ClawHub

SafeProactive

Security checks across malware telemetry and agentic risk

Overview

SafeProactive is a disclosed safety and audit-log framework, with broad activation and local log retention that users should configure carefully.

Install this only if you want a safety framework active across routine agent interactions. Review the trigger settings, keep WAL files in a protected local directory, define retention/redaction rules, and disable or gate any automatic external API/web queries unless they are explicitly needed.

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 (3)

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The README explicitly normalizes autonomous execution of Level 1 actions, including external API queries, without per-action human approval. In an agent skill, this is risky because even 'read-only' exploration can trigger unintended network access, interact with untrusted systems, or expose the operator to actions they did not realize would happen automatically.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The WAL examples show detailed reasoning, constraints, timestamps, sensor data, and execution results being written to disk, but the README does not give a prominent privacy warning about sensitive data retention. This can lead operators to unknowingly persist prompts, model reasoning, external API responses, or environmental data in plaintext logs, increasing exposure in the event of local compromise or log sharing.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The `every_turn` trigger causes the skill to activate on routine interactions, which increases the chance of unintended execution and unnecessary exposure of its logic on unrelated prompts. Although the skill text emphasizes safety controls such as local-only scope and human approval gates, broad invocation still expands attack surface and can interfere with normal agent behavior if those controls are imperfectly enforced.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal