ClawHub
Back to skill

Security audit

Agent Harness

Security checks across malware telemetry and agentic risk

Overview

The skill is a coherent ClawHub code-review helper, but its bundled helper defaults to launching nested Codex review with approval bypass and full filesystem sandbox escape, which deserves manual review before use.

Review this skill before installing or enabling it automatically. It is not showing deception or malware, but users should be comfortable with a review helper that can run nested Codex in full-access approval-bypass mode by default. Prefer `--no-yolo` or `AUTOREVIEW_YOLO=0`, choose fallback reviewers deliberately, and avoid sending sensitive local diffs to external reviewer CLIs unless that is intended.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Vague Triggers

Medium
Confidence
94% confidence
Finding
The manifest description and trigger language are overly broad, including common verbs like 'build', 'plan', 'review', 'ship', and 'debug' that can match many ordinary user requests. This can cause the skill to activate unintentionally and steer unrelated conversations into this workflow, increasing the chance of prompt hijacking, context pollution, or unsafe execution patterns in situations where the user did not intend to invoke the harness.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.