Team Projects

• Outcome: Review
• Latest audit: May 29, 2026 at 9:25 PM UTC
• Version:
  1.3.0

Overview

This is a real multi-agent project-management skill, but it asks for broad agent permissions and includes under-scoped local execution paths that should be reviewed before use.

Review the source patches before installing. Replace wildcard agent allowlists with explicit project-team agents, avoid broad session visibility unless needed, and change the dashboard refresh to a dedicated read-only project API instead of a silent sessions.send command. Do not enable the optional localhost HTTP API unless you add authentication, restricted CORS, and clear controls for mutating project data.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Context-Inappropriate Capability

High

• Confidence: 98% confidence
• Finding: The UI refresh path sends a free-form instruction to an agent session telling it to run a local Node command and return raw JSON. That makes a presentation-layer component depend on agent-mediated command execution rather than a constrained data API, expanding the trust boundary and creating a dangerous execution primitive if the message path, script path, or surrounding agent behavior can be influenced.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The installation instructions explicitly recommend wildcard inter-agent permissions via tools.agentToAgent.allow and subagents.allowAgents set to ["*"], which grants the coordinator broad ability to contact or spawn any configured agent. In a multi-agent environment, this expands blast radius substantially: a compromised or misprompted coordinator can delegate sensitive data, trigger unintended actions, or laterally move work to higher-privilege agents without meaningful restriction.

Missing User Warnings

Medium

• Confidence: 88% confidence
• Finding: The code instructs the agent to run the command 'silently' and provides no UI disclosure that refreshing the dashboard triggers local command execution and filesystem-backed retrieval. In a multi-agent project-management skill, hidden execution is more dangerous because users may treat the dashboard as passive display logic while it is actually causing privileged backend activity.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The walkthrough explicitly recommends enabling inter-agent messaging with `allow: ["*"]` and broad session visibility, which permits unrestricted cross-agent communication and increases the risk of unnecessary data sharing between agents. In a multi-agent project-management skill, agents may handle different user data, prompts, files, or tool outputs, so documenting a permissive default without clear privacy boundaries or least-privilege guidance can lead to cross-agent data exposure or misuse.

VirusTotal

63/63 vendors flagged this skill as clean.