{"id":50376845,"url":"https://github.com/phamduchuong517-hub/hermes-self-evolution","last_synced_at":"2026-06-12T16:34:28.630Z","repository":{"id":352904604,"uuid":"1217025222","full_name":"phamduchuong517-hub/hermes-self-evolution","owner":"phamduchuong517-hub","description":"🧬 Zero-dependency self-evolution for LLM agents. One file + one prefill = Agent learns from corrections forever. No databases, no APIs, no overengineering.","archived":false,"fork":false,"pushed_at":"2026-05-30T10:05:31.000Z","size":227,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-30T10:14:39.991Z","etag":null,"topics":["ai-agent","debugging","hermes-agent","mcp","self-evolution","skills","task-orchestrator","token-optimization","workflow"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/phamduchuong517-hub.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-04-21T13:22:13.000Z","updated_at":"2026-05-30T08:48:26.000Z","dependencies_parsed_at":null,"dependency_job_id":"0ffa62be-a78e-42f4-a903-f574b8339ccf","html_url":"https://github.com/phamduchuong517-hub/hermes-self-evolution","commit_stats":null,"previous_names":["phamduchuong517-hub/openclaw-core-skills"],"tags_count":6,"template":false,"template_full_name":null,"purl":"pkg:github/phamduchuong517-hub/hermes-self-evolution","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/phamduchuong517-hub%2Fhermes-self-evolution","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/phamduchuong517-hub%2Fhermes-self-evolution/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/phamduchuong517-hub%2Fhermes-self-evolution/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/phamduchuong517-hub%2Fhermes-self-evolution/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/phamduchuong517-hub","download_url":"https://codeload.github.com/phamduchuong517-hub/hermes-self-evolution/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/phamduchuong517-hub%2Fhermes-self-evolution/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34253936,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-12T02:00:06.859Z","response_time":109,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-agent","debugging","hermes-agent","mcp","self-evolution","skills","task-orchestrator","token-optimization","workflow"],"created_at":"2026-05-30T10:00:54.918Z","updated_at":"2026-06-12T16:34:28.621Z","avatar_url":"https://github.com/phamduchuong517-hub.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# 🦞 Hermes Self-Evolution — AI Agent 自我进化核心技能包\n\n**7 个技能**，让 AI 助手学会持续进化：记住经验、纠正行为、优化成本、不重复犯错。\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)\n[![Version: 4.1.0](https://img.shields.io/badge/Version-4.1.0-green.svg)](https://github.com/phamduchuong517-hub/hermes-self-evolution/releases/tag/v4.1.0)\n[![Hermes Agent](https://img.shields.io/badge/Hermes-Agent-blue)](https://github.com/project-hermes/hermes-agent)\n[![OpenClaw](https://img.shields.io/badge/OpenClaw-Core-red)](https://github.com/openclaw)\n\n---\n\n## 为什么要用这套技能？\n\n大多数 AI Agent 系统都有一个共同问题：**每次对话都是从零开始**。\n\n- 用户说\"我不喜欢看长分析\" → 下次对话又发长篇\n- 用户说\"用这种格式\" → 下回又忘了\n- 昨天修复的 bug → 今天又犯一次\n- 随着技能积累 → 系统Prompt越来越胖，越来越慢\n\n这套技能包用**文件级**方案解决了\"Agent 如何记住经验和规则\"的问题——不需要数据库、不需要外部服务、不需要多代理编排器。\n\n---\n\n## 📦 技能一览\n\n### 1. memory-system-v2 (记忆系统) ✨\n\n**解决什么问题**\nAgent 需要记住用户偏好、历史纠正、项目约束和外部参考，但不能每条都塞进系统Prompt。传统做法是\"全量注入MEMORY.md\"——不分类，无法合并，重复多。\n\n**怎么解决的**\n四分类记忆提取 + 主题自动合并。用 `[MEM_APPEND:type: 内容]` 标签在回复中标记，脚本自动分类存储到 MEMORY.md 的独立区块。\n\n```\n分类: user(用户偏好) | feedback(纠正教训) | project(项目约束) | reference(外部指向)\n```\n\n**什么时候用它**\n- 每次对话都会自动使用（通过在 prefill 指令中声明）\n- 夜间有 cron 自动做主题合并 + 去重\n\n### 2. self-improvement-core (自我进化核心)\n\n**解决什么问题**\nAgent 进化方案要么是工程浩大的多代理架构（如 Reflexio 的四阶段流水线），要么是行为分析\"只记不改\"的方案。小团队/个人用户跑不动。\n\n**怎么解决的**\n只需一个 `SELF-EVOLUTION.md`（46行规则表） + 行为规则。每次对话开始前，Agent 自动扫描规则表，被用户纠正后自动写新规则。