Back to skill

Security audit

抖音数据助手

Security checks for vulnerabilities and agentic risk

Overview

ClawScan labels this skill benign with medium confidence, while noting concerns about automatic compliance records, an unpinned installer, credential and data disclosures, and a messaging capability; some references were not fully resolved, so the evidence is incomplete.

The overall verdict is benign, not risk-free. Before installing, consider the unpinned installer, the reported automatic compliance records, and workflows that may pass cookies or user data to services without clear disclosure. The scanner completed checks across the listed components but marked the analysis partial because some local references were missing or ambiguous, so the available evidence does not fully resolve those areas.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (11)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code performs no Douyin data retrieval or analysis. Although local logging can be a supporting implementation detail, these writes specifically assert that legal confirmation occurred, rather than merely logging API activity. Automatically establishing that confirmation state is an undeclared behavior materially distinct from the advertised data-query functionality. No network requests are present, so the stated request destination is not contradicted by this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The only supplied command is an installer, not an implementation of the described data-query and analysis functionality. Running it may retrieve and execute the skillhub package and install another skill. The installed skill's behavior and network destinations are not shown, so its conformity to the description cannot be verified. This assessment is limited to the supplied command and does not imply malicious behavior.

Content

No source excerpt is available for this finding.

YARA rule 'exploit_framework': Exploit framework components and payloads [hacktools]

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/atoms/_index.md (reported line 9)May include surrounding context.

md
Index / 原子索引


> 以下原子被分类为高风险,**默认 agent 不应主动调用**。
> 仅在用户明确授权 + 一次性确认后才能执行。具体策略见 SKILL.md 的「高风险能力清单」章节。

| atom_id | 风险类别 | 限制原因 |
|---------|---------|---------|
| 🚨 open_app_message | private_messaging | triggers private-message UI; abuse vector for spam/social-engineering |
| 🚨 extract_sec_uid_batch | bulk_extraction | bulk identifier extractor; scale-collection vector |
| 🚨 extract_aweme_id_batch | bulk_extraction | bulk identifier extractor |
| 🚨 extract_webcast_id_batch | bulk_extraction | bulk identifier extractor |
| 🚨 query_user | query_helper | high-risk write_op marked user-search helper |

---

> 本文件是原子化层(Atomic Layer)的索引。每个端点封装为一个"原子",具备标准化输入/输出/幂等性/链路角色声明。
> Agent 可按 atom_id 快速定位端点,无需全文读取 reference 详

Authenticated analytics workflows pass session cookies to tools without warning about credential sensitivity.

High
Category
Output Handling
Confidence
97% confidence
Finding

L009–L012 instruct the agent to accept a cookie and pass it into analytics tools; the same pattern recurs throughout L020–L056. The file provides no warning that session cookies are sensitive credentials or that these operations access authenticated account analytics.

Content

No source excerpt is available for this finding.

Cookie-based workflows instruct the agent to pass credentials to API endpoints without warning the user.

High
Category
Output Handling
Confidence
96% confidence
Finding

L43 and L52 explicitly direct the agent to supply a cookie when calling Web and App V3 endpoints, rather than merely listing available tools. These workflows contain no warning that cookies may represent authentication credentials or that transmitting them can expose account access.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The installation command executes an unpinned npm package, allowing the code retrieved by a future installation to change without review. A compromised package release could execute with the installing user's privileges and access local files or credentials; this is a supply-chain risk, although the command is an installer rather than an MCP server.

Content

No source excerpt is available for this finding.

The index advertises a private-message UI capability outside the skill's stated data-query and analytics purpose.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

At L009, open_app_message is described as triggering a private-message UI, whereas the manifest describes Douyin data retrieval and content analysis. The authorization requirement at L004–L005 limits its use but does not justify this messaging capability within the stated purpose; this source does not establish that messages are actually sent.

Content

No source excerpt is available for this finding.

Generic keywords can activate specialized recipes without confirming the intended platform or task.

Medium
Category
Trigger Abuse
Confidence
96% confidence
Finding

Triggers such as “高清” (high definition), “合集” (collection), and “频道” (channel) can occur in unrelated requests, yet select specific video-retrieval workflows. The longest-match and tie-breaking rules at L004–L007 resolve competing matches but do not establish whether a recipe should activate at all.

Content

No source excerpt is available for this finding.

Visual search submits a user-provided image URI without a privacy disclosure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

This workflow instructs the agent to pass image_uri to search_vision, rather than merely listing an available tool. It does not warn that the search service receives the image reference, which could contain a private location or access-bearing URL.

Content

No source excerpt is available for this finding.

The recipes instruct API calls with user-supplied identifiers and keywords without disclosing transmission to an external service.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The atomic steps direct the agent to submit uid, sec_user_id, and search keywords to lookup and search endpoints, rather than merely listing available tools. This file contains no warning explaining that these inputs leave the user's environment or identifying the service receiving them.

Content

No source excerpt is available for this finding.

Confirmation rules require Chinese wording and explicitly reject an English affirmative without offering a language choice.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

All accepted confirmation phrases are Chinese, while English 'yes' is explicitly rejected and the clarification prompt directs users toward Chinese wording. Requiring explicit informed consent is reasonable, but restricting that consent to one language is not justified here and no language opt-in is offered.

Content

No source excerpt is available for this finding.