Back to skill

Security audit

Gov Disclosure Check(政务公开披露校验)

Security checks for vulnerabilities and agentic risk

Overview

This local document checklist has inconsistent permissions and overstated validation claims, but shows no harmful behavior.

Use only text or documents you choose, and restrict file access through the host. Do not grant write or web access for this checker. Treat its result as a keyword checklist requiring human review before publication. Permission declarations and CLI documentation need correction; investigation of the scanner's partial coverage found no material missing executable component.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (4)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The manifest declares filesystem access disabled, but the documented workflow requires reading local documents and the implementation supports local file reads. This is a permission-contract mismatch that could expose unpublished government material to unauthorized processing if the host trusts the declaration without enforcing it; no sandbox bypass or external transmission is demonstrated.

Content

No source excerpt is available for this finding.

The prescribed workflow requires filesystem access and console execution despite explicitly denying those permissions.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The workflow instructs the agent to read input files and execute a Python script despite filesystem and subprocess permissions being explicitly disabled. The allowed-tools list also includes Bash and other capabilities inconsistent with those restrictions, creating conflicting authorization signals for an agent handling potentially sensitive documents.

Content

No source excerpt is available for this finding.

The --file option reads local files despite the manifest explicitly disabling filesystem access.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --file argument reaches open(...).read() without application-level path restrictions, contradicting the declared prohibition on filesystem access. If execution is permitted without host enforcement, a caller can cause any UTF-8-readable file accessible to the process to be processed; the script reports keyword presence rather than raw contents and contains no network exfiltration.

Content

No source excerpt is available for this finding.

Unconditional trigger phrases include generic reply-checking requests outside government disclosure.

Low
Category
Trigger Abuse
Confidence
92% confidence
Finding

The skill explicitly requires activation on any listed phrase, including "batch check old replies for missing items," which could equally refer to customer-support messages or ordinary correspondence. Unlike the government-specific purpose, this activation instruction provides no requirement that the replies concern government information disclosure and no exclusion for unrelated contexts.

Content

No source excerpt is available for this finding.