Security audit

imrc-report

Security checks across malware telemetry and agentic risk

Overview

This skill is a plausible internal operations-report tool, but it can produce authoritative-looking reports from hard-coded figures and aggregates internal message/personnel information without clear privacy or scoping controls.

Install only if the user is authorized to access the IMRC pages and Meixin messages being summarized. Treat outputs as sensitive internal reports, verify all figures against source systems before sharing, and avoid relying on the generated executive summary until the hard-coded placeholder data is replaced or clearly labeled.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Intent-Code Divergence

Medium

Confidence: 97% confidence
Finding: The file-level description states the report is generated by combining IMRC data with Meixin messages, but the implementation loads the Meixin file and then ignores its contents in favor of a hard-coded summary. In a reporting skill, this creates a data integrity problem: operators may trust the report as current and evidence-based when it is actually stale or fabricated, which can mislead decisions and conceal real operational changes.

Intent-Code Divergence

Medium

Confidence: 98% confidence
Finding: The summary-generation function claims to produce the first page from input data, but it returns a fully hard-coded report with fixed metrics instead of using IMRC data. Because this script produces management-facing operational reports, hard-coded KPIs can directly cause false business reporting, bad prioritization, and unnoticed anomalies; the skill context makes this more dangerous than a normal placeholder because users are likely to rely on the output as authoritative.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The trigger list includes broad operational phrases such as monthly summary, annual summary, and operations status that are likely to appear in normal user conversations. This can cause unintended invocation of the skill, which is more concerning here because the skill is designed to pull internal IMRC data and compile reports, potentially exposing sensitive business information without clear user intent.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill explicitly instructs collection of internal Meixin messages involving team collaboration, technical sharing, and personnel changes, but provides no privacy notice, consent boundary, data minimization rule, or handling restrictions for sensitive content. Because these messages may contain personal, confidential, or business-sensitive information, the skill creates a clear risk of unauthorized aggregation and disclosure in generated reports.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The script persists collected message data to disk in a workspace memory directory and allows an arbitrary output directory via a CLI argument, but it provides no warning, consent flow, minimization, retention control, or access protection for potentially sensitive internal communications. In the context of a message-collection tool for team collaboration, technical sharing, and personnel changes, the stored JSON may contain private or business-sensitive content, so unguarded local persistence increases exposure if the filesystem is shared, backed up, or later consumed by other tools.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.