Back to skill

Security audit

bossskill

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed business assistant that stores local CRM-style records and uses an optional licensed cloud service, so it is not malicious but should be used carefully with sensitive business data.

Install only if you are comfortable using a local business-memory database and, for paid features, sending the current command input plus license/device information to the vendor service. Do not store passwords, payment details, government IDs, unnecessary employee/customer private details, or trade secrets unless you have a clear business need and authorization. Use the official source and endpoint, protect the SQLite database, license cache, and exports, and avoid running it in an environment where someone else controls BOOSKILL_* environment variables.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (12)

Tainted flow: 'request' from os.environ.get (line 123, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Content
headers={"Content-Type": "application/json"},
        method="POST",
    )
    with urllib.request.urlopen(request, timeout=60) as response:
        return json.loads(response.read().decode("utf-8"))
Confidence
82% confidence
Finding
with urllib.request.urlopen(request, timeout=60) as response:

Context-Inappropriate Capability

Medium
Confidence
84% confidence
Finding
The file performs license checks, machine fingerprinting, and remote core-service communication that are not clearly necessary for a startup coaching/secretary skill as described. In this context, undisclosed remote dependency and host fingerprinting expand the data exposure surface and create privacy and supply-chain risk beyond the user’s likely expectations.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The payload labels privacy mode as 'no_local_database_upload', but it still transmits local user-supplied business content such as text, title, topic, owner, content_json, customer_id, and industry together with license and machine identifiers. This is misleading because users may infer that local/private data is not leaving the machine when substantial local content is still sent to the cloud.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The document states that authorization codes, device identifiers, feature names, and command parameters are sent to cloud services, but it does not indicate a clear end-user warning, consent flow, or privacy notice. Even if local SQLite data is not uploaded by default, device identifiers and command inputs may still contain sensitive business or personal information, creating privacy and compliance risk if users are unaware.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The marketplace description states that the authorization flow and cloud features send an authorization code, device identifier, function name, command parameters, and the user's current input to remote services, but it does not describe an explicit just-in-time warning or consent step at the moment data is transmitted. In a startup coaching skill, user inputs may contain sensitive business data, customer details, team information, or operational plans, so silent or insufficiently signposted cloud transmission creates a real privacy and data-handling risk.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The manifest explicitly advertises local database export functionality and network use for licensing/cloud features, but does not surface any user-facing warning, consent step, or data-handling notice in the operational prompt near those capabilities. Because this skill stores sensitive business data such as customers, team notes, budgets, and follow-up records, undisclosed export/transmission paths increase the risk of privacy leakage or unintended data disclosure.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The activation text is overly broad: phrases like 'what to do next' or wanting a 'follow-up/task operating system' can match many ordinary planning or advisory requests that are not actually asking to invoke this skill. In an agent setting, this can cause unintended skill activation, leading to irrelevant task-generation behavior, context hijacking from the user's actual goal, or unnecessary retention/structuring of business data.

Natural-Language Policy Violations

Medium
Confidence
83% confidence
Finding
The output pattern is hard-coded in Chinese, which can override the user's language preference and cause the agent to respond in an unexpected language. While not a direct security exploit, forced language can degrade usability, obscure important task details, and increase the risk of user misunderstanding in operational workflows like reminders, deadlines, and complaint handling.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
This file defines schemas for storing extensive employee and customer personal data, including contact details, relationship data, performance scores, sensitive notes, and inferred preferences. Although it briefly says to ask permission for customer sensitive data, it lacks clear data minimization, retention, access control, legal basis, and protection guidance, which makes privacy-invasive collection and misuse more likely in a business operations skill centered on CRM and memory storage.

Missing User Warnings

Medium
Confidence
78% confidence
Finding
The license request sends a license key and machine-derived identifier to a remote service without any visible user-facing notice in this file. Even if licensing is legitimate, silent transmission of identifying data is a privacy and transparency issue, especially in a business workflow tool handling potentially sensitive operator environments.

Missing User Warnings

High
Confidence
94% confidence
Finding
The cloud core request submits command arguments that may contain sensitive business or customer information, along with license and machine-identifying data, without visible warning or opt-in in this file. In a startup coaching/secretary context, these arguments are likely to contain confidential operational data, making silent cloud transmission more dangerous than in a purely public-data tool.

Missing User Warnings

Medium
Confidence
84% confidence
Finding
The export command returns the full contents of customers, team members, contacts, tasks, and projects, which can include personal data, birthdays, relationship notes, business pipeline details, and internal task information. In a skill context, bulk export without confirmation, redaction options, or access controls increases the chance of accidental disclosure to an unauthorized user, downstream tool, or logging surface.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.