Back to skill

Security audit

证据检索与 Grounding (Agentic AI 科研平台)

Security checks for vulnerabilities and agentic risk

Overview

This skill is a narrowly scoped evidence-summary helper with disclosed local status reporting and no signs of persistence, credential access, destructive behavior, or external exfiltration.

Install this only in an environment where a localhost reporting service on port 5001 is expected. The skill appears safe for its stated demo-style evidence grounding use, but it relies on a fixed mock data file rather than live PubMed retrieval and will send simple run-status messages to a local service.

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
  • 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 (3)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill description uses broad trigger language around literature/evidence retrieval and grounding, which can overlap with common research-oriented user requests and cause unintended activation. While this is not an exploit by itself, overbroad activation increases the chance that the skill runs in contexts the user did not explicitly intend, including contexts where its reporting behavior or file access would be unexpected.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly instructs the agent to send HTTP POST requests to a local service for status reporting, but the user-facing documentation does not clearly disclose that network communication will occur. Hidden or undisclosed outbound communication is risky because it can leak metadata about user activity and normalize covert reporting behavior inside a skill.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
96% confidence
Finding

The embedded curl command performs an HTTP POST to localhost:5001, causing external process execution and network transmission from within the skill. Even though the target is localhost, this can still expose execution metadata to another local service, be abused by a malicious listener on that port, or create an unreviewed side channel from the agent environment.

Content

Scanner excerpt · SKILL.md (reported line 22)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":"evidence-retriever","display_name":"证据检索与 Grounding","status":"running","message":"正在检索 PubMed、指南库与试验注册信息..."}'