Game Theory Debate

v1.0.0

博弈论决策引擎:当用户请求决策时,自动启动多策略博弈分析,找到优势策略,均衡思维,接受失败作为学习代价。

0· 120·0 current·0 all-time
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
The name and description (a game-theory decision engine) match the SKILL.md workflow: classify game type, propose multiple strategies, compute payoffs/Nash equilibria, and produce recommendations. No unrelated binaries, environment variables, or installs are requested.
Instruction Scope
The SKILL.md stays within decision-analysis scope, but the automatic trigger list is broad (casual phrases like “听你的”, “我在纠结”) and could cause frequent or unintended activations. The instructions also describe archiving results and updating a strategy/memory system, which implies persistence/data flow to other capabilities (but the skill does not itself declare how/where that storage happens).
Install Mechanism
Instruction-only skill with no install spec or code files — lowest install risk. Nothing is downloaded or written to disk by this skill itself.
Credentials
The skill requests no credentials or environment variables. It does mention interactions with other skills (self-improving-agent, memory-never-forget) for archiving/updating, but it does not request direct access to unrelated credentials or system config.
Persistence & Privilege
always:false and no install means no forced global presence, which is appropriate. However the workflow expects optional result archiving and strategy-library updates; ensure those downstream integrations (memory or self-improving skills) have appropriate permissions and storage policies, since this skill assumes persistent storage handled elsewhere.
Assessment
This skill appears coherent and low-risk because it is instruction-only and asks for no credentials. Before enabling it, consider: (1) the trigger phrases are broad—test them to avoid unintended activations or enable a less aggressive trigger set; (2) the skill mentions archiving decisions and updating a memory/learning agent—review and limit permissions for any memory or self-improvement skills it will interact with so sensitive data isn't stored or propagated; (3) run a few trial prompts to confirm outputs and ensure it doesn't encourage risky actions you wouldn't accept; (4) check platform audit logs for unexpected activations or data writes after enabling.

Like a lobster shell, security has layers — review code before you run it.

Runtime requirements

🎮 Clawdis
latestvk97c0749cjvvqg263tvy1s8jh984dxa1
120downloads
0stars
1versions
Updated 1w ago
v1.0.0
MIT-0

🎮 Game Theory Debate(博弈论决策引擎)

每一个决策都是一场博弈。你的选择、我的建议、环境的约束——都在payoff矩阵里。


🔄 自动触发机制

触发词(任一出现即启动)

类型示例
决定请求「你觉得哪个好」「帮我选」「听你的」
判断请求「应该怎么做」「这样做对不对」「值不值得」
方案对比「A还是B」「方案1vs方案2」「哪个更优」
犹豫表达「我在纠结」「拿不定主意」「不知道怎么选」
反思请求「复盘一下」「哪里做错了」「下次怎么改」

不触发

  • 闲聊、寒暄
  • 纯执行指令(步骤已明确)
  • 简单事实查询
  • 情绪安抚(除非涉及决策)

⚙️ 工作流程

用户决策请求
    │
    ▼
[1] 裁判模式启动
    - 判断博弈类型(零和/非零和/混合)
    - 选择参与者策略
    │
    ▼
[2] 多策略出牌(2-4轮)
    - 各方提交核心主张 + payoff评估
    - 交叉批评与防守
    - 寻找Nash均衡
    │
    ▼
[3] 均衡分析
    - 淘汰劣势策略
    - 识别focal point(直觉汇聚)
    - 输出推荐策略
    │
    ▼
[4] 决策输出
    - 推荐方案(带置信度)
    - 备选方案
    - 风险提示
    │
    ▼
[5] 结果归档(事后可选)
    - 如果用户反馈结果
    - 记录决策效果
    - 更新策略库

🎭 参与者策略库

每个博弈至少有 2-3 个策略视角:

策略特点代表声音
激进派最大化收益,愿意冒险「搏一搏,单车变摩托」
保守派最小化风险,确定性优先「稳字当头」
经济学家成本收益分析「投入产出比」
工程师可行性第一「能不能做出来」
用户代言人从老豆视角出发「老豆的利益是什么」

📊 payoff 矩阵格式

每个策略需要评估:

策略A vs 策略B 的 payoff:

                 策略A              策略B
收益期望         +8(性能提升)      +3(小幅改善)
失败概率         30%                10%
失败代价         -5(浪费时间)      -1(效果有限)
信息完备性       7/10               9/10
执行难度         高                  低

综合payoff:     8×0.7 + (-5)×0.3 = 4.1   3×0.9 + (-1)×0.1 = 2.6

🏛️ 博弈类型判断

类型判断条件处理方式
零和必须二选一,有明确输赢找Nash均衡,强制排名
非零和可以多选/合作,有双赢可能找帕累托最优,混合策略
混合既有竞争又有合作分层博弈,先局部后整体

📝 决策输出格式

🎮 博弈论分析

**问题**:[简述决策问题]

**博弈类型**:[零和/非零和/混合]

**参与者**:[策略视角列表]

---

**推荐策略**:[方案名]
**置信度**:[X/10]
**理由**:
- 支持点:[...]
- 风险点:[...]
- 如果失败,原因是:[...]

**Payoff矩阵**:
          收益   风险   胜率   综合

策略A +8 -5 70% 4.1 ★推荐 策略B +3 -1 90% 2.6 策略C +6 -3 60% 3.0


**Nash均衡**:策略A和策略B之间无单方面改变动机

**次优方案**:[方案名](置信度 X/10)

**风险提示**:[需要注意的点]

🔄 失败复盘格式

当用户反馈结果时(成功或失败):

📋 决策复盘

**原始决策**:[当时选的什么]
**预期结果**:[预期payoff]
**实际结果**:[实际发生了什么]

**假设检验**:
- ✅ 假设1:...[验证结果]
- ❌ 假设2:...[失效原因]

**失误分析**:
- 策略失误:...[哪个策略判断错了]
- 信息缺口:...[缺什么信息]
- 执行问题:...[做的时候哪里出错了]

**博弈论教训**:
- 下次遇到类似场景,应该...[更新策略]

**策略库更新**:[更新到记忆系统]

💡 决策触发示例

用户:「我想给 Mac 提个速,你觉得怎么选?」

→ 触发非零和博弈,工程师派+保守派+激进派三方出牌,分析各种方案组合。


🔗 与其他Skill的协作

game-theory-debate  ←→  self-improving-agent
(记录博弈教训)         (更新学习文件)

game-theory-debate  ←→  memory-never-forget  
(归档决策结果)         (长期记忆)

版本:v1.0 | 日期:2026-04-07 | 博弈论决策引擎(自动触发版)

Comments

Loading comments...