Skill Dispatcher

• Outcome: Review
• Latest audit: May 28, 2026 · 11:23 AM UTC
• Version: 1.1.0

Overview

This skill is not malware, but it acts as a broad always-on skill router and asks agents to make persistent changes to routing, memory, and installed skill state.

Install only if you intentionally want a global skill-routing and skill-maintenance layer. Before using it, disable external registry search and require explicit user approval before it edits core rules, memory, routing tables, packaged skill files, or installed skills.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Rogue Agent: Self-Modification, Session Persistence
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (6)

Context-Inappropriate Capability

Medium

• Confidence: 88% confidence
• Finding: The fallback protocol tells the agent to search external registries such as Clawhub when no local match is found, introducing network-based discovery and supply-chain expansion from inside a router skill. For a broadly triggered dispatcher, this can cause unexpected external interactions and pull the agent toward unvetted capabilities.

Context-Inappropriate Capability

Medium

• Confidence: 83% confidence
• Finding: The new-skill onboarding flow includes review, commit, and log integration actions, which materially change the workspace and system state rather than merely selecting an existing skill. Embedding change-management steps in a router increases the chance of unauthorized or accidental modifications during ordinary task routing.

Vague Triggers

High

• Confidence: 92% confidence
• Finding: The skill says it triggers on any non-trivial task and should proactively prevent forgetting skills, which is an extremely broad activation scope. In practice this can let the dispatcher intercept most conversations, override normal tool-selection boundaries, and amplify any unsafe downstream instruction embedded in its routing logic.

Vague Triggers

Medium

• Confidence: 86% confidence
• Finding: The decision tree is full of placeholders and ambiguous categories, leaving invocation criteria underspecified while still instructing the agent to route or chain skills. Ambiguous routing logic is risky because a broadly loaded meta-skill can misclassify tasks and trigger unintended tools or workflows without a clear policy basis.

Vague Triggers

Medium

• Confidence: 87% confidence
• Finding: The enhancer layer auto-applies additional skills based on broad natural-language conditions like long text, complex tasks, or corrections, without user opt-in or strict constraints. This creates hidden secondary behavior where extra skills may be invoked implicitly, increasing the attack surface and reducing transparency.

Self-Modification

High

• Category: Rogue Agent

Content

Only proceed to Step 5 after ALL 3 rounds pass.

## Step 5: Commit
1. Edit SKILL.md — insert routing entry into the correct table section
2. If CP: add branch to Q1 decision tree
3. Update `references/skill-capabilities.md` with the new skill's capabilities
4. Also add the skill name to the Covered Skills Registry at the bottom of SKILL.md

• Confidence: 98% confidence
• Finding: Edit SKILL

VirusTotal

66/66 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.