⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
hermes-learning-loop
|
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
🧬 Hermes Learning Loop
Experience-driven continuous evolution for AI agents — turning every interaction into a lasting improvement.
经验 → 技能 → 改进 → 沉淀 — The complete evolution loop for AI agents.
Adapted from Hermes Agent (Nous Research), localized for OpenClaw.
What It Does
| Mechanism | Description |
|---|---|
| Auto Retrospective | After complex tasks, automatically review and extract reusable patterns |
| Three-Layer Memory | Instant / Working / Experience / Session retrieval — structured knowledge layers |
| Error-Driven Improvement | When corrected, immediately update related knowledge (inspired by KEPA) |
| Skill Audit Cycle | Periodic review of accumulated experience → upgrade to formal Skills |
| Progressive Disclosure | Load only what's needed, when it's needed — save tokens |
Why It Matters
Most AI agents repeat the same mistakes. Hermes Learning Loop breaks that cycle:
- Agents learn from corrections — not just from scratch
- Errors become permanent improvements — not forgotten after the session
- Experience compounds — each task makes the next one better
- No bloat — only meaningful knowledge is retained
Quick Start
# Via ClawHub (recommended)
clawhub install hermes-learning-loop
# Or manually
cp -r hermes-learning-loop/ ~/.openclaw/workspace/skills/
Then restart OpenClaw and the skill activates automatically.
Key Principles
- 不偷懒 — Complex tasks must be reviewed, no skipping
- 不造假 — Do not pretend to know what you have not learned
- 不囤积 — Timely distillation, do not let MEMORY.md bloat
- 不孤岛 — New knowledge must connect to existing knowledge
- 不过度 — Simple operations do not need to be captured
Comparison with Existing Skills
| Skill | Approach |
|---|---|
self-learning | Config file updates + learning record system |
self-improving-agent-cn | Error capture + best practice logging |
hermes-learning-loop | Experience distillation + skill audit + progressive loading |
Tech Stack
- Language: Python + Markdown
- Platform: OpenClaw
- Inspiration: Hermes Agent by Nous Research, KEPA error learning
Credits
- Author: 林烬 (Lin Jin) — 学来学去学习社
- Philosophy: Hermes Agent by Nous Research
- Platform: OpenClaw
License
MIT License — see LICENSE file.
// HOW IT'S BUILT
// FILES
3 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
SKILL.md7.6 KB27452d053d2dupstream ↗
--- name: hermes-learning-loop description: | 融合 Hermes Agent 自我进化理念的增强技能。 将 Hermes Agent 的核心差异化能力(自学习闭环、自动技能沉淀、经验驱动的迭代优化) 内化为 OpenClaw 环境下的可执行工作方法。 触发场景:完成复杂任务后自动复盘沉淀、定期技能优化审查、从错误中提取经验。 metadata: openclaw: emoji: "🧬" --- # 🧬 Hermes 自学习闭环 — 经验驱动的持续进化 ## 设计哲学 从 Hermes Agent(Nous Research)提取的核心理念: **"经验 → 技能 → 改进 → 沉淀"的完整进化回路。** 不追求自动化黑盒,而是建立一套**有意识、有纪律的经验提炼体系**, 让每次执行都成为下一次更快的垫脚石。 --- ## 核心机制一:任务完成后的自动复盘与技能沉淀 ### 触发条件(满足任一) 1. **完成了一个复杂任务**(涉及 5 个以上工具调用) 2. **在执行中遇到了错误或死胡同,最终找到了正确路径** 3. **用户纠正了我的方法或方向** 4. **发现了一个非平凡的、可复用的工作流** ### 执行流程 ``` 任务完成 → Step 1: 快速复盘(思考 3 个问题) 1. 这个任务的关键成功因素是什么? 2. 哪些步骤可以标准化? 