Back to skill

Security audit

研究问题解析 (Agentic AI 科研平台)

Security checks for vulnerabilities and agentic risk

Overview

This skill looks like a demo parser, but it reads a fixed mock-data file instead of actually parsing the user's question, so it needs review before installation.

Install only if you expect a demo workflow tied to `/home/ubuntu/workspace/demo/mock_data/task_create.json` and a local reporting service on port 5001. Do not rely on it as a real research-question parser until it is changed to parse the user's actual input and clearly disclose any mock-data or status-reporting behavior.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
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 (5)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill's stated purpose is to parse user questions, but the instructions explicitly say to use mock/demo data even when the user's query is unrelated. This mismatch is dangerous because it normalizes deceptive behavior, undermines output trustworthiness, and can mask unauthorized data access from local files instead of processing the user's actual input.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documentation claims the skill parses the user's natural-language research question, but the procedure instead reads a fixed local file containing mock data. This is a security and integrity issue because the skill can access local filesystem content unrelated to the user's request, and it misrepresents what data is being processed, creating a path for unauthorized data exposure or deceptive outputs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest description and output requirements are written as mandatory Chinese-language behavior, including user-facing status messages and formatting guidance, with no indication that the user can choose another language. This creates a natural-language policy concern because the skill appears to impose a specific language/locale by default rather than offering an opt-in or respecting user preference.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The skill contains an HTTP POST to a local service, which is an external transmission relative to the skill boundary. Even though it targets localhost and the sample body contains only status text, this still creates an exfiltration/instrumentation channel that could expose user workflow metadata or be repurposed to send sensitive parsed content to another process.

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

1. 上报开始

bash
curl -s -X POST http://localhost:5001/api/report \
  -H "Content-Type: application/json" \
  -d '{"skill":"research-question-parser","display_name":"研究问题解析","status":"running","message":"正在解析研究问题,提取结构化参数..."}'

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill performs HTTP status-reporting requests that are not necessary for the stated purpose of parsing and structuring a research question. Even though the payload shown is only status metadata, this establishes an outbound communication path and could be extended to leak user-derived data or operational metadata without user awareness.

Content

No source excerpt is available for this finding.