Back to skill

Security audit

数据分析理调

Security checks for vulnerabilities and agentic risk

Overview

The skill is a data-analysis guidance document, but it asks for command execution and broad filesystem authority that the document itself says it should not need.

Install only if you are comfortable reviewing or restricting its permissions. Prefer allowing it to read only user-selected datasets, and do not grant exec, write, glob, or administrator-level execution unless a specific future workflow clearly requires them and is separately approved.

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 (2)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:23
Finding

Unnecessary Command Execution and Filesystem Capabilities

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 23-27
Vulnerability Type: Excessive tool permissions that violate least privilege
Risk Level: Medium

Vulnerable Code

yaml
tools:
  - read
  - write
  - exec
  - glob

The declared capabilities are inconsistent with the statement at line 149 that this is a pure Markdown skill for which command-line execution is unnecessary.

Technical Analysis

The skill is a methodology framework for data analysis and contains no executable implementation. Nevertheless, it requests unrestricted command execution, filesystem modification, and broad file-discovery capabilities.

The exec, write, and glob permissions are not required for the documented statistical guidance. Granting them expands the authority available to the agent beyond the skill's legitimate needs. If attacker-controlled analysis input can influence tool selection or arguments, the unnecessary capabilities could be misused for local reconnaissance, command execution, or filesystem modification.

The file contains no direct instruction to exploit these tools, so this finding concerns an unsafe privilege boundary rather than confirmed malicious execution.

Attack Path

  1. A user or external data source supplies attacker-controlled content as analysis input.
  2. The content attempts to influence the agent into inspecting unrelated local paths or performing system operations.
  3. The agent relies on the skill's declared glob or read capability to discover and access files outside the intended dataset.
  4. The unnecessary exec or write capability is invoked to run a local command or modify an accessible file.
  5. The operation executes with the permissions of the agent process.

This path is conditional on the hosting agent honoring these tool declarations and failing to independently constrain attacker-controlled instructions.

Impact Assessment

Potential impact ...[truncated 481 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove exec, write, and glob from the declared tools.
  • Retain read only if the skill must inspect user-authorized dataset files.
  • Restrict permitted reads to an explicit workspace or allowlist of user-selected files.
  • Require separate, explicit user authorization before enabling filesystem modification or command execution for any future feature.
  • Enforce sandboxing and operating-system-level least privilege independently of the Markdown declaration.
  • Validate that untrusted dataset content cannot be interpreted as agent instructions.
  • Add automated policy checks that reject tool declarations inconsistent with the skill's documented functionality.

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:233
Finding

Unsafe Recommendation to Run with Administrator Privileges

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 233
Vulnerability Type: Unsafe privilege-escalation guidance
Risk Level: Medium

Vulnerable Code

The troubleshooting table contains the following guidance, translated into English:

markdown
| Insufficient permissions | The current user lacks read/write permissions | Check file permissions and run as administrator |

Technical Analysis

Recommending administrator execution as a generic response to insufficient file permissions violates least-privilege principles. A permission error normally indicates that the requested resource is outside the process's authorized scope, that ownership or access-control settings should be corrected narrowly, or that the operation should be stopped pending explicit authorization.

This guidance is especially risky because the same skill declares exec and write capabilities. Elevating the hosting agent would increase the authority of every command and filesystem operation performed during that run. The project does not contain code that automatically obtains elevation, so exploitation requires a user or operator to follow the recommendation.

Attack Path

  1. An operation encounters a read or write permission failure.
  2. The operator follows the troubleshooting guidance and restarts the agent with administrator or equivalent elevated privileges.
  3. The skill remains configured with command-execution and filesystem-write capabilities.
  4. Attacker-controlled input, an agent error, or an unintended tool invocation causes a command or file operation.
  5. The operation now executes with elevated privileges and may affect resources that were previously protected.

Impact Assessment

If the recommendation is followed, potential impact expands from the current user's accessible resources to those available to the elevated account. Depending on the platform and agent sandbox, this could include modification of ...[truncated 359 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove the recommendation to run the agent as an administrator.
  • Replace it with instructions to verify that the requested file belongs to the explicitly authorized workspace.
  • Grant only the minimum required permission to the specific file or directory rather than elevating the entire agent process.
  • Stop the operation and request explicit user authorization when a resource is inaccessible.
  • Run the agent in a sandbox under a dedicated, unprivileged account.
  • Deny access to system directories, credentials, other users' files, and unrelated project paths.
  • Log denied access attempts without recording sensitive file contents.
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)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description says to use the skill whenever users need data analysis, report generation, statistical insights, or data visualization, which are broad everyday requests that can overlap with many general-purpose assistant tasks. It does not provide specific trigger phrases, scope boundaries beyond one exclusion, or negative examples to clarify when this skill should or should not activate.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The description explicitly states '支持中文交互' and the document repeatedly frames the skill as operating in Chinese, but it does not offer the user a language choice or state that Chinese is optional. This creates a language-policy concern because it imposes a locale/language behavior without explicit user opt-in.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This skill is primarily a methodology and analysis-guidance document, so declaring command-execution capability is unjustified on its face. Unnecessary exec permission expands the attack surface and could let downstream prompt injection or accidental agent behavior trigger shell commands unrelated to the stated purpose of the skill.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill claims it is a pure Markdown, no-exec capability, yet the manifest grants exec access. This mismatch violates least privilege and can cause agents or operators to trust the skill as non-executing when it is actually allowed to run commands, increasing the chance of unintended code execution if later prompts or integrations invoke that tool.

Content

No source excerpt is available for this finding.