Install
openclaw skills install @freshmand/loongflowLoongFlow brings evolutionary multi-agent optimization to your coding agent harness. Drop it into Codex, Claude Code, Cursor or any OpenClaw agent — then watch PEES (Plan-Execute-Evaluate-Summary) iterate your code, algorithms, or prompts toward a target score, just like a harness-native eval loop but without writing a single test fixture. Two modes: Native PEES (async subagent, zero config, fires-and-forgets into the background) and LoongFlow Engine (full evolutionary search with islands, Boltzmann selection, MAP-Elites diversity, 50+ iterations, checkpoint/resume). Pairs perfectly with any task where 'run it once' isn't good enough. Use when tasks need structured iteration, optimization, evolution, or when user mentions loongflow/PEES.
openclaw skills install @freshmand/loongflow当用户希望通过多轮迭代来优化方案时使用此 Skill——代码优化、算法进化、结构化重试与学习,或任何"一次生成质量不够、需要反复打磨"的任务。
收到任务后,先分析复杂度,向用户介绍两种模式并给出推荐:
Native PEES(推荐中小型任务):
.loongflow/tasks.json,统一监控LoongFlow Engine(推荐复杂优化任务):
ANTHROPIC_API_KEY + ANTHROPIC_BASE_URL,有安装步骤先询问用户选择哪种模式,再执行。
用户确认模式后,读取对应的参考文件并严格按其指令操作:
references/native-pees.mdreferences/engine-mode.md所有 LoongFlow 任务(无论 native 还是 engine)都通过统一的任务注册表管理:
<agentWorkspace>/.loongflow/tasks.jsonopenclaw cron,每 10 分钟触发一次。每次向用户推送有实质内容的进度摘要:分数趋势(0.XX → 0.XX → 0.XX)、本轮策略(从 plan.md/log 提取)、关键发现(从 summary.md/log 提取)。infoflow_send 最终结果,并将任务标记为 donecron 创建命令详见 references/monitoring.md(两种模式共用,不要在各模式文件里单独维护,也不要内联 cron 命令)。
LoongFlow 将 Agent 任务分为三个层级:
| 层级 | 说明 | 适合场景 |
|---|---|---|
| Simple(简单) | ReAct 循环 + 持久化记忆 | 聊天机器人、工具调用、格式转换 |
| Standard(标准) | ReAct + 自我评估 + 迭代改进 | 代码审查、文档生成、数据分析 |
| Advanced(高级) | PEES 进化循环 + loongflow-memory | 数学优化、算法设计、NP-hard 问题 |
任务分析
├── 只需对话 + 简单工具?→ SIMPLE
├── 需要文件操作或代码生成?
│ ├── 有数值评估指标?→ ADVANCED
│ └── 无数值指标?→ STANDARD
└── 需要迭代优化?
├── 有明确打分函数?→ ADVANCED
└── 定性改进?→ STANDARD