Back to skill

Security audit

Cca Quiz

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to be a Chinese-language CCA quiz, but it unnecessarily requests shell command access that is not needed to ask or grade quiz questions.

Review this skill before installing because it asks for shell access despite being a static quiz. Prefer a version with Bash removed, and ideally with only the minimum tool permissions needed to load and present the quiz.

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:1
Finding

Unnecessary Shell Access Violates Least-Privilege Requirements

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 1-5
Vulnerability Type: Excessive tool permissions
Risk Level: Medium

Vulnerable Code

yaml
---
name: cca-quiz
description: CCA 模拟测验,12道场景题覆盖全部5个领域。当用户说"CCA测验"、"模拟考试"、"cca quiz"、"测试CCA水平"时使用。
allowed-tools: Read, Bash
---

Technical Analysis

The skill grants access to both Read and Bash, although its documented function is limited to presenting a static 12-question quiz, evaluating answers, and reporting scores. All quiz questions, answers, and scoring instructions are embedded directly in SKILL.md; no shell execution or additional filesystem reads are required.

Bash provides a general-purpose command-execution channel. Depending on the host sandbox and process privileges, it can execute programs, inspect local files, modify accessible resources, and initiate network requests through installed command-line utilities. Granting this capability without a functional requirement violates the principle of least privilege and unnecessarily expands the consequences of prompt injection or future unsafe modifications to the skill.

The file does not instruct the agent to invoke Bash, and the audited package contains no scripts or payloads. Therefore, this finding identifies an exposed capability rather than evidence that commands are currently being executed maliciously.

Attack Path

  1. A user invokes the quiz skill, causing it to operate with the declared Bash permission.
  2. Attacker-controlled conversational content or a future modification introduces instructions that induce an unintended tool call.
  3. Because Bash is already authorized, the agent may execute a shell command rather than being blocked by the tool allowlist.
  4. The command can act with the filesystem, process, and network access available to the hosting agent.
  5. Accessible local information could be read or altered, or additional commands could be launched, subject to the host's sandbox and operatin ...[truncated 773 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove Bash from the skill's allowed-tools declaration.
  2. Remove Read as well if the platform does not require it to load the skill definition; the complete quiz content is already embedded in the skill.
  3. Prefer an empty tool allowlist for this purely conversational workflow. For example:
yaml
---
name: cca-quiz
description: CCA 模拟测验,12道场景题覆盖全部5个领域。当用户说"CCA测验"、"模拟考试"、"cca quiz"、"测试CCA水平"时使用。
allowed-tools: []
---
  1. If the platform requires Read, authorize only that tool and restrict it to the skill's own static resources where path-level controls are supported.
  2. Add a review requirement for future permission changes so that each newly authorized tool must have a documented functional justification.
  3. Enforce host-level sandboxing independently of the skill configuration, including restricted filesystem access, disabled outbound networking where unnecessary, and execution under a minimally privileged operating-system account.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description and core instructions are written as if the skill operates in Chinese by default, including the invocation phrases and the instruction that the agent is the exam writer/grader. There is no opt-in or alternative language path, which can violate language/locale policy for general-purpose skills.

Content

No source excerpt is available for this finding.