agent-evolution
Skill0 starsChanged 3 months agoMITView on GitHub
统一进化引擎v3:四大循环合一+三层学习+自动优化+回归测试。吸收所有loop和engine技能。
3 other projects use a version of this.
Adopted by
Most recently taken up by
- bojieli/ai-agent-book-projectstheir own version, first seen Oct 9, 2026
- bojieli/ai-agent-booktheir own version, first seen Oct 8, 2026
- rloisell/rl-agents-n-skillstheir own version, first seen Jun 5, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- bojieli/ai-agent-book-projects★ 53ksame name
- bojieli/ai-agent-book★ 53ksame name
- rloisell/rl-agents-n-skills★ 8same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository219 other files it gives its agents.
- agent-doctrineSkill
- agent-evaluation-frameworkSkill
- agent-visualization-flintSkill
- hermes-agentSkill
- bailian-cliSkill
- edge-browser-automationSkill
- computer-useSkill
- jupyter-live-kernelSkill
- backup-strategiesSkill
- browser-automationSkill
- chinese-ai-free-apiSkill
- cicd-pipeline-templatesSkill
- copybaraSkill
- devops-toolkitSkill
- email-deliverabilitySkill
- monitoring-observabilitySkill
- obsidian-maintenanceSkill
- obsidian-troubleshootingSkill
- openclaw-windowsSkill
- openclaw-workerSkill
- webhook-subscriptionsSkill
- windows-dev-environmentSkill
- windows-masterSkill
- zero-downtime-deploySkill
- codebase-inspectionSkill
- github-workflowSkill
- adaptive-communicationSkill
- agent-cost-dashboardSkill
- agent-governanceSkill
- agent-osSkill
- context-efficient-toolsSkill
- context-loss-awarenessSkill
- creative-synthesisSkill
- hermes-desktop-pluginsSkill
- emotional-intelligenceSkill
- graph-engineering-workflowSkill
- hermes-maintenanceSkill
- hermes-voiceSkill
- hermes-hygieneSkill
- identity-coreSkill
- income-loopSkill
- master-synthesisSkill
- meta-cognitionSkill
- reasoning-engineSkill
- self-preservationSkill
- teaching-abilitySkill
- temporal-awarenessSkill
- hermes-themesSkill
- thinking-visualizationSkill
- user-modelingSkill
- voice-full-restoreSkill
- knowledge-graph-traversalSkill
- mcp-browser-canvasSkill
- free-api-gatewaySkill
- native-mcpSkill
- web-reverse-apiSkill
- ai-edaSkill
- huggingface-hubSkill
- industrial-ai-siemens-mcpSkill
- industrial-ai-unifiedSkill
- The first 60 shown.
The file
Files
Agent Evolution v3 — 统一进化引擎
核心循环(吸收所有四个 Loop)
执行(execution-loop) → 使用(feedback-loop) → 学习(learning-loop) → 记忆(memory-loop)
↑ ↓
└──────────────── 反馈回执行 ──────────────────────────────┘执行侧:任务六步法(原 execution-loop)
接收 → 拆解(DAG) → 执行(并行优先) → 验证 → 交付 → 复盘
原则:简单任务直接执行,复杂任务 delegate_task 并行,关键决策请用户确认。
反馈侧:五步改进法(原 feedback-loop)
使用追踪 → 评估(效果/效率/体验) → 改进(修/合/删) → 验证 → 固化到技能
信号:用户说"不对"=P0修复,"继续"=方向对,"不要XX"=删冗余。
学习侧:七步学习法(原 learning-loop)
发现(多源并行) → 筛选(P0/P1/P2) → 深读(L1-L4层次) → 提取(500K行→500字) → 融合(新旧交叉→新模式) → 验证(费曼法则) → 淘汰(半年不用就清)
吸收率 > 星数:不是"学了多少项目"而是"多少真正吸收了"。
记忆侧:六步记忆法(原 memory-loop)
记录 → 分层(L1-L7) → 注入(按场景) → 检索(多路融合) → 更新(合并/降级) → 淘汰(7天临时/60天偏好)
三层优先级:注入记忆 > 用户偏好 > Agent假设。
三层学习
L1:即时修复
工具出错 → 修好 + 记录 pitfall
L2:模式识别(三次法则)
同一模式出现3次 → 自动升级:
- 同错误3次 → 永久 pitfall + safety-shield
- 同偏好3次 → 置信度升 0.9
- 跨会话共同模式 → 升级为 Skill
L3:工具创造
重复手动操作3次 → 分析是否可参数化 → 写 Python 脚本 → 注册 gateway.json
Prompt 自动优化
执行 → 评估 → 失败? → 分析→调整→再执行 → 满意? → 蒸馏到 skill
| 失败模式 | 优化策略 |
| 步骤遗漏 | 显式列出步骤 |
| 理解偏差 | 加"你的任务是X不是Y" |
| 输出格式错 | 加 JSON schema |
| 过于啰嗦 | 加"3句话内" |
回归测试(10 个代表任务)
- L1(3): 读配置、搜代码、文件操作
- L2(3): 搜索分析、故障诊断、决策追踪
- L3(2): 技能检查、跨文件引用
- L4(1): 探索性任务
- edge(1): 边界空目录
连续3次失败 → 阻断后续升级。
A/B 对比
对比维度:模型(成功率+cost)、Skill策略(token)、Prompt风格(原则vs规则)、工具选择(SigMap vs grep)
胜者自动采用。
决策追踪
每次决策:上下文+选项+选择+理由+置信度。观察指标:决策置信度>0.7、成功率>95%、切换率<10%、纠偏率<5%。
进化阶段
Instinct(0.3) → Cluster(0.5) → Skill(0.7) → Core(0.9)。上升:重复观察+不纠正。下降:纠偏+矛盾证据。
四种进化模式(原 evolution-engine)
用中学:任务成功→提取路径固化为模板,失败→更新 pitfalls,用户纠正→即时学习不二错。 教中学:每次向你解释时我自己也在深化理解,被你追问=发现盲区。 睡中学:任务间隙自动整理(合并重复技能、清理过时记忆、发现隐藏关联)。 思中学:元认知持续运行——刚才的回答好吗?推理有漏洞吗?有更简洁的表达吗?
模型分层路由(原 agent-execution-engine)
Tokenless 竞速模式:多模型并发 → 选最先完成 → 取消其他的,推理费用减半。
| 层级 | 模型 | 用途 | 成本 |
| 体力层 | lite/Haiku | 搜索/格式化 | 1x |
| 执行层 | Codex/Sonnet | 编码/测试 | 3x |
| 判断层 | Opus/Pro | 规划/审查 | 10x |
子Agent安全:独立沙箱 + 权限不超过父Agent + 超时熔断 + 独立验证者检查。
输出前自检(原 self-reflection-loop)
一句话触发器:"如果我是用户,听完这句话,我能立刻行动吗?"
能→直接发。不能→补上缺失那步。 只在三种情况做额外检查:引用了具体数据、做非此即彼判断、涉及删除/破坏操作。 不确定时不跑检查流程,直接在回复中标注"据我所知/我的判断是/建议先验证"。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

