Skill创作发明家

• Outcome: Review
• Latest audit: Jul 28, 2026 · 6:06 AM UTC
• Version: 3.7.0

Overview

This is a disclosed local helper for creating, auditing, and publishing skills, with minor routing and README privacy caveats but no hidden or destructive behavior.

Install this if you want an agent to help author, audit, and prepare skills for publication. Be aware that skill-related requests may activate it broadly, the audit script can scan multiple local skill directories if no target is supplied, and the README includes remote images/badges even though the actual audit workflow is local.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (8)

Intent-Code Divergence

Medium

• Confidence: 88% confidence
• Finding: The README states that everything runs locally with no telemetry, data collection, or cloud calls, but the document includes externally hosted links and an embedded remote image. Rendering the README or clicking its links can cause network requests, which makes the privacy claim materially inaccurate and could mislead users operating in restricted or privacy-sensitive environments.

Vague Triggers

Medium

• Confidence: 73% confidence
• Finding: The trigger examples are broad natural-language phrases such as 'Create a skill that does X' and 'Audit my-skill for quality,' which could match ordinary user requests and invoke the skill unintentionally. In an agent environment, overbroad activation language can cause accidental execution of powerful workflows, including auditing, packaging, or script guidance, outside the user's intended context.

Vague Triggers

Medium

• Confidence: 80% confidence
• Finding: The trigger phrases in the workflow table are broad, natural-language requests such as asking to create, audit, or publish a skill. In an agent environment, overly generic triggers can cause accidental invocation during ordinary conversation, leading the skill to activate outside intended contexts and potentially perform file inspection or modification actions unexpectedly.

Vague Triggers

Medium

• Confidence: 82% confidence
• Finding: The example invocation phrases are short and broadly phrased, overlapping with common requests a user might make in general conversation. In systems that rely on semantic matching for skill activation, this increases the chance of unintended routing to this skill, which can expose workspace contents to the skill's audit logic or trigger unwanted workflow steps.

Vague Triggers

Medium

• Confidence: 80% confidence
• Finding: The eval uses a very generic prompt to create a new Skill without clearly constraining the trigger conditions. In a routing or policy-evaluation context, this can train or validate behavior that over-triggers the skill on ordinary 'create' requests, causing misrouting and unsafe delegation to the wrong workflow.

Vague Triggers

Medium

• Confidence: 77% confidence
• Finding: The audit eval prompt asks to 'check' a modified SKILL.md, but the expected behavior describes a broad audit mode without explicit scope boundaries. This can normalize triggering audit routines on vague review requests, potentially invoking privileged scanning or specialized workflows when a narrower code-review action was intended.

External Transmission

Medium

• Category: Data Exfiltration

Content

If this project helped you build better AI agent skills, give it a star ⭐

[![Star History Chart](https://api.star-history.com/svg?repos=muippt/mu-skill-creator&type=Date)](https://star-history.com/#muippt/mu-skill-creator&Date)

> The quality gatekeeper for AI agent skills — 53 items, 10 layers, zero compromises.

• Confidence: 83% confidence
• Finding: https://api.star-history.com/

External Transmission

Medium

• Category: Data Exfiltration

Content

如果这个项目帮你构建了更好的 AI Agent 技能，给个 star ⭐

[![Star History Chart](https://api.star-history.com/svg?repos=muippt/mu-skill-creator&type=Date)](https://star-history.com/#muippt/mu-skill-creator&Date)

> AI Agent 技能的质量守门员 —— 53 项，10 层，零妥协。

• Confidence: 93% confidence
• Finding: https://api.star-history.com/

VirusTotal

64/64 vendors flagged this skill as clean.

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical

• Code: suspicious.dynamic_code_execution
• Location: scripts/skill-audit.sh:195