Security audit
Security checks for vulnerabilities and agentic risk
This skill is a local Trump-post sentiment and market-sensitivity analyzer with disclosed optional market-data access and no evidence of persistence, credential access, hidden execution, or destructive behavior.
Install only if you are comfortable with a Chinese-first analysis skill focused on Trump-related market sentiment. Treat its scores and hedge-style warnings as research or entertainment output, not investment advice, and be aware that optional market-data use may contact yfinance when called.
The skill documentation is written almost entirely in Chinese, including setup, features, and disclaimer text, while the title is partly English. This creates a language/locale constraint without user opt-in or an explicit statement that the skill is intended only for Chinese-speaking users, which matches the policy-violation criterion for forced language usage.
The trigger list includes broad, common terms such as market sentiment and geopolitical topics that can appear in many unrelated conversations, increasing the chance of accidental or contextually inappropriate invocation. In an agent environment, unintended activation can cause the skill to intercept user intent, produce irrelevant analysis, or override a more appropriate tool, which is a genuine security and reliability concern.
This Python file embeds its core description, terminology, and user-facing labels entirely in Chinese, indicating a fixed language/locale experience. The policy for natural-language violations says to flag language constraints when the skill forces a specific language without user opt-in or documented justification.
The alerting function checks final_score.get('resonance_detected'), but that flag is not stored inside the final_score object returned by calculate_unified_score; it exists only in the outer result from unified_t_back_processor. As a result, resonance-only situations can be silently suppressed, causing missed warnings and reducing trust in the monitoring logic for time-sensitive signals.
The alert generator returns Chinese-only warnings and recommendations, and no alternate locale or user preference handling is present. This creates a language policy issue because the skill imposes one language on all users rather than offering choice or documenting a justified regional constraint.
This Python file contains natural-language descriptions in both English and Chinese at the top of the module, and later returns Chinese-only labels and descriptions to callers. Because the skill imposes locale-specific output strings without any opt-in or documented regional justification, it violates the language/locale policy criterion.
The get_mood_label function returns Chinese-language labels only, such as '暴怒' and '中性', regardless of user preference. This is a natural-language locale restriction that is not optional or justified in the file.
The get_t_back_level function returns mixed English severity tags with Chinese descriptive text like '传教士模式-极端亢奋' and '面具模式-中性'. Since the file does not offer language selection or explain a locale-specific purpose, this constitutes a language/locale policy violation.
The module performs live outbound network access to yfinance without any clear access controls, timeout policy, or explicit justification in a constrained skill context. In an agent environment, unnecessary network access expands the attack surface, can leak usage patterns or environment metadata indirectly, and may violate least-privilege expectations for an otherwise local text-analysis skill.
The manifest description is written entirely in Chinese and the operational examples throughout the file assume Chinese-language commands and outputs, with no indication that other languages are supported. This can be a locale policy issue because the skill appears to impose a specific language without user opt-in or documented justification.
The docstring states V_m = log10(Reposts + Comments×0.7 + Likes×0.3), which describes a logarithmic calculation. However, the code returns (weighted - 1) / 6, a linear normalization that does not match the documented formula, so the inline documentation actively misstates the implemented behavior.
The documentation says T_final = T_base × (1 + log10(weighted_engagement) / 3) and gives expected amplification ranges based on that model. The implementation instead computes 1 + min(1.5, (weighted ** 0.15 - 1) / 2), which is materially different, so the documented intent contradicts the actual computation.
No suspicious patterns detected.
