ClawHub
Back to skill

Security audit

ia-orchestrating-swarms

Security checks across malware telemetry and agentic risk

Overview

This is a coherent multi-agent orchestration skill, with disclosed background-agent and local team-state behavior plus a privacy caveat for debugging inbox files.

Install only if you intend to coordinate multiple agents. Be aware that some debugging examples read local Claude team config, task, and inbox files, which may include prompts, messages, or operational context; use them only for teams you are authorized to inspect and clean up persistent teammates when finished.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (3)

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The documentation explicitly instructs users to read teammate inbox and task files under ~/.claude, which can expose message contents, task metadata, agent identities, and other potentially sensitive operational context. Even though presented as debugging guidance, it normalizes direct inspection of private agent state without any warning about sensitivity, increasing the risk of unnecessary data exposure or misuse.

Agent Config Directory Access

High
Category
Agent Snooping
Content
cat ~/.claude/teams/{team}/config.json | jq '.members[] | {name, agentType, backendType}'

# Check teammate inboxes
cat ~/.claude/teams/{team}/inboxes/{agent}.json | jq '.'

# List all teams
ls ~/.claude/teams/
Confidence
93% confidence
Finding
This example shows direct access to ~/.claude team configuration and inbox files, which are likely to contain internal coordination data, prompts, agent identifiers, and message history. In a multi-agent orchestration skill, that is more dangerous because these files may aggregate sensitive cross-agent context and could be abused for reconnaissance or unauthorized data harvesting.

Agent Config Directory Access

High
Category
Agent Snooping
Content
ls ~/.claude/teams/

# Check task states
cat ~/.claude/tasks/{team}/*.json | jq '{id, subject, status, owner, blockedBy}'

# Watch for new messages
tail -f ~/.claude/teams/{team}/inboxes/team-lead.json
Confidence
90% confidence
Finding
The commands enumerate teams, inspect task state files, and tail a team-lead inbox, all of which facilitate broad visibility into agent operations and message flow. In this skill's context, those artifacts are central to swarm coordination, so exposing them via documentation meaningfully increases the chance of leaking sensitive workflow, prompt, or task information.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.