Back to skill

Security audit

安全公告流监控专业版

Security checks for vulnerabilities and agentic risk

Overview

The skill’s advisory tracking and optional notifications fit its purpose, but its example code needs repairs before reliable use.

Repair and test the examples before relying on alerts. Independently verify affected versions and scores, and preserve a trusted integrity baseline. Configure only webhook destinations you trust: notifications include advisory descriptions and other advisory details. The review covered the complete bundled artifact, but did not establish that the monitor works at runtime.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Product correlation marks installed versions as affected without performing the documented version-range analysis.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The capability documentation at L410-L411 claims affected-product matching and identification of affected version ranges. However, L99-L105 checks only whether the product name appears in the advisory, copies the installed version without evaluating it, and unconditionally sets is_affected to true. Consequently, patched or otherwise unaffected installed versions can be reported as affected, misleading security triage.

Content

No source excerpt is available for this finding.

The Webhook workflow transmits advisory content to an external endpoint without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The setup workflow invokes send_webhook at L401, which posts advisory titles, descriptions, links, identifiers, and exploitability scores to the configured endpoint. Although Webhook notifications are advertised, the skill does not warn that user-supplied advisory content will leave the local environment; the general HTTPS and secret-handling precautions at L470–L475 do not disclose this data transfer.

Content

No source excerpt is available for this finding.