Security audit

喂料专家五步投喂工作流

Security checks for vulnerabilities and agentic risk

Overview

This document-only skill has no executable code, but its knowledge-ingestion rules tell agents to remove third-party attribution, which creates provenance and compliance risk.

Review before installing. The package does not appear to execute code or access credentials, but users should override the no-attribution instructions and require source URLs, authors or publishers, dates, and licensing notes for third-party material before adding it to any knowledge base or RAG corpus.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
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)

Intent-Code Divergence

High

Confidence: 99% confidence
Finding: The instruction to rewrite third-party content and omit attribution before ingesting it encourages removal of provenance information and can facilitate copyright infringement, plagiarism, and laundering of untrusted material into the knowledge base. In this skill context, it is especially dangerous because the document governs repeated ingestion behavior, so the practice could systematically strip source traceability across the dataset.

Natural-Language Policy Violations

High

Confidence: 96% confidence
Finding: The template explicitly instructs users to rewrite external source material so that third-party attribution is removed, while retaining factual content. This creates a provenance-obscuring workflow that can facilitate plagiarism, misrepresentation of source authority, and defeat downstream verification or auditability, which is especially risky in a RAG or multi-agent setting where trust in source lineage is critical.

Natural-Language Policy Violations

High

Confidence: 98% confidence
Finding: Discouraging attribution of third-party sources undermines auditability and makes it harder to distinguish official material from rewritten external content, increasing legal, compliance, and integrity risks. Because this file is framed as a global rule set for all feeding behavior, the harmful effect is amplified: operators may repeatedly ingest unattributed material and later redistribute it as if it were original.

Static analysis

No suspicious patterns detected.