Back to skill

Security audit

Maxhub Youtube

Security checks for vulnerabilities and agentic risk

Overview

The scanners report a YouTube data-query skill with no AIG findings, and the verdict is benign because the noted issues point to installation, scope, and privacy concerns rather than clear malicious behavior.

The verdict is benign with medium confidence, not a guarantee of zero risk. Before installing, consider the unpinned installer, generic activation examples, and the possibility that query terms and target identifiers are sent to MaxHub; scanner coverage was partial because some referenced local paths could not be resolved.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
Findings (8)

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · references/api-video.md (reported line 188)May include surrounding context.

l": "...", "width": 168, "height": 94}, {"url": "...", "width": 336, "height": 188} ], "moving_thumbnail": "https://i.ytimg.com/an_webp/zd3yCa1bJCM/mqdefault_6s.webp...", "duration": "16:57", "duration_accessibility": "16分钟57秒钟", "view_count": "343,369次观看", "short_view_count": "34万次观看", "published_time": "18小时前", "description": "Today, we're trapped in a super weird dream...", "is_live": false, "is_verified": true, "url": "https://www.youtube.com/watch?v=zd3yCa1bJCM", "playback_url": "https://rr5---sn-ogueln67.googlevideo.com/initplayback..." } ], "continuation_token": "下一页token" }

text
 ### 清洗后的字段说明:
- `video_id`: 视频ID
- `title`: 视频标题
- `thumbnail`: 最高清晰度缩略图URL
- `thumbnails`: 所有分辨率的缩略图列表
- `moving_thumbnail`: 动态缩略图URL(webp格式,鼠标悬停预览)
- `duration`: 视频时长

MCP server referenced without pinned version: 'npx clawhub'.

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The installation command can download and execute an unpinned version of the clawhub npm package. A compromised future release could execute code with the installing user's privileges; this is an installation supply-chain risk, although clawhub is a CLI rather than an MCP server.

Content

No source excerpt is available for this finding.

Example invocation prompts include generic words without YouTube-specific scope.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The example prompts include standalone terms such as “详情” (details), “信息” (information), “搜索” (search), and “回复” (reply), which commonly occur in unrelated requests. Although the README identifies this as a YouTube skill, these invocation examples do not require YouTube context or exclude unrelated uses, making unintended activation plausible.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx clawhub'.

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The installation command leaves the executable clawhub npm package unpinned, allowing its resolved code to change between installations. If a subsequently resolved release is compromised, installation could execute attacker-controlled code with the user's privileges; the MCP-server label is inaccurate but does not eliminate the supply-chain risk.

Content

No source excerpt is available for this finding.

The example invocation phrases are generic words without a YouTube-specific activation constraint.

Medium
Category
Trigger Abuse
Confidence
93% confidence
Finding

The example commands include “搜索,搜” (search) and “评论,回复” (comments, reply), which also occur in ordinary requests unrelated to YouTube. Although the introduction identifies the skill as a YouTube assistant, the invocation examples do not require YouTube context or provide exclusion conditions, leaving activation scope ambiguous.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 53)May include surrounding context.

md
maxhub_auth_header="Authorization: Bearer ${MAXHUB_API_KEY}"

# GET example
curl -s "https://www.aconfig.cn/api/v1/youtube/{endpoint}?%7Bparams=}" \
  -H "$maxhub_auth_header"

# POST example

The autonomous query workflow lacks a privacy warning about sending user-supplied search terms and target identifiers to MaxHub.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The workflow instructs the agent to search for channels and retrieve video data autonomously through the third-party MaxHub API. Although the provider is named, the skill does not warn that user-supplied query terms and channel/video identifiers will be disclosed to that provider; the existing disclaimer addresses data reliability rather than privacy.

Content

No source excerpt is available for this finding.

The fallback instructions include Douyin API calls despite the manifest limiting the skill to YouTube data queries.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest at L003 describes a YouTube data-query assistant, but the operational fallback table at L282–L285 directs requests to /api/v1/douyin/ endpoints for videos, search, and user profiles. These are operations on a different platform, not alternate versions of the declared YouTube functionality.

Content

No source excerpt is available for this finding.