Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only thinking-format skill that changes response style but does not request tools, data access, persistence, or privileged actions.
Install this if you want Chinese-language response templates for reasoning and planning. Be aware that common phrases such as requests for concise output or structured planning may activate a mode and change the answer format unexpectedly.
The natural-language content of the skill is entirely presented in Chinese, including trigger phrases and output formats, with no indication that users may select another language. Under the stated policy, forcing a specific language without user opt-in can constitute a language/locale policy violation unless clearly justified.
The trigger phrase '分步分析' is generic and likely to appear in ordinary user requests, which can cause the skill to activate when the user did not intend to invoke this specific mode. In a skill-routing or prompt-composition system, overly broad activation terms can create unintended prompt injection of the skill’s formatting and behavior, reducing reliability and potentially interfering with higher-priority safety or task instructions.
The phrases '全局规划' and '先搭框架' are common natural-language expressions rather than unique invocation tokens, so they may collide with normal requests. This can lead to unintended skill activation and prompt-shaping side effects, causing the assistant to follow the skill template instead of the user’s actual preferred format or system-level routing expectations.
GranolaAI’s listed triggers like '简洁模式' and '快速输出' are broad requests that many users would naturally make without intending to invoke a skill. If these phrases auto-activate the skill, the model may suppress explanation or alter response structure unexpectedly, which can degrade transparency and lead to incomplete or mismatched outputs.
The trigger phrase '结构化梳理' is highly generic and fits many everyday requests, making accidental invocation plausible. In an agent environment, such unintended activation can hijack the output format, introduce unrequested diagramming behavior, and reduce determinism in how user intents are mapped to skills.
No suspicious patterns detected.
