ClawHub
Back to skill

Security audit

teamwork

Security checks across malware telemetry and agentic risk

Overview

The skill is coherent for team orchestration, but it asks agents to run on schedules, send external notifications, and automatically commit and push repository changes without clear approval boundaries.

Review before installing in any shared or sensitive repository. Use it only where autonomous scheduled agents are allowed to modify task state, archive project files, notify DingTalk/email recipients, and push to Gitee; otherwise add human approval, protected branches, recipient allowlists, and redaction rules first.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Vague Triggers

Medium
Confidence
75% confidence
Finding
The skill defines broad activation and execution behavior for a coordinator role without clear scoping or user-consent boundaries, which can cause the agent to autonomously act on repository task files whenever it encounters them. In this skill’s context, that ambiguity is more dangerous because it is paired with downstream actions like task dispatch, notifications, and git push, increasing the chance of unintended operational changes.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The skill explicitly instructs agents to automatically run git add, commit, and push, which can modify and publish repository state without an explicit confirmation step. This is especially dangerous here because the skill treats push as mandatory completion behavior, creating a strong risk of unintended code/data publication, propagation of mistakes, and exfiltration of sensitive material committed by automation.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The skill directs automated DingTalk notifications using task and result metadata but does not require a privacy review, consent check, or data minimization guardrail before sending content to an external service. In this workflow, that risk is elevated because task summaries, blockers, and outputs may contain internal project information, enabling unintentional disclosure to third parties or the wrong recipient.

Natural-Language Policy Violations

Medium
Confidence
62% confidence
Finding
The skill appears to mandate a specific language/locale in operational outputs without checking user preference, which can reduce usability and lead to misunderstandings in multilingual environments. While not as severe as code execution or data exfiltration, it becomes relevant in this orchestration context because status updates, audit records, and human escalation messages may be misread or mishandled if forced into one locale.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.