Back to skill

Security audit

任务编排与调度 (Agentic AI 科研平台)

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent clinical research demo orchestrator, with disclosed local progress reporting and fixed mock-data reads but no hidden persistence or destructive behavior.

Install this only in the intended demo environment. Be aware that it will POST progress updates to a local monitoring service and read fixed mock clinical research files; do not use it on real patient data unless privacy, consent, de-identification, and monitoring-log handling are governed separately.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill repeatedly instructs the agent to send HTTP POST progress reports to a local monitoring endpoint even though its stated purpose is clinical research task planning. This creates an unnecessary network side effect and expands the trust boundary, enabling task metadata to be transmitted off the core workflow without clear user consent or justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill includes repeated background progress-reporting calls to a monitoring backend but does not warn the user that task metadata will be transmitted externally to the conversational response path. In a clinical research context, even seemingly harmless status messages can reveal sensitive study topics, workflow timing, and operational metadata that should be disclosed and governed.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The skill explicitly embeds a curl command that transmits status data to an HTTP endpoint, which is an external transmission channel relative to the skill's file-reading and planning purpose. Even though the destination is localhost, this still enables data egress to another service and could be abused if that service is misconfigured, logs sensitive metadata, or bridges to broader systems.

Content

Scanner excerpt · SKILL.md (reported line 24)May include surrounding context.

步骤 0:启动任务

bash
curl -s -X POST http://localhost:5001/api/report \
  -H "Content-Type: application/json" \
  -d '{"skill":"task-planner","display_name":"任务编排与调度","status":"running","message":"开始拆解研究任务,规划 5 步执行计划..."}'

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill instructs the agent to read and summarize mock clinical and patient-related datasets without any explicit privacy, minimum-necessary-use, or de-identification guidance. In healthcare-adjacent workflows, the absence of handling constraints increases the risk that similar real deployments will process or reveal protected or sensitive information inappropriately.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The manifest description and all user-facing instructions are Chinese-only, which effectively fixes the interaction language without documenting user choice or opt-in. The policy requires flagging language or locale constraints unless the skill offers a choice or clearly justifies the restriction.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file documents a sequence consisting of 步骤 0 through 步骤 5, with multiple prior sections clearly enumerated. The final status payload at L132 says '全部 4 个执行步骤已完成', which conflicts with the documented step structure and can misrepresent what the skill actually executed.

Content

No source excerpt is available for this finding.