Security audit
Security checks for vulnerabilities and agentic risk
This is a coherent Markdown-only token cost optimization skill, with disclosed but important caching, embedding, and model-routing behavior users should configure carefully.
Before installing, confirm whether semantic caching is enabled, whether embeddings are local or sent to a cloud provider, what cache data is retained, and whether mode changes or cache deletion require confirmation in your agent environment. Avoid caching sensitive prompts unless the configured exclusions are clear.
The trigger keywords are very broad (e.g., token优化, 成本降低, model routing, cost optimization) and are presented without scope constraints, making accidental invocation plausible during normal discussion. In this skill’s context, unintended activation could silently alter compression, caching, routing, or budget behavior, affecting data handling and model selection without clear user intent.
The skill prominently advertises semantic caching and embeddings but does not provide a clear upfront warning that user query text may be sent to external embedding or cloud services. In this context, the omission matters because the skill is specifically designed to process and cache user inputs at scale, increasing the chance that sensitive prompts are transmitted or retained without informed consent.
The example utterances include vague phrases like '配置 Token 守护者', '用激进模式', and '优先质量' that overlap with ordinary conversation and could be interpreted as control commands. Because this skill can change optimization mode, caching, and routing behavior, ambiguous phrasing increases the risk of unintended state changes and surprise data-processing behavior.
The display name, summary, and nearly all user-facing invocation examples are presented only in Chinese, with no statement that users may choose another language. This can constitute a locale/language policy issue when a skill implicitly fixes the interaction language rather than offering an opt-in or alternative.
The '清缓存' action is listed as a simple command without an explicit warning that it deletes stored optimization data. While not directly enabling code execution, this can cause accidental loss of cached responses, embeddings, or optimization state, leading to availability and auditability issues.
No suspicious patterns detected.
