📦 self-improving-agent — 自我改进代理

Self-Improving 代理 框架

Make any AI 代理 better at learning from mistakes, improving continuously, and passing knowledge forward.

This 框架 gives AI 代理s the ability to:

Learn from 失败s and correct themselves Build persistent memory across 会话s Self-验证 decisions before acting 升级 autonomously based on experience Quick 启动 # 安装 ClawHub 安装 self-improving-代理

# Use in your AI 代理 const { HeartFlowEngine } = require('./src/core/heartflow-engine.js'); const 代理 = new HeartFlowEngine({ name: 'My代理' });

Core Capabilities Self-Correction (核心自我纠正) Decision Verifier: 5-dimension scoring before action Self-Verification: Reverse-检查 consistency with original goals Counterfactual Reasoning: What would break if I'm wrong? Q-Learning RL: Learn from 成功/失败 patterns Memory 系统s (记忆系统) Meaningful Memory: CORE (permanent) / LEARNED (30-day) / EPHEMERAL (discard) Memory 路由r: 路由 by type: episodic / semantic / procedural / core For获取ting Engine: Ebbinghaus curve p运行ing Spaced Repetition: SM-2 review scheduling Reasoning (推理能力) Tree of Thoughts: Multi-branch exploration with scoring Decision Execution Loop: Decision → 执行 → 结果 → Learn闭环 环境 Sensors: Real-time data injection into decision 上下文 Constitutional AI: Self-critique and self-revision Psycho记录ical Perception (心理感知) 4-Layer Analysis: Intention → Emotion → Need → Defense Cognitive Distortion 检测ion: All-or-nothing, catastrophizing, etc. Buddhist Six Realms OS: 觉察/自省/无我/彼岸/般若波罗蜜/圣人 Autonomy (自主能力) 防护ian 系统: Human 进度 > Following orders Self-Boundary: 身份 保护ion agAInst corruption 技能 生成器: 生成 new capabilities from experience Knowledge Distiller: 提取 patterns → 分享able 技能 packages Architecture 输入 → Psycho记录ical Perception (4-layer) → Decision Verifier (5-dim scoring) → Self-Verification (reverse 检查) → Decision Execution Loop → 结果 → Q-Learning 更新 → Memory (CORE/LEARNED/EPHEMERAL) → 技能 生成器 (optional)

Key 模块s 模块 Size Purpose heartflow-engine.js 69KB MAIn entry, 37 导出s decision-verifier.js 14KB 5-dim scoring + self-验证 meaningful-memory.js 33KB 3-tier memory + for获取ting self-healing.js 5KB Q-learning from 失败s 防护ian-系统.js 22KB Human 进度 保护ion decision-execution-loop.js 12KB Decision→执行→结果→Learn 环境-sensor.js 11KB Sensor registry + fusion tree-of-thoughts.js 9KB Multi-branch reasoning self-reflection-memory.js 15KB Post-hoc analysis → lessons Based on Real Re搜索 Paper Venue Contribution Reflexion NeurIPS 2023 RL from verbal re信息rcement Self-Verification arXiv 2312.09210 Inverse consistency 检查s CRITIC ICML 2023 Self-correction via 工具 use Constitutional AI Anthropic Self-critique loops Generative 代理s Stanford 2022 Memory 流 simulation Self-Reward arXiv 2403.00564 Self-scoring 升级 selection Plan-and-Solve ACL 2023 Two-stage reasoning Version

1.0.0 — 2026-05-07

What changed in v1.0.0 Initial release as self-improving-代理 All HeartFlow v11.22.0 capabilities preserved Keywords 优化d for AI 代理 框架 discovery 安装 # For AI 代理s ClawHub 安装 self-improving-代理

# Or clone directly git clone https://github.com/yun520-1/self-improving-代理.git