ClawHub
Back to skill

Security audit

Agent Swarm Dev Team

Security checks across malware telemetry and agentic risk

Overview

This skill is a plausible multi-agent coding orchestrator, but it makes elevated, approval-bypassing agent execution the default and leaves key runtime scripts opaque or missing.

Install only if you intentionally want a powerful local coding swarm. Before use, remove or override the dangerous defaultArgs, inspect or supply the missing run-agent.sh and check-agents.sh scripts, change the hardcoded install path, and ensure monitoring and agent restarts require explicit per-task approval with a clear stop command.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (3)

Context-Inappropriate Capability

High
Confidence
99% confidence
Finding
The configuration assigns both coding agents dangerous permission-bypass flags as defaults, effectively disabling approval and sandbox controls for all runs. In an orchestration skill that coordinates multiple AI coding agents, this materially increases the chance of unintended filesystem changes, secret exposure, command execution, or destructive actions propagating across the development environment.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The default arguments include ambiguous but clearly high-risk flags that skip normal safety controls without any visible trigger constraints or contextual gating. Because they are always enabled in settings rather than conditionally applied, the skill makes unsafe execution the baseline behavior, which is especially dangerous for autonomous coding agents interacting with code, shells, and repositories.

Natural-Language Policy Violations

High
Confidence
98% confidence
Finding
The settings force permission-bypass behavior for both agents by default, with no evidence of informed user consent, policy justification, or compensating controls. In the context of a multi-agent development swarm, this is more dangerous because multiple agents may act concurrently or sequentially with elevated privileges, amplifying the blast radius of prompt injection, bad code generation, or operational mistakes.

VirusTotal

No VirusTotal findings

View on VirusTotal

Static analysis

Detected: suspicious.dangerous_exec

Shell command execution detected (child_process).

Critical
Code
suspicious.dangerous_exec
Location
scripts/init-swarm.cjs:65