Security audit

智能会议机器人

Security checks across malware telemetry and agentic risk

Overview

This meeting-bot skill describes joining, listening to, transcribing, and saving meetings, but it does not clearly define consent, retention, or limits on automatic participation.

Install only if you are prepared to tightly control when it can join meetings, confirm all participant consent and platform policy requirements, and decide where transcripts are stored, who can access them, and when they are deleted. Treat the requested read/write/exec authority and generic callback/API configuration as needing manual review before use.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (8)

Intent-Code Divergence

Medium

Confidence: 91% confidence
Finding: The skill’s description mixes a meeting-joining bot with unrelated project-management and workflow-planning uses, which broadens the apparent scope of when the agent may invoke it. This ambiguity is dangerous because it can cause the skill to activate in contexts where users did not intend meeting access, transcription, or other higher-privilege behavior.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The later usage scenarios advertise broad capabilities like file batch processing and environment management that do not match a meeting bot’s stated purpose. In an agent ecosystem, this mismatch can expand trust boundaries and lead users or orchestrators to grant unnecessary tools or invoke the skill for unrelated tasks, increasing abuse potential.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The activation language is overly broad and overlaps with common project-management requests, making accidental or unintended triggering more likely. Because this skill is capable of joining meetings and handling sensitive conversation data, vague triggering materially increases the risk of privacy-invasive actions occurring without clear user intent.

Vague Triggers

High

Confidence: 95% confidence
Finding: The skill claims it will proactively join meetings when one starts, but it does not define what signal authorizes that behavior or what boundaries prevent silent autonomous participation. In context, this is especially risky because joining a live meeting implies access to audio, participant identity, and potentially confidential business discussions.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The skill describes automatic joining, listening, transcription, and record handling without a prominent user-facing privacy warning or consent model. This is dangerous because it normalizes covert monitoring and collection of sensitive meeting content, potentially violating participant expectations, organizational policy, or law.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The documentation states that full transcripts and records are saved, but it does not clearly warn about retention of potentially sensitive business, personal, or regulated information. This increases the chance that users enable the feature without understanding that lasting copies of confidential discussions will be created and stored.

Ssd 3

Medium

Confidence: 90% confidence
Finding: Automatically saving full meeting transcripts and records creates a data retention surface that can expose confidential conversations, credentials spoken aloud, strategic plans, or personal data. The risk is amplified because the skill is designed for continuous meeting participation rather than one-off user-supplied text.

Ssd 3

Medium

Confidence: 92% confidence
Finding: Continuous real-time transcription, speaker tracking, and summary generation inherently collect and attribute participant data, which can reveal identities, opinions, and behavioral patterns. Without strict consent, minimization, and access boundaries, these features can become a privacy and surveillance risk even if the author’s intent is functional rather than malicious.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.