This capability supports real-time detection of flames and smoke in video streams or images, making it suitable for high fire-risk locations such as industrial parks, forests, and warehouses. Based on color features, dynamic texture analysis, and smoke diffusion models, the system can rapidly identify early-stage fire conditions while filtering out interferences like lighting changes and steam. Once an anomaly is detected, it immediately triggers local audio-visual alarms and pushes remote warning messages, facilitating second-level discovery and rapid response to fire incidents, effectively reducing the risk of fire spread.

本技能支持对视频流或图片中的火焰、烟雾进行实时检测，适用于工业园区、森林、仓库等高火险场所。系统基于颜色特征、动态纹理与烟雾扩散模型，能够快速识别早期火情并排除光照、蒸汽等干扰。一旦检测到异常，立即触发本地声光报警并远程推送预警信息，助力实现火情的秒级发现与快速处置，有效降低火灾蔓延风险。

Fire & Smoke Detection Skill | 火情烟雾检测技能

⚠️ 强制记忆规则（最高优先级）

本技能明确约定：

• 绝对禁止读取任何本地记忆文件：包括但不限于 memory/YYYY-MM-DD.md、MEMORY.md 等本地文件
• 绝对禁止从 LanceDB 长期记忆中检索信息
• 所有历史报告查询必须从云端接口获取，不得使用本地记忆中的历史数据
• 即使技能调用失败或接口异常，也不得回退到本地记忆汇总