cm
使用 CM 构建持久的跨代理记忆
AI 编码代理会积累宝贵的知识,但在会话之间会丢失这些知识。CM 捕获所有代理的模式和学习成果,将分散的经验转化为持久的程序性记忆,并通过置信度评分随时间不断改进。
Baixar o ZIP da skill
Upload no Claude
Vá em Configurações → Capacidades → Skills → Upload skill
Ative e comece a usar
Testar
A utilizar "cm". cm context "fix authentication timeout bug" --json
Resultado esperado:
- Relevant rules found: Always check token expiry before auth debugging (Score: 8.5/10)
- Anti-patterns: Don't cache auth tokens without expiry validation
- Similar sessions: 3 days ago - Fixed OAuth timeout in user service
- Suggested searches: cass search "authentication timeout" --robot
A utilizar "cm". cm trauma list
Resultado esperado:
- Active trauma patterns: DROP TABLE (Severity: critical)
- Blocked commands matching trauma patterns will show: BLOCKED: This command matches a trauma pattern
Auditoria de Segurança
Baixo RiscoDocumentation-only skill describing a legitimate memory system. Static analyzer flagged 227 patterns, but all are false positives: the trauma guard safety examples were misidentified as dangerous commands, scoring terminology was confused with crypto, and documentation examples were flagged as executable code. No actual code or scripts are present.
Fatores de risco
📁 Acesso ao sistema de arquivos (1)
🌐 Acesso à rede (1)
⚙️ Comandos externos (1)
Pontuação de qualidade
O Que Você Pode Construir
在代理之间共享知识
当 Cursor 学习到调试模式时,CM 会自动将其提供给 Claude Code。
构建团队手册
将个人代理经验转化为共享的团队知识,包括经过验证的模式和反模式。
个人编码记忆
永远不会忘记如何解决棘手的错误。CM 会记住成功的模式并警告有问题的方法。
Tente Estes Prompts
Run cm context "implement user authentication with JWT" --json to get relevant patterns from your playbook
Use cm playbook add "Always validate JWT expiry before debugging auth errors" --category security
Run cm reflect --days 7 --json to see extracted patterns and provide feedback
Install trauma guard with cm guard --install to block commands matching harmful patterns
Melhores Práticas
- 在开始非平凡任务之前运行 cm context 以获取相关指导
- 留下内联反馈注释,例如 // [cass: helpful b-8f3a2c],说明规则是有帮助还是有害
- 仅在审查隐私影响后才启用跨代理增强功能
Evitar
- 不要在没有证据的情况下手动添加规则 - 让 CM 从实际会话中提取规则
- 避免共享包含敏感项目信息的手册文件
- 在不了解模式被阻止原因的情况下,不要禁用创伤防护
Perguntas Frequentes
我的会话数据会被发送到外部服务吗?
CM 支持哪些代理?
置信度衰减是如何工作的?
我可以在没有 LLM API 密钥的情况下使用 CM 吗?
如果 CM 出错了会怎样?
CM 与搜索会话历史有何不同?
Detalhes do Desenvolvedor
Autor
DicklesworthstoneLicença
MIT
Repositório
https://github.com/Dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations/tree/main/skills/cmReferência
main
Estrutura de arquivos
📄 SKILL.md