Back to skill

Security audit

Auto Monitor

Security checks for vulnerabilities and agentic risk

Overview

This monitoring skill is small and non-executable, but it tells the agent to perform proactive monitoring and immediate automatic repairs without clear user approval or limits.

Install only if you intend the agent to monitor the relevant server context, and require explicit approval before any repair action. Treat it as suitable for reporting and recommendations unless you add a narrow allowlist, logging, rollback guidance, and clear enable/disable controls.

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:43
Finding

Unbounded Autonomous Remediation Without User Approval

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 43
Vulnerability Type: T05: Unauthorized Access and Privilege Escalation
Risk Level: Medium

Complete relevant snippet translated into English:

markdown
## Proactivity

- Do not wait for the user to ask, "How is the server?"
- Proactively report discovered problems
- Immediately handle anything that can be automatically repaired

Technical Analysis

The final instruction authorizes the agent to perform unspecified automatic remediation immediately after detecting a problem. It does not define an allowlist of permitted operations, require user approval, restrict the privileges under which remediation may run, identify protected resources, or require rollback and post-action validation.

Monitoring disk, memory, CPU, network, and process state is consistent with the declared purpose. Unconditionally modifying the system is not necessary for monitoring and reporting, however. The broad phrase "anything that can be automatically repaired" could be interpreted as permission to terminate processes, delete files, change configuration, restart services, modify network settings, or perform other state-changing operations.

The package contains no executable implementation or explicit privilege-escalation mechanism. Consequently, this instruction cannot independently obtain operating-system privileges beyond those already available to the hosting agent. The risk arises because it encourages the agent to use its existing tools and permissions beyond a clearly bounded monitoring role.

Attack Path

  1. The skill periodically examines system health indicators.
  2. A normal transient condition, manipulated metric, or genuine fault exceeds one of the documented thresholds.
  3. The agent classifies the condition as something that can be repaired automatically.
  4. Based on the line 43 instruction, the agent selects a state-changing remediation without requ ...[truncated 974 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the unrestricted automatic-repair instruction with diagnosis, reporting, and remediation recommendations only.
  2. Require explicit user approval before every state-changing operation.
  3. If automation is required, define a narrow allowlist of reversible actions and prohibit commands outside that allowlist.
  4. Specify resource boundaries, including the exact services, processes, directories, and configuration keys that may be changed.
  5. Run monitoring under a dedicated least-privilege account without administrative access.
  6. Separate read-only monitoring permissions from remediation permissions, granting the latter only for an approved action.
  7. Require a dry run, pre-change state capture, timeout, rollback procedure, and post-action health validation.
  8. Log the detected condition, selected action, authorization decision, command or tool invocation, result, and rollback status.
  9. Add safeguards against destructive operations, including file deletion, process termination, firewall modification, package installation, and service reconfiguration.
  10. A safer replacement instruction would be: “Detect and report anomalies with recommended remediation. Do not modify system state unless the user explicitly approves a specific, bounded action.”
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill states that it may "automatically repair immediately" but does not warn the user that it performs ongoing monitoring or autonomous remediation. This is dangerous because self-initiated changes to processes, services, or system configuration can disrupt production systems, cause data loss, or exceed the operator's intended authority without informed consent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill is framed as proactively monitoring and reporting "without waiting for instructions," but it does not define explicit activation boundaries, authorization requirements, or the conditions under which monitoring begins. In an agent setting, this ambiguity can cause unauthorized background behavior, unexpected access to system state, and user surprise about continuous monitoring or autonomous action.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

All user-facing descriptive text in the skill file is presented in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-language context. Under the policy, a fixed language requirement without opt-in or justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest description says the skill proactively monitors system status but does not define triggers, scope, cadence, or authorization boundaries. For an autonomous monitoring skill, that ambiguity is security-relevant because it can enable unexpected background execution, overbroad data access, or user surprise about when monitoring starts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The only natural-language description in the manifest is written in Chinese, with no indication that users can choose another language or that the skill is intentionally region-specific. This can be a language/locale policy issue because it implicitly fixes the skill metadata to one language without opt-in or justification.

Content

No source excerpt is available for this finding.