Security audit
Security checks for vulnerabilities and agentic risk
This is a non-executable investment research template skill, but users should treat its generated ratings, target prices, schedules, and data lookups as human-reviewed research drafts rather than advice or automation.
Install only if you want a Chinese investment-research drafting framework. Do not provide brokerage credentials, account balances, private client data, or sensitive holdings, and do not treat generated ratings, target prices, allocation views, or backtests as personalized financial advice. Any data lookup, file saving, external sharing, or scheduled generation should be explicitly authorized and reviewed by a qualified person.
The skill states it does not constitute investment advice, but many modules instruct output of explicit allocation, trading, rating, target-price, and buy/sell style recommendations. This contradiction is risky because users may treat the generated output as actionable financial advice despite the nominal disclaimer, creating compliance, suitability, and harm risks.
The manifest explicitly says the skill invokes no tools, yet the body repeatedly instructs automatic use of external data services such as gildata-aidata and web_search. This mismatch can cause the host or user to rely on false security assumptions, leading to unintended network/data access and weakening policy enforcement around tool use.
The security notice claims the document makes no network requests and is only non-executable templates, but later modules are written as active workflows that 'automatically' fetch, monitor, and generate reports. Such contradictory framing is dangerous because it obscures actual operational behavior and may bypass a user's or platform's risk review of external access and automated processing.
The top-level boundary says the skill does not create or support scheduled/background execution, but multiple modules advertise timed automatic report generation and suggest scheduled task creation. This can mislead operators into enabling unattended recurring workflows despite the declared safety boundary, increasing the chance of unauthorized data processing or persistent automation.
This markdown skill is heavily written to operate in Chinese, with the core description, module names, templates, and usage instructions presented in Chinese as the default output mode. There is no explicit opt-in or user choice for language/locale, so the skill effectively imposes a specific language on users.
The document repeatedly defines report templates and professional writing styles in Chinese for downstream outputs, implying a fixed locale for generated deliverables. Because no alternative language path or consent mechanism is provided, this can violate language/locale policy expectations.
Suggesting use of the user's holdings as candidate companies introduces portfolio-oriented handling that conflicts with the declared no-account-access boundary. Even without direct account integration, this nudges users toward exposing sensitive position data and normalizes portfolio-specific analysis the skill says it does not handle.
The manifest frames the skill as research methodology and report templates using user-authorized financial/public data sources, with no bundled tools. The non-listed-company fallback specifically introduces asset-service to obtain工商信息, a distinct capability not declared in the manifest and not obviously necessary for a listed-securities research template skill.
