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The proactive activation rules rely on ambiguous conversational cues such as anxiety, hesitation, or discussing multiple options, which can be present in many unrelated conversations. Because the skill then proactively applies stored preferences and decision frameworks, this creates a risk of unsolicited profiling, unwanted intervention, and accidental persistence of sensitive user context.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The quick-start instructions create local directories and files for persistent decision history, but the skill does not present a clear, prominent warning that it will store user decisions, preferences, and potentially sensitive personal data on disk. This is dangerous because users may reveal intimate business, technical, or personal decision context without understanding that it will be retained locally across sessions.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The skill explicitly says it will continuously learn user style, risk preference, and framework preference from decision reviews, which constitutes persistent behavioral profiling. In context, this is sensitive because the subject matter includes career, business, architecture, and personal life decisions, allowing the agent to accumulate a rich profile over time.

Ssd 3

Medium

Confidence: 91% confidence
Finding: The design states that HOT memory is fully loaded at session start, meaning previously stored preference and risk-profile data is broadly reused in each conversation. That increases the chance of overexposure, inappropriate cross-context use, and accidental disclosure of sensitive prior information when the current conversation does not require it.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The skill automatically infers and stores risk-preference signals from casual phrases like 'I'd rather wait for more data' or 'I don't want to regret it.' This is dangerous because it turns ordinary conversation into persistent behavioral profiling without a reliable indication that the user intended those statements to become stored long-term attributes.

Ssd 3

Medium

Confidence: 95% confidence
Finding: The skill persistently captures framework preferences from user phrasing, building a behavioral profile across interactions. While not obviously malicious, this creates a durable record of how the user thinks and decides, which can be sensitive and may be reused in ways the user did not expect.

Ssd 3

Medium

Confidence: 90% confidence
Finding: The skill includes commands to print the user's stored decision style and risk profile from memory files, which can expose sensitive behavioral data on demand. If invoked in a shared environment, by mistake, or through an over-broad trigger path, this could reveal personal profiling information that the user did not intend to surface in that moment.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.