Back to skill

Security audit

OpenLX 携程酒店运营助手

Security checks for vulnerabilities and agentic risk

Overview

This skill has high-impact hotel account automation, but the artifacts describe it openly and include meaningful user authorization, account-mapping, and readback controls.

Install only if you are comfortable giving this skill access to a dedicated hotel workspace and an isolated Chrome session for the relevant Ctrip merchant account. Keep the workspace private because it may contain reports, evidence captures, action logs, browser session state, license data, and business information. Enable automatic pricing, positive-review replies, or content submission only after checking the exact policy, hashes, account mapping, and rollback/readback plan.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (29)

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · references/acceptance-T01-T46.csv (reported line 1)May include surrounding context.

text
test_id,scenario,expected,capability_ids,mock_status,live_status,evidence_ids,note
T01,全新环境安装,得到可运行技能及真实依赖检查,不是仅复制文件,INSTALL;RELEASE,NOT_RUN,LOCAL_INSTALL_PASS_MAC_LINUX,,28项回归是局部场景覆盖,不等于T01-T46全部通过;真实账户尚未验证
T02,Chrome首次登录与重启,有效会话可复用;失效可人工续登,LOGIN_EBOOKING;LOGIN_CONTENT,NOT_RUN,NOT_RUN,,28项回归是局部场景覆盖,不等于T01-T46全部通过;真实账户尚未验证
T03,日常默认Profile隔离,不改动用户日常Chrome配置,LOGIN_EBOOKING,NOT_RUN,NOT_RUN,,28项回归是局部场景覆盖,不等于T01-T46全部通过;真实账户尚未验证

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README explicitly encourages importing real store data and verifying live accounts, but it does not clearly disclose how sensitive business data, customer-related information, credentials, session state, or generated reports are handled, stored, or protected. In a tool that reuses Chrome login state and performs authorized actions, this omission can lead users to expose account access or private operational data without understanding retention, local storage, report leakage, or misuse risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The description is entirely framed around executing hotel-operations tasks for Ctrip merchants in Chinese, including review replies and content production, with no indication that users may choose another language. This creates a natural-language locale constraint that appears mandatory rather than optional.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
78% confidence
Finding

The skill states that persona files include '语言风格' and that content generation reads this configuration, but it does not say the user can opt into or override the language/locale used. This can amount to enforcing a particular language or linguistic style through configuration without clear user choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The capability title explicitly specifies '中文离线完整免费HTML及打印报告', which indicates the skill is designed to produce reports in Chinese. In this manifest-style file there is no accompanying indication that users can choose another language or that the locale restriction is optional or region-justified, so this reads as a forced language policy choice.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest describes a hotel operations skill for audits, reviews, pricing/inventory/orders, competitor analysis, content production, offline reporting, and authorized browser actions. This file additionally tracks first-class LICENSE and PAYMENT capabilities, including refund handling, device migration, real payment callbacks, and entitlement snapshots, which are product-commerce/platform-management functions rather than obvious hotel-ops functionality.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The catalog entry presents the skill name entirely in Chinese, and the manifest provides no indication that users can choose another language or that the skill is restricted to a China-specific audience for compliance or regional reasons. Under the policy, locale constraints should be opt-in or clearly justified when embedded in natural-language metadata.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This function writes draft content to disk with writeJson(file,draft) and may automatically submit, read back, and publish content via operate(...), but there is no confirmation prompt, print/log disclosure, or inline warning to the user at the point of execution. Because these actions can affect user data and external systems, the absence of any user-visible disclosure in the code is a quality/safety concern.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The content workflow can autonomously submit content to an external platform when auto_submit is enabled and a broad policy matches topic/facts/assets, without a per-post human approval step. In a hotel-operations skill connected to real accounts, this increases the risk of unintended publication, policy drift, or reputational damage if the generated draft is inaccurate, stale, or contextually inappropriate despite passing hash-based checks.

Content

Scanner excerpt · scripts/automation.mjs (reported line 54)May include surrounding context.

js
const factsHash=hash((s.facts||[]).filter(f=>f.verified===true));
  const policyValid=p?.hotel_id===s.hotel.id&&p?.account_id&&p.facts_hash===factsHash&&p.persona_hash===(persona?hash(persona):null)&&Array.isArray(p.allowed_asset_hashes)&&draft.assets.every(a=>p.allowed_asset_hashes.includes(a.hash))&&p.authorized_topics?.includes(topic)&&Date.parse(p.valid_until)>Date.now()&&Number.isFinite(Date.parse(p.valid_until));
  // Broad topic/fact/asset authorization is sufficient for unchanged rule content.
  // New model prose remains a reviewable draft, never silently auto-approved.
  if(c.auto_submit===true&&policyValid&&draft.mode==='RULE_FACTS_ONLY'&&s.source.type!=='MOCK'){
    const mapping=path.join(w.base,'adapter-content.json');
    if(!fs.existsSync(mapping)||JSON.parse(fs.readFileSync(mapping)).account_id!==p.account_id)return {module:'content',status:'DRAFT_READY',file,submission:'ACCOUNT_MAPPING_REQUIRED'};

Ssd 3

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The capture command reads the entire page body text and writes it verbatim to a local evidence file. In this skill's context, the pages are authenticated hotel merchant portals and may contain sensitive business data, account identifiers, customer/order details, or other protected content, so indiscriminate plaintext capture increases the risk of local data exposure and over-collection.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code begins enforcing a Chinese-only user experience with hard-coded Chinese error text, and the pattern continues throughout the file in generated replies and HTML (lang="zh-CN"). The file does not provide any user opt-in or locale selection, which is a natural-language locale policy concern under the rule.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The thrown error string '缺少--workspace' is hard-coded in Chinese, which imposes a specific language without any visible user opt-in or documented locale constraint. This is a natural-language policy concern because the skill does not appear to offer language selection or explain why Chinese-only messaging is required.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The install/upgrade path backs up any existing target directory and then replaces it by copying the new source tree into place. This can alter or remove an existing skill installation and potentially affect user data or operational state if the target path is unexpected or the copy fails partway, especially because the operation proceeds without a prominent prior warning or confirmation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The installer automatically runs npm ci and then executes scripts/ops.mjs doctor in the target directory during install/upgrade. Even though this is typical installer behavior, it causes dependency lifecycle scripts and a local diagnostic program to run with the user's privileges without an explicit confirmation or warning in this file, which increases risk if the package, dependencies, or distribution channel are compromised.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The function sends hotel-related input, including verified facts and hotel_name, to a configurable external model endpoint via HTTP(S). Although the code includes useful safeguards such as HTTPS-or-loopback enforcement, allowlisted fields, and redaction, this file does not provide any user-facing notice or explicit consent check before transmitting data off-box. In a skill that handles real store data, silent external transfer can create privacy, compliance, and data-governance risk if users are unaware of where their data is being sent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The system message is written in Chinese and instructs the model to produce a draft for a specific Chinese platform, which effectively constrains language/locale behavior. There is no indication in this file that the user can choose the language or explicitly opt into this locale-specific output.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Several user-visible messages and notes are hard-coded in Chinese, such as the device note and error/help text, with no indication that the skill is region-specific or that users can choose their language. This creates a language/locale policy issue because the skill appears to enforce one language without opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

Multiple command errors and status-related strings throughout the file are hard-coded in Chinese, including initialization and unknown-command messages. Because the file does not offer language selection or justify a Chinese-only locale, these natural-language strings indicate a forced-language behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
76% confidence
Finding

The refresh-license command invokes refreshLicense(w,true), which by name and context appears to contact a remote licensing service. In this file there is no confirmation prompt, user-facing log, or comment/docstring disclosing that a network operation may occur or that device/workspace-related data could be transmitted.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The note string '按已有签名有效期继续;免费报告和数据保留' is a natural-language message presented in a specific language, but this file does not indicate user opt-in, language selection, or a documented region-specific constraint. That can violate language/locale policy by forcing a locale implicitly.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest focuses on merchant data analysis, report generation, and authorized operational actions in Ctrip-related systems. The WEBSITE capability adds official site page deployment, public HTTPS health/readback, and related live-write behavior, which is not described as part of the skill's user-facing purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The manifest describes business-facing hotel operations and reporting behavior, but this file also treats repository release, build artifacts, download publication, and installation verification as product capabilities. Those software release-management functions are implementation lifecycle concerns not conveyed by the manifest description.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

This manifest-style JSON file contains user-facing natural-language values exclusively in Chinese, including names, descriptions, review text, and metric labels. Because the file does not state that it is intended only for a Chinese-language or region-specific context, it may violate the policy against forcing a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

This manifest-style JSON contains user-facing natural-language fields such as titles, labels, open gaps, and publication notes exclusively in Chinese. For an all-file-types language policy check, this can be a locale/language constraint that is not presented as optional or justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The code defaults the content schedule timezone to Asia/Shanghai when none is provided, which imposes a locale-specific behavior without explicit user opt-in. The policy allows locale constraints only when the user is given a choice or when the constraint is clearly justified as region-specific.

Content

No source excerpt is available for this finding.