Back to skill

Security audit

Lemon8数据助手

Security checks for vulnerabilities and agentic risk

Overview

The scanners report a Lemon8 analysis skill with no AIG findings and several concerns about consent, privacy disclosure, generic triggers, installation, and local audit writes; the assigned verdict is benign, with medium confidence.

The verdict is benign, not risk-free. Before installing, consider the reported persistent local compliance and audit writes, the lack of a privacy disclosure for sending share links to a third-party API, and the unpinned installer. Skillspector inspected all 16 listed components but marked analysis partial because some references were missing or ambiguous, so evidence is incomplete; AIG completed with no findings.

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
  • 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 (10)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Ordinary local logging could be a supporting implementation detail. However, this code goes beyond logging API activity: it unconditionally asserts that legal confirmation occurred and updates persistent confirmation state. That consent-related behavior is not represented by the declared public-data analysis purpose or its default read-only operation.

Content

No source excerpt is available for this finding.

The consent gate requires Chinese confirmation wording and rejects affirmative responses in other languages without offering a language choice.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The accepted confirmation phrases are exclusively Chinese, while English affirmations such as 'yes' are explicitly rejected. Because confirmation is mandatory to proceed, this imposes a language requirement without user opt-in; the legal context does not explain why consent must be expressed in Chinese.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding

The installation command invokes an unpinned npm CLI through npx, potentially downloading and executing whichever version resolves at installation time. A compromised future package release could execute with the installing user's permissions and access local credentials; this is a conditional supply-chain exposure, not evidence of an existing compromise or an MCP server.

Content

No source excerpt is available for this finding.

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

Medium
Category
Output Handling
Confidence
97% confidence
Finding

All accepted confirmation phrases are Chinese, while the English affirmative “yes” is explicitly rejected. This forces users to use Chinese to proceed, without an opt-in or a documented justification for the language restriction.

Content

No source excerpt is available for this finding.

The share-link workflow sends user-provided links to a third-party API without a privacy disclosure.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow directs the agent to submit a user-provided Lemon8 sharing link as share_text to the API hosted at www.aconfig.cn (L003), rather than processing it locally or sending it directly to Lemon8. The file does not warn users that their supplied links will be transmitted to this third-party service; listing the base URL alone does not explain that privacy implication.

Content

No source excerpt is available for this finding.

The trigger “看看评论” (“look at the comments”) does not require Lemon8-specific context.

Medium
Category
Trigger Abuse
Confidence
90% confidence
Finding

This phrase can refer to comments on other platforms, product reviews, or code, yet it is listed as an activation condition for the post-detail-and-comments recipe. The trigger does not explicitly require an established Lemon8 context or provide exclusions, so unrelated requests could activate this workflow.

Content

No source excerpt is available for this finding.

Confirmation rules require Chinese responses without offering a language choice.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

Every accepted confirmation is Chinese, while the explicit English affirmation 'yes' is rejected. This imposes a language constraint on user consent without opt-in or an explanation of why confirmation must be in Chinese; citing regional laws elsewhere does not establish that requirement.

Content

No source excerpt is available for this finding.

The explicit no-write constraint contradicts mandatory persistent compliance-state and audit-log writes.

Low
Category
Excessive Agency
Confidence
91% confidence
Finding

The core constraint at L245 states that the skill does not perform writes, but L312–L315 create a directory, overwrite a confirmation file, and append an audit log under the user's home directory. These are required persistent side effects, and L412–L419 additionally require a filesystem write after every successful API call.

Content

No source excerpt is available for this finding.

Generic triggers can invoke Lemon8 recipes for requests unrelated to Lemon8.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

Triggers such as “看看评论” (look at the comments), “用户链接” (user link), and “看看主页” (look at the homepage/profile) do not identify Lemon8 and can occur in requests about other sites. The longest-match and tie-breaking rules resolve competing recipe matches but do not require a Lemon8 context before activation.

Content

No source excerpt is available for this finding.

Generic keywords can trigger compliance reminders for unrelated requests.

Low
Category
Trigger Abuse
Confidence
93% confidence
Finding

The batch-operation triggers include generic terms such as 全部 ('all') and 循环 ('loop'), while the personal-data triggers include 笔记 ('notes'). Even within Lemon8, these terms do not establish bulk data collection or personal-information access, and no contextual matching constraints or exclusions are specified.

Content

No source excerpt is available for this finding.