Back to skill

Security audit

快手数据助手

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report undeclared compliance-record writes, overly broad triggers, and an unpinned installer; the review is based on incomplete evidence.

Consider avoiding installation until the undeclared compliance-record behavior and broad triggers are addressed, the installer version is pinned, and the handling of links and identifiers is made clear. The scan covered the listed components but was marked partial because some local references could not be resolved, so the evidence is incomplete.

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
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (15)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The mismatch is not merely incidental logging or caching: the code unconditionally asserts that legal version 1.0.0 was confirmed and records a first_time_confirmation event with status confirmed. Creating this compliance state is not disclosed by the public-data-query description, and the supplied code contains no consent check or interaction supporting that assertion.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

Ordinary operational logging could be a supporting implementation detail, but this code specifically writes a completed legal-review assertion. That compliance-state change is not disclosed by the public-data querying description and is materially different from logging an API request or analysis operation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

Ordinary operational logging could be a supporting implementation detail, but this code specifically asserts a completed legal review without performing or validating one. That compliance-state assertion is unrelated to the declared public-data querying and analysis purpose. The write succeeds only if the destination directory exists and is writable.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The installation command invokes an npm executable without pinning its version, so a fresh installation can execute a changed or compromised release with the user's privileges. This is a supply-chain exposure even though the referenced tool is an installer rather than an MCP server.

Content

No source excerpt is available for this finding.

The consent gate requires Chinese-language confirmation without offering a language choice.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The acceptance rules allow only specified Chinese confirmations and explicitly reject English responses such as 'yes', directing users to enter Chinese wording instead. This imposes a language requirement without user opt-in or a documented justification, despite the manifest listing English as a supported locale.

Content

No source excerpt is available for this finding.

L295–L300 authorize querying the MaxHub account balance, which is outside the stated public Kuaishou analytics purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The whitelist permits GET /api/maxhub/balance, an account-balance query rather than a public Kuaishou content or analytics query. Access to API-provider account financial information is not declared in the manifest and is not necessary to perform the described content-analysis operations.

Content

No source excerpt is available for this finding.

The generic trigger “打开链接” (“open link”) can invoke a video-specific recipe for unrelated links.

Medium
Category
Trigger Abuse
Confidence
96% confidence
Finding

The generic phrase 'open link' selects a Kuaishou video recipe without requiring a Kuaishou-specific intent or URL check. In this third-party API skill, that can route an unrelated URL, potentially containing private identifiers or access tokens, to MaxHub even though the user only intended to open it.

Content

No source excerpt is available for this finding.

The triggers “热门作品” and “热门视频” do not distinguish general trending-content requests from requests for a particular user's popular posts.

Medium
Category
Trigger Abuse
Confidence
90% confidence
Finding

These phrases can describe platform-wide popular content, but this recipe fetches a specific user's profile and popular posts. Although user_id is a required input, the activation description does not require user-specific intent or exclude general trending-content requests.

Content

No source excerpt is available for this finding.

The trigger 「推荐」 ('recommend') can activate the video-feed workflow for unrelated recommendation requests.

Medium
Category
Trigger Abuse
Confidence
97% confidence
Finding

「推荐」 is common in requests for products, books, restaurants, and other recommendations. This recipe requires no input and does not explicitly restrict activation to requests for a Kuaishou video feed, so the trigger can select an unrelated workflow.

Content

No source excerpt is available for this finding.

The lookup workflows send user-provided share links and identifiers to a third-party API without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

These chain recipes direct lookups using user-provided share links, usernames, and identifiers against the aconfig.cn API identified at L003, rather than directly against Kuaishou. Although the API host is documented, the file does not warn users that their lookup inputs will be disclosed to this third-party provider.

Content

No source excerpt is available for this finding.

The URL-resolution workflow sends user-provided share text to a third-party API without a privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

L125–L135 direct the agent to use the user's Kuaishou link or share text as the share_text parameter. The configured API host is www.aconfig.cn (L003), but the file does not warn users that their supplied content will be transmitted to this third-party service rather than directly to Kuaishou.

Content

No source excerpt is available for this finding.

The confirmation rules require Chinese wording without offering a language choice.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

Every accepted confirmation is Chinese, while English confirmations such as 'yes' are explicitly rejected and the fallback prompt directs users to enter Chinese wording. This imposes a language requirement on consent without user opt-in or a documented reason why confirmation must be in Chinese.

Content

No source excerpt is available for this finding.

L050 mandates a fixed Chinese response without offering a language choice.

Low
Category
Output Handling
Confidence
88% confidence
Finding

For unsupported requests, the instruction tells the agent to reply directly with a specific Chinese sentence. Unlike an illustrative example, this prescribes user-facing output without allowing adaptation to the user's language or documenting a justified locale constraint.

Content

No source excerpt is available for this finding.

The trigger 「作者」 ('author') is too broad for a video-author lookup.

Low
Category
Trigger Abuse
Confidence
94% confidence
Finding

The recipe treats a generic reference to an author as an activation signal, although the workflow specifically retrieves a video's creator profile. This could match requests about book authors or other unrelated creators; no explicit activation restriction requires an existing video context.

Content

No source excerpt is available for this finding.

Batch-operation triggers include ordinary words that do not establish an intent to collect data in bulk.

Low
Category
Trigger Abuse
Confidence
93% confidence
Finding

The keyword list includes 全部 ('all'), 所有 ('every/all'), and 循环 ('loop'), which can occur in routine requests unrelated to bulk data collection. Although the category description mentions bulk collection, the matching rules do not explicitly require that context or provide exclusions, allowing unnecessary compliance prompts.

Content

No source excerpt is available for this finding.