Back to skill

Security audit

助手

Security checks for vulnerabilities and agentic risk

Overview

This skill looks like a generic assistant, but it asks for broad file and command powers without enough boundaries or clear user consent.

Install only if you are comfortable granting a generic assistant file read/write and command-execution capability. Use it in a sandbox or restricted workspace, do not run it with administrator privileges, and require explicit confirmation before it modifies files, runs commands, sends messages, or uses credentials.

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:35
Finding

Unrestricted Tool Permissions Violate Least-Privilege Requirements

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:35-38
Vulnerability Type: Excessive and insufficiently constrained tool permissions
Risk Level: Medium

Complete Vulnerable Configuration:

yaml
tools:
  - read
  - exec
  - write

The skill also provides generic guidance to run with administrator privileges when file permissions are insufficient at SKILL.md:216, and instructs the agent to execute commands at SKILL.md:236.

Technical Analysis

The skill declares unrestricted file-reading, file-writing, and command-execution capabilities even though its documented primary functions are task organization, schedule management, and message drafting. The configuration does not define permitted paths, approved commands, argument validation, sandbox boundaries, or confirmation requirements.

Granting exec alongside filesystem read and write access creates a broad authority surface. If attacker-controlled task descriptions, paths, or command parameters are treated as operational instructions, the agent could execute unintended commands or access files unrelated to the skill's legitimate purpose. The generic recommendation to use administrator privileges further increases the risk because an ordinary permission error could lead to execution with elevated operating-system authority.

No malicious command, direct privilege-escalation implementation, or confirmed data-exfiltration mechanism is present in the reviewed project. Exploitation therefore depends on how the hosting agent interprets user-controlled input and enforces tool permissions.

Attack Path

  1. A user or attacker supplies a crafted task containing a filesystem path, command argument, or instruction to process an external resource.
  2. The agent interprets that input as an operation and invokes the unrestricted read, write, or exec capability.
  3. The operation accesses or modifies data outside the workspace, or executes an uninten ...[truncated 882 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the exec capability unless a specific, documented feature requires local command execution.
  2. Restrict file reads and writes to a dedicated workspace directory using canonical-path validation.
  3. Define an explicit allowlist of permitted operations and reject arbitrary commands, shell metacharacters, and unvalidated arguments.
  4. Require explicit user confirmation before file modification, command execution, message delivery, or external communication.
  5. Do not recommend generic administrator execution. Diagnose path ownership and permission errors, and request only the minimum required access.
  6. Run the skill in a sandbox under a non-privileged account with network, filesystem, process, and execution restrictions.
  7. Separate message drafting from message delivery and require confirmation immediately before sending.
  8. Document why each tool is required and ensure the host platform enforces the declared restrictions rather than relying only on natural-language instructions.
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 (6)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill description is broad and lacks clear activation triggers or invocation boundaries, making it unclear when the agent should use this skill. Over-broad routing can cause the skill to be invoked for unrelated tasks, which is more dangerous here because the skill advertises read/write/exec tooling and operational breadth.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill is presented as a task, communication, and scheduling assistant, but its documented capabilities implicitly broaden into generic file handling, API use, and command execution via declared tools and operational guidance. This capability mismatch can cause an agent or user to authorize the skill in low-risk contexts while it actually has higher-impact system access, increasing the chance of unsafe execution or data exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation indicates the skill can send messages and, elsewhere in the file, manage files, execute commands, and call APIs, but it does not provide clear user-facing warnings or consent boundaries for actions with real-world or system impact. This creates a risk of unintended outbound communication, file modification, or command execution under the guise of routine assistant behavior.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The request/output contract describes a grading-style JSON schema that does not match the stated assistant purpose of scheduling, task management, and communications. This inconsistency makes the skill's true behavior unpredictable, which can lead orchestrators or users to mis-handle outputs, bypass expected validation, or invoke the skill in inappropriate contexts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The skill presents a Chinese display name and substantial Chinese-language documentation while also embedding English text, but does not state whether language selection is user-configurable. This can amount to a language-policy issue if the skill defaults users into a locale or language without explicit opt-in.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file claims command execution is limited to a whitelist, but the documentation provides no actual enforcement mechanism, policy definition, or validation logic. In a skill that declares exec capability, unenforced security claims create a false sense of safety and can permit arbitrary command execution if the surrounding agent trusts the documentation.

Content

No source excerpt is available for this finding.