ClawHub

Self Improvement

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed local learning and reminder workflow, with optional hooks that are broad but currently only emit reminders.

Install this only if you want agents to keep local learning notes and reminders. Prefer project-scoped hooks over global hooks, avoid enabling the PostToolUse error detector unless you are comfortable with command output being inspected for error patterns, and do not store secrets, tokens, full transcripts, or raw private outputs in .learnings or promoted agent instruction files.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding
The declared description says the skill captures learnings and reviews them, but the body also prescribes hook installation, error-monitoring behavior, inter-session sharing, and automated skill extraction that create broader operational effects. This mismatch can cause users or orchestrators to enable a skill with more persistence, surveillance, and file-writing capability than its metadata suggests, reducing informed consent and making risky behaviors easier to smuggle in.

Vague Triggers

Low
Confidence
80% confidence
Finding
An empty hook matcher causes the hook to fire on every prompt, creating broad always-on interception behavior with unclear scope limits. In this skill's context, that can lead to persistent monitoring, unnecessary processing of all user inputs, and greater chance of logging sensitive content or normal conversation without sufficient need.

Session Persistence

Medium
Category
Rogue Agent
Content
└── FEATURE_REQUESTS.md
```

### Create Learning Files

```bash
mkdir -p ~/.openclaw/workspace/.learnings
Confidence
88% confidence
Finding
Create Learning Files ```bash mkdir -p ~/.openclaw

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal