Back to skill

Security audit

A Share Guard

Security checks for vulnerabilities and agentic risk

Overview

This finance skill is read-only in intent, but it overstates unfinished risk checks and can return reassuring stock-risk results when key data is missing.

Review this before installing if you need dependable financial-risk analysis. The skill appears intended for read-only stock analysis, but several advertised checks are unfinished and may make missing data look safe. Use pinned installation, avoid adopting the browserless private-IP template as-is, and do not treat its output as investment advice.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:64
Finding

Unpinned Package Execution Through npx

Content
View full analysis
Remediation
View remediation
install xiaodi-a-share-guard ``` - Publish and verify package integrity hashes or signed release artifacts. - Configure installation to use an explicitly trusted package registry. - Use a lockfile where the installation workflow permits it. - Document the expected package version and verify it before execution. - Avoid automatically accepting a newly published package version in production installation procedures. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/scanner.py:111
Finding

Missing Financial Data Is Classified as Zero Risk

Content
View full analysis
30% 数据源:东方财富 F10(需要 browserless 抓取) """ # 简化版:返回模拟数据(实际需要 browserless 抓取 F10 页面) # TODO: 实现 browserless CDP 抓取 return { "goodwill": None, # 商誉(亿元) "total_assets": None, # 总资产(亿元) "ratio": None, # 商誉占比 "status": "pending", # pending / safe / warning / danger "source": "东方财富F10", } def fetch_pledge_ratio(code6): """ 大股东质押率 雷点阈值:>50% 数据源:东方财富 F10 """ # 简化版:返回模拟数据 # TODO: 实现 browserless 抓取质押数据 return { "pledge_ratio": None, # 质押率 "pledge_amount": None, # 质押股数 "status": "pending", "source": "东方财富F10", } def fetch_operating_cashflow(code6, quarters=4): """ 经营性现金流净额(最近N季度) 雷点阈值:连续2季度负值 数据源:东方财富财务分析 """ # 简化版:返回模拟数据 # TODO: 实现财务数据抓取 return { "recent_quarters": [], # 最近N季度现金流 "negative_count": 0, # 负值季度数 "status": "pending", "source": "东方财富财务分析", } ``` The scoring logic initializes the result as safe and only increases risk when non-null values satisfy a threshold: ```python score = 0 reasons = [] signal = "SAFE" gw = radar_data.get("goodwill", {}) gw_ratio = gw.get("ratio") if gw_ratio and gw_ratio > RED_LINE_THRESHOLDS["goodwill_ratio"]: score += 30 reasons.append(f"商誉占比过高({gw_ratio}% > 30%)") signal = "CRITICAL_RISK" pl = radar_data.get("pledge", {}) pl_ratio = pl.get("pledge_ratio") if pl_ratio and pl_ratio > RED_LINE_THRESHOLDS["pledge_ratio"]: score += 25 reasons.append(f"大股东质押率高({pl_ratio}% > 50%)") sig ...[truncated 2597 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
references/config-template.json:34
Finding

Privileged Browser-Control Endpoint Uses a Hard-Coded Plaintext URL

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (26)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims broader web retrieval, DuckDuckGo-based filtering, and integrated source-credibility analysis, but the described code path appears to only perform narrow keyword sentiment scoring. Such mismatches can mislead users into relying on incomplete or non-existent safeguards, especially in a financial analysis context where false confidence in data provenance and filtering can drive poor decisions or unnecessary permission grants.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims broader web retrieval, DuckDuckGo-based filtering, and integrated source-credibility analysis, but the described code path appears to only perform narrow keyword sentiment scoring. Such mismatches can mislead users into relying on incomplete or non-existent safeguards, especially in a financial analysis context where false confidence in data provenance and filtering can drive poor decisions or unnecessary permission grants.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README advertises broad natural-language triggers such as asking the agent to 'scan risk' or 'identify financial problems' without clearly constraining what actions, data sources, symbols, or workflow boundaries are allowed. In an agent setting, vague trigger scope can cause unintended invocation, overbroad interpretation of user intent, or execution paths beyond the narrow stock-risk function, which is especially sensitive in a finance-related skill.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
82% confidence
Finding

The skill advertises use of network-capable tools (web_search, web_fetch, and references to external APIs) but does not declare an explicit tool scope such as permissions or allowed-tools. This creates ambiguity around what the skill is authorized to access and can lead to overbroad network use, especially in agent ecosystems where undeclared capabilities may still be reachable via runtime defaults or inherited permissions.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding

Using npx clawhub install xiaodi-a-share-guard without pinning a version introduces supply-chain risk because the package resolved at install time may change unexpectedly or be replaced by a malicious release. In agent/tooling environments, this can result in users installing different code than what was reviewed, undermining trust in the analyzed skill.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The natural-language triggers are broad and overlap with ordinary stock-related conversation, which can cause unintended invocation of a skill that uses network and execution-capable tooling. Accidental activation increases the chance of unnecessary external requests, unintended code execution paths, and user surprise, particularly in chat environments where discussing a stock should not automatically launch a multi-tool analysis workflow.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest requests the exec tool for a finance-focused stock risk diagnosis skill, but the stated functionality in metadata does not justify shell execution. Granting command execution unnecessarily expands the attack surface and could enable arbitrary local command execution if the skill prompt or downstream logic is manipulated.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill declares a browser automation capability via a browserless plugin even though its stated purpose is stock risk diagnosis, which does not inherently require remote browser control. Unnecessary browser access expands the attack surface and could enable unintended web interaction, data access, or lateral movement through the configured internal CDP endpoint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This markdown file contains only Chinese-language instructions and explanations, with no opt-in, alternative locale, or justification for restricting the content to a single language. Under the policy rule, forcing a specific language without user choice is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file claims a three-layer consensus model, but the implementation only consumes two scored inputs and still produces definitive outputs. In a financial risk-diagnosis skill, this mismatch can mislead downstream agents or users into believing an additional independent validation layer exists when it does not, weakening trust and potentially producing overconfident investment guidance.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The fallback branch is commented as handling 'three-layer consensus', but in reality it processes every non-special-case combination, including partial disagreement states. This can cause maintainers, auditors, or calling systems to misinterpret disagreement as consensus and trust a final rating that was not derived under the stricter policy described in the comments.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The module docstring describes this as a 'browserless CDP version' that uses browserless CDP to render JS pages, and a browserless URL constant is defined. However, the actual implementation performs plain urllib requests to Eastmoney AJAX endpoints and does not invoke the browserless service at all, so the documentation actively misstates how the scraper works.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The function is documented as obtaining pledge data and notes JS rendering is needed, which implies some retrieval behavior. In reality it performs no fetching at all and always returns a static placeholder with pending status, so the documentation overstates the implemented capability.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/f10_scraper.py (reported line 224)May include surrounding context.

python
暂时返回占位数据,实际需要 browserless
    """
    # 东方财富质押页面
    # https://data.eastmoney.com/gp/gplist.html
    
    return {
        "pledge_ratio": None,

Session Persistence

Medium
Category
Rogue Agent
Confidence
75% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/f10_scraper.py (reported line 224)May include surrounding context.

python
暂时返回占位数据,实际需要 browserless
    """
    # 东方财富质押页面
    # https://data.eastmoney.com/gp/gplist.html
    
    return {
        "pledge_ratio": None,

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The module description, CLI description, and output labels are written only in Chinese, which imposes a specific language/locale without indicating user choice or opt-in. The policy requires flagging language constraints unless the tool offers a choice or clearly documents a justified region-specific limitation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill is marketed as an expert-grade stock risk diagnosis system, yet core risk-diagnosis components are stubbed out and always return non-real results. Because the tool still produces a final risk level, users may trust materially incomplete assessments, which is especially dangerous in a financial decision-support context.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documented behavior promises real financial risk indicators, but the implemented functions return placeholder values marked as pending. In a stock-risk diagnostic skill, this can mislead users into treating incomplete analysis as authoritative, causing unsafe financial decisions through false reassurance or missing red flags.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The argparse description and help text force a single language for user interaction, and there is no indication that users can choose another language or that the locale restriction is intentional. This is a natural-language policy issue rather than a code-security flaw.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The module docstring describes active collection of web content and filtering of PR-style articles as core tasks. In actual code, there is no network access, search integration, or content fetching; analyze_stock_sentiment merely analyzes caller-supplied texts or returns a pending placeholder message when no texts are provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill instructions and invocation examples are entirely in Chinese and describe interaction through Chinese natural-language prompts, but the document does not state that the tool is intentionally limited to Chinese users or offer any language choice. This can violate language/locale policy when a skill implicitly requires one language without user opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The natural-language description for the slash command is only in Chinese ("一键扫描个股风险"), which implies a fixed language presentation for invocation guidance. The file does not provide any user-selectable language option or document that this skill is intentionally limited to a Chinese-language audience.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file’s natural-language specification mandates output categories and report content entirely in Chinese, and the implementation later hardcodes Chinese report strings. There is no indication that users can choose another language or that the locale restriction is documented as a justified region-specific policy requirement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

User-facing strings such as the report headings, rating labels, and argparse descriptions are fixed in Chinese, which can violate a language/locale policy when no opt-in or alternative locale is provided. The file does not expose any parameter or documented setting for language selection.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

This code performs HTTP requests to third-party services and transmits user-supplied stock codes, but there is no print/log statement, confirmation, or explanatory warning near the operation. For a code file, outbound network access should have some visible disclosure unless clearly communicated elsewhere.

Content

No source excerpt is available for this finding.