ClawHub
Back to skill

Security audit

Skill Collision Engine

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only meta-skill for generating and evolving other skills, but it asks for broad recurring access to skills, logs, external sources, and conversation records without clear user approval or scope limits.

Install only if you explicitly want a meta-skill that can influence other skills. Use it manually, approve each source and generated skill before reuse or publication, restrict which logs and external content it may read, avoid private conversations or secrets, and verify any separate implementation before running the Python examples.

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
91% confidence
Finding
The skill defines collision inputs as effectively arbitrary skills and arbitrary external knowledge, with no meaningful trigger boundaries, trust model, or exclusion criteria. In a meta-skill that combines other skills and ingests external content, this broad scope increases the chance of prompt injection, unsafe chaining, and unintended processing of sensitive or adversarial material.

Vague Triggers

Low
Confidence
87% confidence
Finding
The usage section shows powerful operations but does not define when the skill should or should not activate, leaving invocation decisions overly open-ended. For a framework-generation engine, ambiguous activation can cause accidental execution in contexts involving untrusted content, private data, or tasks outside intended boundaries.

Ssd 3

Medium
Confidence
94% confidence
Finding
The skill explicitly frames 'files as memory' and positions human-machine closed loops as a core pattern, which creates a natural path for retaining user-provided text and reusing it in later operations. In a system that synthesizes new skills from prior inputs, this can lead to cross-session leakage, unconsented retention, and propagation of sensitive natural-language data into derived artifacts.

Ssd 3

Medium
Confidence
97% confidence
Finding
The external knowledge source list explicitly ranks human conversation records as 'highest value,' encouraging ingestion and reuse of dialogue logs as knowledge inputs. That is a direct privacy and data-governance risk because conversations often contain secrets, personal data, or proprietary information, and the engine is designed to transform inputs into reusable outputs.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.