luoke-knowledge - SKILL.md Agent Skill

name: luoke-knowledge

description: '洛克王国：世界的记忆存储技能，用于保存已学习映射与游戏事实。支持保存/查询/删除意图到动作的知识（如 "open map" -> "press m"）以及可复用事实。适用于用户教学或回忆历史游戏知识。'

metadata:

requires:

bins: ["python3"]

游戏知识库

平台：跨平台 - 纯 Python，无 Windows 依赖

学习并召回游戏特定映射与自定义动作。

能力

保存：存储按键绑定、动作映射、游戏事实
查询：按意图或关键词检索知识
列出：查看全部已学习知识
删除：移除错误或过时知识

知识类型

| 类型 | 示例 | 存储形式 |

|------|------|---------|

| key_binding | "按 M 打开地图" | {"intent": "open map", "action": "press m"} |

| custom_action | "治疗是先按 1 再按 e" | {"intent": "heal", "action": "press 1, press e"} |

| game_fact | "Boss 在 10 分钟刷新" | {"fact": "boss spawn time", "value": "10 minutes"} |

用法

保存知识


# 按键绑定

python3 {baseDir}/scripts/save_knowledge.py --intent "open map" --action "press m"



# 自定义动作序列

python3 {baseDir}/scripts/save_knowledge.py --intent "heal" --action "press 1, wait 0.5 sec, press e"



# 游戏事实

python3 {baseDir}/scripts/save_knowledge.py --fact "boss spawn time" --value "10 minutes"

查询知识


# 按意图查询

python3 {baseDir}/scripts/query_knowledge.py "open map"

# 返回: {"intent": "open map", "action": "press m"}



# 按关键词查询

python3 {baseDir}/scripts/query_knowledge.py "heal"

# 返回: {"intent": "heal", "action": "press 1, wait 0.5 sec, press e"}

列出全部知识


python3 {baseDir}/scripts/list_knowledge.py

删除知识


python3 {baseDir}/scripts/delete_knowledge.py --intent "open map"

与 Agent 的联动

当用户说：

“记住按 M 打开地图” -> save_knowledge.py --intent "open map" --action "press m"
“打开地图” -> query_knowledge.py "open map" -> 执行返回动作

存储位置

知识保存到：~/.luoke-aw/knowledge.json


{

  "key_bindings": [

    {"intent": "open map", "action": "press m", "learned_at": "2026-03-30T10:30:00"}

  ],

  "custom_actions": [

    {"intent": "heal", "action": "press 1, press e", "learned_at": "2026-03-30T10:31:00"}

  ],

  "game_facts": [

    {"fact": "boss spawn time", "value": "10 minutes", "learned_at": "2026-03-30T10:32:00"}

  ]

}

示例


# 用户：“记住按 M 打开地图”

save_knowledge.py --intent "open map" --action "press m"



# 稍后用户：“打开地图”

RESULT=$(query_knowledge.py "open map")

ACTION=$(echo $RESULT | jq -r '.action')

parse_action.py "$ACTION" --execute