\n\n```\n架构: 在线层（读规则→执行→记录）→ 离线层（cron异步分析）\n```\n\n**什么时候用它**\n- 系统启动即自动加载（prefill 指令驱动）\n- 每次被用户纠正后，规则自动更新\n\n### 3. hermes-self-evolution (自我进化流程)\n\n**解决什么问题**\nAI 犯的错只被纠正一次，没有根因分析和防复发机制。同一个类型的错误可能反复出现。\n\n**怎么解决的**\n五步闭环：观察错误 → 分析根因 → 制定纠正 → 验证效果 → 固化规则到 SELF-EVOLUTION.md。\n\n**什么时候用它**\n- 当用户纠正 Agent 行为时\n- 当系统发现重复错误时（与 error-logger 联动）\n\n### 4. task-orchestrator (任务编排器)\n\n**解决什么问题**\n复杂任务（4+步骤）Agent 容易跑偏：做到一半忘了目标、遗漏关键步骤、跳过验证直接报完成。\n\n**怎么解决的**\n四阶段闭环：规划（拆解步骤）→ 执行（逐步完成）→ 检查（验证每个步骤）→ 反思（总结优化）。支持任务分解、调度、监控、验收。\n\n**什么时候用它**\n- 任务超过3个步骤时\n- 任务涉及多个子系统时\n- 需要分阶段验证结果时\n\n### 5. token-optimization (Token优化)\n\n**解决什么问题**\n长会话上下文膨胀 + 历史注入导致 API 费用飙升、响应变慢。\n\n**怎么解决的**\n三层策略：上下文压缩（保留核心，丢弃噪音）→ 智能缓存（高频使用不重复加载）→ 压缩率 50%+。\n\n**什么时候用它**\n- 长会话（50+轮）显慢时\n- API 账单突然上涨时\n- 想主动控制成本时\n\n### 6. error-logger (错误日志)\n\n**解决什么问题**\n系统报错只报一次，下次遇到同样的错误又重头排查。无根因分析。\n\n**怎么解决的**\n记录错误→分类→根因分析→防止复发。每个错误带时间戳、上下文、根因链，防止同一错误重复排查。\n\n**什么时候用它**\n- 系统报错时自动记录\n- 排查重复问题时优先查日志\n- 与 self-improvement-core 联动做进化输入\n\n### 7. TaskBalancer (任务平衡器)\n\n**解决什么问题**\n多代理系统（如四系统架构：本地/云端 × Hermes/OpenClaw）任务分发无规则，忙的忙死、闲的闲死。\n\n**怎么解决的**\n智能任务分发 + 负载均衡规则。按 代理角色（实操/外勤/贴身/机要）+ 任务类型（编码/研究/运维/沟通）分配。\n\n**什么时候用它**\n- 运行多代理系统时\n- 子代理之间互相冲突时\n- 某个代理过载时\n\n---\n\n## 🧩 技能关系图\n\n```\n                    ┌──────────────────┐\n                    │  任务编排器        │\n                    │  task-orchestrator │\n                    └────────┬─────────┘\n                             │ 编排\n         ┌───────────────────┼───────────────────┐\n         ▼                   ▼                   ▼\n┌──────────────────┐ ┌──────────────┐ ┌──────────────────┐\n│ 自我进化核心      │ │ 记忆系统 v2   │ │ Token 优化       │\n│ self-improve-core│ │memory-system  │ │ token-optimize   │\n│ + 自我进化流程    │ │              │ │                  │\n└────────┬─────────┘ └──────────────┘ └──────────────────┘\n         │ 进化输入\n         ▼\n┌──────────────────┐ ┌──────────────────┐\n│ 错误日志          │ │ 任务平衡器        │\n│ error-logger     │ │ TaskBalancer     │\n└──────────────────┘ └──────────────────┘\n```\n\n---\n\n## ⚡ 快速选择：我该用哪个？\n\n| 你的痛 | 先装这个 |\n|--------|---------|\n| AI 记不住我偏好和纠正 | **memory-system-v2** |\n| AI 犯错后下次还犯 | **self-improvement-core** + **error-logger** |\n| 复杂任务 AI 老是跑偏 | **task-orchestrator** |\n| API 费用高 / 响应慢 | **token-optimization** |\n| 我在跑多个 Agent 实例 | **TaskBalancer** |\n\n**新手入门路线：** memory-system-v2 → self-improvement-core → task-orchestrator\n\n---\n\n## 📂 项目结构\n\n```\nskills/\n├── self-improvement-core/    # 自我进化核心 v7 (106行)\n├── hermes-self-evolution/     # 自我进化流程 v4.3\n│   ├── memory-system-v2/      # 记忆系统 v2 (新增✨)\n│   │   ├── SKILL.md           # 技能文档\n│   │   ├── scripts/           # 记忆追加脚本 + 合并脚本\n│   │   ├── prefill-evolution.txt\n│   │   └── SELF-EVOLUTION.md\n│   └── ...\n├── task-orchestrator/         # 任务编排器 v3\n├── token-optimization/        # Token 优化\n├── error-logger/             # 错误日志 v3\n├── TaskBalancer/              # 任务平衡器\n└── README.md                  # 本文件\n```\n\n---\n\n## 📥 安装\n\n```bash\n# 1. 克隆\ngit clone https://github.com/phamduchuong517-hub/hermes-self-evolution.git\n\n# 2. 按需安装技能\n# 记忆系统 v2\ncp -r skills/hermes-self-evolution/memory-system-v2/scripts/* ~/.hermes/scripts/\nchmod +x ~/.hermes/scripts/memory_appender.sh\ncat skills/hermes-self-evolution/memory-system-v2/prefill-evolution.txt \u003e\u003e ~/.hermes/prefill-evolution.txt\n\n# 自我进化核心\ncp skills/self-improvement-core/SKILL.md ~/.hermes/skills/core/self-improvement/\n\n# 3. 重启或 rehash\nhermes skills reload\n```\n\n---\n\n## 📋 许可证\n\nMIT — 自由使用、修改、商用。无任何担保。\n\n---\n\n*由 Hermes Agent × OpenClaw 社区维护 — 最小化方案，最大效果。*\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fphamduchuong517-hub%2Fhermes-self-evolution","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fphamduchuong517-hub%2Fhermes-self-evolution","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fphamduchuong517-hub%2Fhermes-self-evolution/lists"}