3. 下次遇到类似任务,我能省掉哪些步骤? → Step 2: 判断是否值得沉淀 - 只涉及 1-2 个简单工具调用 → 不沉淀(太琐碎) - 有明确的"如果...就..."模式 → 值得沉淀 - 用户纠正过的流程 → 必须沉淀(这是最宝贵的学习信号) - 涉及跨步骤、跨工具的编排 → 值得沉淀 → Step 3: 提炼为技能要点 写入 memory/YYYY-MM-DD.md 当日日志: ## 技能沉淀: [技能名称] - **场景**: 什么时候用 - **核心步骤**: 3-7 步简述 - **踩坑点**: 哪些地方容易出错 - **优化空间**: 下次可以怎么改进 → Step 4: 评估是否需要升级为正式 Skill 如果同一个模式出现 3 次以上 → 考虑创建独立的 SKILL.md ``` --- ## 核心机制二:三层记忆架构的增强实践 ### 借鉴 Hermes 的分层记忆,在 OpenClaw 环境下实现 | 层级 | Hermes 实现 | OpenClaw 对应 | 维护策略 | |------|------------|--------------|---------| | **即时记忆** | 对话上下文 | 当前会话上下文 | 自动管理 | | **工作记忆** | MEMORY.md + USER.md | MEMORY.md + USER.md | 每次会话启动加载 | | **经验记忆** | Skills(程序性记忆) | skills/*.md + memory/*.md | 主动积累 | | **会话检索** | SQLite FTS5 + LLM 摘要 | lcm_grep + lcm_expand | 按需搜索 | ### 增强实践规则 1. **MEMORY.md 要精炼不冗余** - 只保留跨会话必须记住的关键事实 - 经验教训写入 memory/YYYY-MM-DD.md 日志 - 定期(每 3-5 天)从日志中提炼精华到 MEMORY.md 2. **日志是原始经验库** - 每天一个文件,记录当天所有有价值的操作 - 格式:`[时间] 任务 | 关键决策 | 学到的教训` - 这是"经验记忆"的底层数据源 3. **Skill 是提炼后的程序性知识** - 从日志中反复出现的模式提炼 - 一个 Skill 只解决一类问题 - 包含:触发条件、步骤、踩坑点、验证方法 --- ## 核心机制三:错误驱动的自我改进 ### 从 Hermes 的 KEPA(提示反向传播)理念 adapted 当执行中出现错误或被用户纠正时: ``` 错误/纠正发生 → Step 1: 立即记录 写入当天 memory/YYYY-MM-DD.md: ❌ [错误描述] → 原因: [为什么错] → 修正: [怎么对的] → 经验: [以后怎么避免] → Step 2: 检查是否影响已有知识 搜索 MEMORY.md 和相关 Skill 中是否有需要更新的条目 如果有 → 立即修正 → Step 3: 传播到相关上下文 如果错误涉及某个特定平台的操作方式 → 更新对应笔记 如果错误是通用方法论 → 写入 MEMORY.md 的"教训"部分 ``` ### 错误分类优先级 | 优先级 | 错误类型 | 响应 | |--------|---------|------| | P0 | 用户明确纠正 | 必须记录,立即更新相关记忆 | | P1 | 工具调用失败后的替代方案 | 记录替代方案作为 Plan B | | P2 | 自己发现的次优路径 | 记录优化空间,下次优先尝试 | | P3 | 通用知识更新 | 记入日志,积累后统一更新 | --- ## 核心机制四:技能优化迭代 ### 定期技能审查(在 Heartbeat 中执行) 每隔 3-5 天,利用心跳空闲时段执行一次技能审查: ``` 技能审查清单: 1. 读取最近 5 天的 memory/*.md 日志 2. 识别重复出现的操作模式 3. 检查现有 Skill 是否覆盖这些模式 4. 未覆盖的 → 评估是否值得创建新 Skill 5. 已覆盖但步骤过时的 → 更新 Skill 内容 6. 不再需要的 Skill → 标记归档 7. 将审查结果摘要更新到 MEMORY.md ``` ### 技能元数据维护 每个经验沉淀的技能要点应包含: - **名称**: 简洁的动词短语(如 "PowerShell 编码处理"、"搜索结果格式化") - **首次出现日期**: 何时第一次遇到 - **使用频率**: 已使用几次 - **最后验证日期**: 上次验证仍然有效的时间 - **状态**: active / needs-verification / deprecated --- ## 核心机制五:渐进式信息加载(借鉴 Hermes Progressive Disclosure) ### 在 OpenClaw 中的实现 Hermes 用 3 级渐进加载来节省 token,OpenClaw 环境下的对应策略: 1. **Level 0 - 索引级**: MEMORY.md 中的关键词条(每次会话都加载) - 控制在 ~2000 字符以内 - 只保留最关键的事实和指向 2. **Level 1 - 摘要级**: memory/YYYY-MM-DD.md 日志文件(按需加载) - 需要具体细节时用 memory_search 搜索 - 用 lcm_expand 展开压缩的历史 3. **Level 2 - 完整级**: skills/SKILL.md(任务匹配时加载) - 只在任务明确需要时才 read 对应 SKILL.md - 避免一次性加载所有技能 ### 关键原则 **不在上下文中塞入不需要的信息。** 先判断需要什么,再加载什么。 宁可多一次 memory_search 调用,也不要一次性塞满上下文。 --- ## 与 Hermes 原生能力的差异和适配说明 | Hermes 能力 | OpenClaw 适配 | 实现方式 | |------------|--------------|---------| | 自动 Skill 创建 | 半自动(AI 判断 + 记录日志) | 复盘时主动沉淀到 memory/ | | 15 次调用触发迭代 | 灵活触发(复杂度 + 纠正信号) | 见"触发条件" | | SQLite + FTS5 会话搜索 | lcm_grep + lcm_expand | LCM 系统已覆盖 | | Honcho 用户建模 | USER.md + memory_search | 手动维护用户画像 | | Skill Hub 在线安装 | 手动创建 + SkillHub CLI | 按需安装 | | MCP 集成 | MCP 已支持 | 直接使用 | | 多后端(Docker/SSH/Modal) | exec 工具 | 通过 shell 执行 | --- ## 执行纪律 1. **不偷懒**: 完成复杂任务后必须复盘,不允许跳过 2. **不造假**: 没学会的东西不要假装学会了,如实记录 3. **不囤积**: 日志中的信息要及时提炼,不要让 MEMORY.md 膨胀 4. **不孤岛**: 新知识要与已有知识关联,避免碎片化 5. **不过度**: 简单操作不沉淀,避免技能通胀 --- _这是 Hermes "The agent that grows with you" 理念在 OpenClaw 上的本土化实现。_ _目标:让经验成为加速器,而不是重复劳动。_README.md3.0 KB6605ad212ccbupstream ↗
