Back to skill

Security audit

云存储

Security checks for vulnerabilities and agentic risk

Overview

This cloud-storage skill asks for broad local read/write/command authority but is poorly scoped and internally inconsistent, so it needs review before installation.

Review this skill carefully before installing. Use it only with least-privilege cloud credentials, avoid administrator execution, and require explicit confirmation for uploads, permission changes, migrations, or any operation that can expose, move, or modify cloud data.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (1)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:23
Finding

Overprivileged Command Execution Capability and Unsafe Administrator-Elevation Guidance

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:23-26 and SKILL.md:202
Vulnerability Type: Excessive tool privileges and unsafe privilege-elevation guidance
Risk Level: Medium

Vulnerable Code

SKILL.md:23-26 declares broad local capabilities:

yaml
tools:
- read
- exec
- write

SKILL.md:202 instructs users handling insufficient permissions to check file permissions and run the operation as an administrator. This is an English translation of the original troubleshooting instruction.

Technical Analysis

The Skill requests unrestricted local read, write, and command-execution capabilities even though its documented purpose is cloud-storage management. It does not define an executable implementation, a command allowlist, argument validation rules, permitted filesystem locations, or provider-specific restrictions that would constrain use of these tools.

The permission-error guidance additionally recommends running as an administrator without identifying the exact resource or permission required. A generic elevated retry can cross a least-privilege boundary if the failed operation was influenced by untrusted input. The risk is heightened because cloud-management environments commonly contain provider CLI configuration files, access tokens, and other credentials.

No executable script or concrete elevation mechanism is included in the reviewed package. Exploitation therefore depends on an Agent or operator interpreting the Skill documentation and executing an attacker-influenced command with the declared tools.

Attack Path

  1. The Skill is loaded with read, write, and exec access.
  2. An attacker supplies a crafted file path, cloud object name, provider argument, or other input that is incorporated into a local command by the hosting Agent.
  3. The operation fails because the current account lacks access to the targeted local resource.
  4. The generic troubleshooting guidance is fo ...[truncated 1060 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove exec unless command execution is demonstrably required for a specific cloud-storage operation.
  2. Replace unrestricted tools with provider-specific APIs or narrowly scoped wrappers.
  3. Define an explicit allowlist of executable programs, operations, and accepted arguments.
  4. Validate provider names, operation names, local paths, object names, bucket names, and destination paths against strict schemas.
  5. Avoid shell command construction. Invoke commands with structured argument arrays and disable shell interpretation.
  6. Restrict filesystem access to explicitly approved working directories and prevent traversal through canonical-path validation.
  7. Remove the generic instruction to run as administrator. Instead, identify the exact resource requiring access and grant only the minimum permission necessary.
  8. Require explicit user confirmation before destructive operations, cross-cloud migration, permission changes, or any privileged action.
  9. Use separate, least-privileged cloud credentials for each provider and operation. Do not expose general-purpose administrator credentials to the Skill.
  10. Add tests confirming that untrusted input cannot alter command structure, access paths outside approved directories, or trigger elevated execution.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest description conflates cloud file management with unrelated database/SQL/data-storage tasks, which can cause the agent to invoke this skill in contexts far outside its intended scope. In a skill with read/exec/write capabilities, this ambiguity increases the risk of inappropriate execution against unrelated user data or infrastructure.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation language is overly broad and internally inconsistent, including common file-processing and database-related phrases that could match many unrelated user requests. Because the skill exposes exec/write/read tools and targets cloud resources, accidental triggering could lead to unintended file transfer, enumeration, modification, or cost-incurring operations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill describes cloud file operations but lacks a clear warning that actions may be destructive, privacy-impacting, or billable, especially when operating across providers. In this context, users or calling agents may initiate uploads, downloads, permission changes, or migrations without adequate confirmation or understanding of data exposure and operational consequences.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The documented output example shows a code-review scoring report instead of storage-operation results, creating a strong mismatch between advertised behavior and actual expected outputs. This can mislead orchestration logic or users into trusting the skill in the wrong workflow, causing incorrect automation decisions and masking potentially destructive storage actions.

Content

No source excerpt is available for this finding.