70High
The skill is vulnerable to prompt injection and can cause
Behavior Analysis
Description verified — behavior matches claim
Claims to do
吞贼·净化魄 (Tun Zei - Purifier): > **七魄之四·吞贼** > 职掌：错误修正、冗余清理、容错自愈
Actually does
This skill defines capabilities to process error objects for fixing, specify cleanup targets (temp, cache, logs) for data removal, and perform health checks. It integrates with other skills for video processing error correction and Douyin video publishing exception handling. No direct external tools, commands, or URLs are specified within this skill's definition.
Skill Info
Namelt8899789/tun-zei
Registryclawhub
Version1.0.5
PURLpkg:clawhub/lt8899789/tun-zei@1.0.5
Downloads90
Source
SHA-256384aeb3e0221...
SHA-1ad4c4234f78b...
MD5b648d987bd80...
TLSHad31731a9e27...
Size1,862 bytes
Install
This skill has been flagged as potentially malicious. Review the findings below before installing.
ClawHubDocs
openclaw skills install lt8899789/tun-zeiAssessments (3)
AI Agent Traps ↗Prompt Injection via `fix` capability's `error.stack`highprompt injection
The `fix` capability accepts an `error` object, including a `stack` field, and outputs a `solution` description. An attacker could inject malicious instructions or data into the `error.stack` field, influencing the agent's reasoning to generate a `solution` that, if acted upon by the agent or another component, could lead to unintended or malicious actions.
80% confidenceline 30
error: type: 错误类型 message: 错误消息 stack: 堆栈信息 ... fix: attempted: true/false success: true/false solution: 解决方案描述
Resource Abuse via `cleanup` capability's `target: all`mediumresource abuse
The `cleanup` capability allows deleting data from `temp`, `cache`, `logs`, or `all` system areas. If an attacker can repeatedly trigger the `cleanup` action with `target: all`, it could lead to denial of service or data loss by indiscriminately removing critical system or application files.
80% confidenceline 44
target: temp|cache|logs|all ... cleanup: freedSpace: 释放空间(MB) filesRemoved: 删除文件数 duration: 耗时(ms)
Prompt Injection Detectedmediumprompt injection
DeBERTa classifier detected prompt injection (confidence: 0.62)
62% confidence
Potential Cognitive State Manipulation via `self-improving`lowmemory poisoning
The skill aggregates a `self-improving` capability. Without further details on its mechanism, this could potentially be a vector for an attacker to influence the agent's learned policies or knowledge base if malicious input is incorporated into its self-improvement process.
50% confidenceline 101
| `self-improving` | 调用 | 自我改进与错误修复 |
Badge
Markdown
[](https://mondoo.com/ai-agent-security/skills/clawhub/lt8899789/tun-zei)HTML
<a href="https://mondoo.com/ai-agent-security/skills/clawhub/lt8899789/tun-zei"><img src="https://mondoo.com/ai-agent-security/api/badge/clawhub/lt8899789/tun-zei.svg" alt="Mondoo Skill Check" /></a>Image URL
https://mondoo.com/ai-agent-security/api/badge/clawhub/lt8899789/tun-zei.svgSecure your AI agents
Skills can read files, run commands, and access credentials. Mondoo helps organizations manage the security risks of AI agent skills across their entire fleet.
- Continuous scanning of skills across all registries
- Policy enforcement before skills reach your agents
- Integration with your existing security workflow