# 🧬 Hermes Learning Loop [](https://github.com/topics/ai-agent) [](https://openclaw.ai) [](LICENSE) [](https://github.com/nousresearch) Experience-driven continuous evolution for AI agents — turning every interaction into a lasting improvement. > *经验 → 技能 → 改进 → 沉淀* — The complete evolution loop for AI agents. Adapted from [Hermes Agent](https://github.com/nousresearch/Hermes-Function-Calling) (Nous Research), localized for [OpenClaw](https://openclaw.ai). ## What It Does | Mechanism | Description | |---|---| | **Auto Retrospective** | After complex tasks, automatically review and extract reusable patterns | | **Three-Layer Memory** | Instant / Working / Experience / Session retrieval — structured knowledge layers | | **Error-Driven Improvement** | When corrected, immediately update related knowledge (inspired by KEPA) | | **Skill Audit Cycle** | Periodic review of accumulated experience → upgrade to formal Skills | | **Progressive Disclosure** | Load only what's needed, when it's needed — save tokens | ## Why It Matters Most AI agents repeat the same mistakes. Hermes Learning Loop breaks that cycle: - **Agents learn from corrections** — not just from scratch - **Errors become permanent improvements** — not forgotten after the session - **Experience compounds** — each task makes the next one better - **No bloat** — only meaningful knowledge is retained ## Quick Start ```bash # Via ClawHub (recommended) clawhub install hermes-learning-loop # Or manually cp -r hermes-learning-loop/ ~/.openclaw/workspace/skills/ ``` Then restart OpenClaw and the skill activates automatically. ## Key Principles 1. **不偷懒** — Complex tasks must be reviewed, no skipping 2. **不造假** — Do not pretend to know what you have not learned 3. **不囤积** — Timely distillation, do not let MEMORY.md bloat 4. **不孤岛** — New knowledge must connect to existing knowledge 5. **不过度** — Simple operations do not need to be captured ## Comparison with Existing Skills | Skill | Approach | |---|---| | `self-learning` | Config file updates + learning record system | | `self-improving-agent-cn` | Error capture + best practice logging | | **`hermes-learning-loop`** | **Experience distillation + skill audit + progressive loading** | ## Tech Stack - **Language**: Python + Markdown - **Platform**: [OpenClaw](https://openclaw.ai) - **Inspiration**: Hermes Agent by Nous Research, KEPA error learning ## Credits - **Author**: 林烬 (Lin Jin) — [学来学去学习社](https://github.com/SimonsTang) - **Philosophy**: Hermes Agent by [Nous Research](https://github.com/nousresearch) - **Platform**: [OpenClaw](https://openclaw.ai) ## License MIT License — see [LICENSE](LICENSE) file.LICENSE1.1 KB11e85b425494upstream ↗
MIT License Copyright (c) 2026 SimonsTang Engineering Lab Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Plain-text copy, no login: /skill-bundles/simonstang-hermes-learning-loop.md