Back to skill

Security audit

Temp Heartbeat - 临时心跳

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed temporary scheduler, but it asks agents to store and later execute free-form task text without clear safety boundaries or auditability.

Review this carefully before installing. It is not clearly malicious, but it should only be used in an environment where scheduled tasks are limited to inert reminders or tightly approved actions, and where deletion of task files after execution is acceptable. Avoid scheduling tasks that can modify files, access private data, call external services, or perform business-impacting actions unless the host agent rechecks permission at execution time.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:95
Finding

Persistent Delayed Execution of Untrusted Task Instructions

Content
View full analysis
This task is a temporary heartbeat and is automatically deleted after execution ## Task Info - **Created At:** {creation time} - **Execute At:** {execution time} - **Task ID:** {unique ID} - **Status:** pending ## Task Content {task content} ## Execution - [ ] Task executed - [ ] Result recorded - [ ] File deleted ``` The reference implementation interpolates task content directly into that persistent file: ```typescript async function createTempHeartbeat( executeAt: Date, content: string ): Promise { const taskId = `task_${Date.now()}_${Math.random().toString(36).substr(2, 6)}`; const filePath = `memory/temp-heartbeat-${Date.now()}.md`; const taskContent = `# Temp Heartbeat Task > ⚠️ This task is a temporary heartbeat and is automatically deleted after execution ## Task Info - **Created At:** ${new Date().toISOString()} - **Execute At:** ${executeAt.toISOString()} - **Task ID:** ${taskId} - **Status:** pending ## Task Content ${content} ## Execution - [ ] Task executed - [ ] Result recorded - [ ] File deleted `; await writeFile(filePath, taskContent); return { id: taskId, executeAt, content, filePath, status: 'pending', }; } ``` The periodic processing workflow later reads the stored file and passes its contents to an execution routine without showing any safety validation or authorization check: ```typescript async function checkAndExecuteTasks(): Promise { const tempHeartbeatFiles = await glob('memory/temp-heartbeat ...[truncated 3426 chars]
Remediation
View remediation
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 (9)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrase '定时任务' ('scheduled task') is very broad and likely to collide with ordinary user requests about scheduling, causing the skill to activate unintentionally. In an agent environment, overbroad activation can route unrelated user intent into this skill, leading to confusion, unintended task creation, or accidental cancellation operations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill declares /heart cancel {ID} and /heart cancel all as supported management commands and provides full output examples for them, but the technical implementation section only includes functions for parsing time, creating tasks, checking/executing tasks, and listing tasks. There is no shown cancellation function or deletion-by-ID/all logic to back up those documented commands, creating an intent-code mismatch within the file.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill prescribes automatic deletion of task files after execution and on cancellation, but provides no safeguards such as confirmation, recovery, audit retention, or clear warning about irreversible loss. Because task content may contain reminders, operational instructions, or evidence of scheduled actions, silent deletion can destroy user data and erase traceability if a task is triggered or cancelled unexpectedly.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The documentation states that English relative time formats such as 'in 10 minutes' are supported, but the provided parsing logic only applies time arithmetic for patterns whose regex string contains the Chinese units 分钟/小时/天. English matches therefore likely return the current time unchanged or are mishandled, causing tasks to execute immediately or at the wrong time. In a scheduler, inconsistent parsing is security-relevant because time-gated actions may run earlier than the user intended.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The skill includes scheduled checks of GitHub API status, which implies deferred network access, but it does not disclose that network requests may occur later or what data may be sent. In an agent context, deferred execution reduces user visibility and can lead to unexpected outbound traffic, accidental disclosure of repository or environment context, or policy violations in restricted environments.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger word "定时任务" is generic and likely to overlap with many unrelated user requests about scheduled tasks, which can cause the skill to activate unintentionally. In an agent ecosystem, overly broad triggers increase the chance of misrouting user input to this skill, potentially causing unintended task creation or confusing behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill documentation, trigger phrases, examples, and output are presented entirely in Chinese, and no alternative language option or opt-in is described. This may violate language/locale policy if users are not explicitly given a choice or if the regional limitation is not documented.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

Most of the skill instructions, outputs, and operational text are written in Chinese, while only limited English trigger support is mentioned. For a natural-language policy review, this can be a locale/language policy issue when the skill effectively forces one language without explicit opt-in or a documented regional justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The manifest uses Chinese for the display name, description, keywords, and categories, with no indication that users can choose another language or locale. This may conflict with language/locale policy expectations if the skill is presented to a broader audience without explicit opt-in or documented regional scope.

Content

No source excerpt is available for this finding.