Back to skill

Security audit

知乎数据助手

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report undeclared installation and upgrade actions, records asserting legal confirmation, and overly broad search triggers alongside its stated read-only Zhihu analysis purpose.

Consider the Review recommendation before installing. The findings point to actions beyond the stated read-only purpose and leave some evidence incomplete; the available results do not establish malicious intent. If you proceed, review the install and upgrade steps and the handling of local compliance records and questions sent to the external service.

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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (16)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The maintenance table recommends ordinary and forced skill upgrades in response to API failures without specifying separate user approval or a reviewed target revision. This can turn a remote service failure into a local software and instruction-trust change, extending beyond the user's data-query authorization.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The mismatch is not merely incidental local logging. The code unconditionally asserts that legal terms were confirmed and persists that assertion in both a status file and an audit log. This is a substantive consent-management action not disclosed by the description of read-only Zhihu data analysis. The supplied chunk contains no interaction or check establishing actual confirmation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Ordinary operational logging could be a supporting implementation detail, but this code specifically records a completed legal review without performing or verifying one. That compliance assertion is distinct from the declared Zhihu data-analysis purpose. The snippet does not make network requests or demonstrate data exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The only supplied executable behavior is npx skillhub install zhihu-pro, which performs an installation operation not described in the declared purpose. The installed skill may provide the advertised functionality, but its implementation is not supplied and cannot be verified. No specific downstream behavior or malicious activity can be inferred from this command alone.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Local logging alone would be a supporting implementation detail, not a mismatch. However, this code also establishes a substantive legal-confirmation state and records confirmation as completed. That behavior is not disclosed by the Zhihu analytics description. The supplied chunk performs no Zhihu queries or network requests.

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 unpinned npm package through npx, allowing the installer code to change between executions. A compromised release could execute with the invoking user's permissions; this is a package supply-chain issue, although the scanner incorrectly describes the command as an MCP server.

Content

No source excerpt is available for this finding.

The mandatory consent gate accepts only Chinese confirmations and explicitly rejects English confirmation.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The confirmation rules accept only specified Chinese phrases, explicitly reject “yes,” and require the user to use wording such as “同意” before proceeding. This imposes a language requirement without user opt-in or a documented justification for restricting consent to Chinese.

Content

No source excerpt is available for this finding.

The read-only analytics skill recommends upgrading its installed code in response to API errors.

Medium
Category
Excessive Agency
Confidence
92% confidence
Finding

The documented recovery path couples 404/410 responses to upgrading the installed skill, including a forced upgrade for repeated failures. An attacker able to influence service responses could induce an unnecessary update attempt, and an unreviewed replacement skill could subsequently change agent behavior.

Content

No source excerpt is available for this finding.

The AI-search workflow sends the user's question to a third-party service without a privacy disclosure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The chain recipe uses the AI-search endpoint with a user-provided natural-language question. The endpoint is hosted at www.aconfig.cn (L03), and message_content contains the user's question (L57), but the file does not warn that this content will leave the assistant's environment and be shared with an external service.

Content

No source excerpt is available for this finding.

The everyday phrase “找人” (“find someone”) can activate a platform-specific user-search recipe without a clear scope restriction.

Medium
Category
Trigger Abuse
Confidence
96% confidence
Finding

The routing rules can select Zhihu user search from the generic phrase 'find someone' without first establishing that the user wants a Zhihu search. If acted upon automatically, this can transmit a person's name or other search terms to an unintended third-party service.

Content

No source excerpt is available for this finding.

The chain graph directs article identifiers into question recipes, contrary to the manifest's explicit requirement to avoid mixing ID types.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The graph treats generic id fields from article results as valid inputs to question-oriented recipes and explicitly encourages direct parameter forwarding. Because numeric identifier namespaces can overlap, this can silently retrieve unrelated records and contaminate analysis or attribution, although it does not demonstrate an access-control bypass.

Content

No source excerpt is available for this finding.

The trigger “找人” (“find someone”) is overly broad and does not constrain activation to a platform-specific user search.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The recipe associates the unqualified phrase 'find someone' with an external user search and profile lookup, without a platform-scope precondition. Automatic activation could disclose personal search terms to MaxHub when the user intended a different service or merely wanted advice.

Content

No source excerpt is available for this finding.

The whitelist authorizes querying the MaxHub account balance, which is unrelated to public Zhihu content and user analytics.

Low
Category
Excessive Agency
Confidence
95% confidence
Finding

L307–L313 allow GET /api/maxhub/balance through the skill's endpoint whitelist. This exposes an account-level billing capability rather than a Zhihu research capability, and the manifest does not declare account or billing management as part of the skill's purpose.

Content

No source excerpt is available for this finding.

The standalone trigger “社交圈” (“social circle”) can activate the recipe for unrelated social-network discussions.

Low
Category
Trigger Abuse
Confidence
87% confidence
Finding

“社交圈” is an everyday expression that can refer to offline relationships or networks on other platforms. The activation condition does not require a Zhihu-specific request, so it may select a user-profile and follower-fetching workflow outside its intended context.

Content

No source excerpt is available for this finding.

Batch-operation reminders can be triggered by everyday words unrelated to bulk collection.

Low
Category
Trigger Abuse
Confidence
95% confidence
Finding

The batch-operation keyword list includes broad terms such as “全部” (all), “所有” (every/all), and “循环” (loop). Although the category description mentions bulk data collection, the matching rules provide no contextual requirement or exclusions to prevent these ordinary words from triggering an unrelated compliance confirmation.

Content

No source excerpt is available for this finding.

Confirmation rules require Chinese wording without offering a language choice.

Low
Category
Output Handling
Confidence
92% confidence
Finding

Every accepted confirmation is Chinese, while the English affirmative “yes” is explicitly rejected. This requires users to confirm in Chinese without an opt-in or a documented justification for restricting the confirmation language; the platform-specific compliance context does not explain that restriction.

Content

No source excerpt is available for this finding.