Back to skill

Security audit

Accounting And Finance

Security checks for vulnerabilities and agentic risk

Overview

This finance-analysis skill is mostly documentation, but it asks for command execution and broad file/API abilities that are not clearly needed or scoped.

Review this skill before installing. It does not appear malicious, but only install it in an environment where exec/write access can be denied or tightly sandboxed, and avoid providing confidential financial data unless callback/API behavior is disabled or explicitly controlled.

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:26
Finding

Unnecessary Command-Execution Capability Violates Least Privilege

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 26–29
Vulnerability Type: Excessive tool permission
Risk Level: Medium

Vulnerable Code

yaml
tools:
- read
- write
- exec

The inconsistency is confirmed elsewhere in the same file:

markdown
- **Category**: Markdown-only instructions; command-line execution capability is unnecessary.

The document also claims command-execution functionality:

markdown
- **Command execution**: Execute system commands in a secure sandbox and collect the results.

Technical Analysis

The skill requests the exec tool even though it describes itself as a pure Markdown instruction set that does not require command-line execution. Its documented financial calculations and report generation do not establish a legitimate need for arbitrary system-command access.

If the hosting agent grants capabilities from this metadata, loading the skill unnecessarily exposes a command-execution channel. This expands the attack surface and violates the principle of least privilege. Untrusted financial data or later task instructions could potentially induce the agent to use this capability outside the skill's legitimate analytical scope.

The reviewed file does not contain a command that directly exploits this access. The issue is therefore an excessive-permission vulnerability rather than evidence of an embedded malicious payload.

Attack Path

  1. A hosting platform loads SKILL.md and grants the tools declared in its metadata.
  2. The skill receives exec access despite requiring only document processing and financial analysis.
  3. An attacker supplies crafted task content or financial-document text intended to induce command execution.
  4. The agent follows the malicious content and invokes exec.
  5. Commands execute with the filesystem, process, and network permissions available inside the hosting sandbox.
  6. The resulting damage depends on sandbox isolation and may include unauthorized fi ...[truncated 647 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove exec from the declared tool list.
  2. Retain only narrowly scoped read access needed to process user-provided financial documents.
  3. Retain write access only if writing reports to files is an explicit requirement; otherwise remove it as well.
  4. Remove generic command-execution claims from the documentation.
  5. If a future feature genuinely requires command execution:
    • Define the exact permitted commands and arguments.
    • Enforce an allowlist in platform code rather than relying on prompt instructions.
    • Prohibit shell interpretation and user-controlled command concatenation.
    • Run commands in a restricted sandbox with minimal filesystem and network access.
    • Apply strict time, CPU, memory, and output limits.
    • Log and audit every invocation.
  6. Replace the generic recommendation to run with administrator privileges when permission errors occur with guidance to verify ownership, select an authorized destination, or contact the responsible administrator.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (6)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill simultaneously claims it is pure Markdown that does not require exec while the manifest grants exec capability and other sections describe command execution. This kind of capability/documentation mismatch can mislead reviewers and users into approving a skill with unnecessary code-execution privileges, increasing the chance of arbitrary command execution if later prompts or workflow logic invoke exec.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The summary says the skill includes 58 professional finance analysis capabilities, implying they are part of the skill. However, the paid-feature table at L050-L052 states core enterprise financial analysis features and named models are not supported in the free version, creating a direct contradiction about what this skill actually provides.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The phrase "Use when 用户需要财务相关功能时使用" is extremely broad and could match many ordinary finance-related conversations without clearly defining trigger boundaries. The description gives only a vague exclusion for "complex needs" and does not provide specific trigger phrases or negative examples to narrow invocation scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The description states "支持中文交互" and the display/content strongly positions the skill around Chinese-language interaction, but does not indicate that users may choose another language. This can violate a language/locale policy when a specific language is effectively imposed without opt-in or documented justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The skill mentions callback URLs and external/API-style behavior without clearly warning users that provided data may be transmitted off-platform. In a finance context, inputs may contain nonpublic financials, customer records, or other sensitive business data, so unclear transmission boundaries raise confidentiality and compliance risks.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The skill advertises broad file handling, API integration, and command execution capabilities unrelated to a finance-analysis use case. Unnecessary general-purpose capabilities expand the attack surface, making it easier for a prompt injection or user-supplied task to pivot the skill into data access, exfiltration, or arbitrary system interaction beyond intended financial analysis.

Content

No source excerpt is available for this finding.