Recursive Self Improvement
递归自我改进系统,能够自动检测错误并修复,或持续优化和重构。包含修复模式和优化模式,支持并发执行、自动化测试、性能监控、智能调度、自适应学习、错误预测和异常恢复。用于需要持续自我优化的系统。
MIT-0 · Free to use, modify, and redistribute. No attribution required.
⭐ 6 · 4.5k · 26 current installs · 26 all-time installs
MIT-0
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The skill claims to detect errors, modify code/logic, run tests, monitor performance, and perform deployments/refactors. Yet the package declares no binaries, no credentials, no config paths, and supplies no connectors or concrete tooling. Performing the stated tasks in a real environment would normally require repository/CI credentials, test runners, monitoring APIs, or filesystem access — none of which are requested or described.
Instruction Scope
SKILL.md contains actionable workflows that include '代码/逻辑变更', '单元/集成/回归测试', '性能监控', and '递归调用'. These are high-level but permissive instructions that could lead an agent to read/write source files, invoke test suites, call CI or monitoring endpoints, or deploy changes. The instructions lack explicit limits, approval gates, file/path scopes, or safe-guarding procedures (e.g., require manual review before committing changes).
Install Mechanism
Instruction-only skill with no install spec and no code files. That minimizes direct supply-chain risk since nothing is downloaded or installed by the skill itself.
Credentials
The manifest requests no environment variables or credentials, but the described capabilities inherently require access to source repositories, CI/CD, test runners, and monitoring systems. The absence of declared required credentials or explicit scoping is a mismatch and could lead to the agent asking for or attempting to use broad credentials at runtime.
Persistence & Privilege
always:false and no install means the skill does not demand permanent system presence. However model invocation is allowed (normal), and combined with instructions that tell the agent to autonomously modify code, this creates a higher-risk scenario: an autonomously-invoking agent could enact changes if given tool access or permissions. There are no instructions in SKILL.md that explicitly require persisting credentials or modifying other skills.
What to consider before installing
This skill defines an autonomous 'self-improvement' process that includes modifying code and running tests but provides no concrete connectors, required credentials, or safety gates. Before installing or enabling it: 1) Treat it as potentially high-risk — run it only in an isolated test environment or sandbox. 2) Do not grant it repository, CI, or monitoring credentials with write/deploy scope — instead create narrow, read-only or test-only accounts if needed. 3) Require human approval for any code commits, merges, or deployments (add explicit gating/PR review). 4) Ask the author for provenance and for concrete connector specs (which repos/CI/metrics it needs) and for built-in safeguards (rollback, authorization, audit logs). 5) If you want lower risk, limit the skill to producing repair/optimization suggestions rather than applying changes automatically, or disable autonomous invocation so every action requires your explicit permission.Like a lobster shell, security has layers — review code before you run it.
Current versionv1.0.0
Download ziplatest
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
SKILL.md
递归自我改进系统
核心模式
系统有两种基本工作模式,根据系统状态自动切换:
1. 修复模式 (REPAIRING)
触发条件: 检测到错误或异常
工作流程:
- 错误识别(类型/位置/影响)
- 根因分析
- 修复方案设计
- 代码/逻辑变更
- 单元测试
- 集成测试
- 验证通过?→ 是则标记已修复,否则回到步骤1
系统状态: REPAIRING → REPAIRED → STABLE
2. 优化模式 (OPTIMIZING)
触发条件: 系统稳定运行,无错误超过 N 轮
工作流程:
- 性能指标收集
- 代码复杂度分析
- 重构方案设计
- 迁移计划
- 分步实施
- 回归测试
- 指标对比
- 是否有提升?→ 是则标记已优化,否则保留原样
系统状态: OPTIMIZING → OPTIMIZED → STABLE
状态标记
INITIAL: 初始状态REPAIRING: 修复模式中OPTIMIZING: 优化模式中STABLE: 稳定运行ERROR: 检测到错误OPTIMIZED: 已优化完成
并发执行引擎
系统支持多任务并发执行:
任务池 → 智能调度 → 并发执行 → 结果收集
调度策略:
- 基于任务复杂度
- 考虑历史成功率
- 预测执行时间
- 动态调整并发数
默认配置:
- 并发工作池大小:4
- 超时时间:5秒
- 重试次数:3
自动化测试框架
系统内置测试框架:
测试类型:
- 单元测试:验证单个功能
- 集成测试:验证模块间交互
- 性能测试:验证性能指标
测试覆盖率:
- 目标覆盖率:80%+
- 关键路径覆盖率:100%
性能监控仪表盘
实时监控以下指标:
系统状态:
- 当前版本
- 运行轮次
- 系统模块
性能指标:
- 并发任务数
- 平均执行时间
- 吞吐量(任务/分钟)
- CPU 使用率
- 内存使用率
智能任务调度器
基于历史数据和预测的智能调度:
优先级计算:
- 任务复杂度评估
- 历史成功率分析
- 最近性能趋势
- 截止时间紧迫性
调度策略:
- 高优先级任务优先执行
- 同优先级任务 FIFO
- 动态调整资源分配
自适应学习引擎
从执行中学习,持续优化:
学习内容:
- 任务执行成功率
- 性能瓶颈识别
- 模式识别
预测能力:
- 任务成功率预测
- 性能趋势预测
- 资源需求预测
错误预测系统
提前识别潜在错误:
预测维度:
- 任务类型模式
- 资源使用模式
- 时间分布模式
预测阈值:
- 低置信度:60%
- 中置信度:80%
- 高置信度:90%
异常恢复系统
智能错误处理和恢复:
内置策略:
TIMEOUT: 重试 + 指数退避MEMORY_ERROR: 并行化处理CONCURRENCY_LIMIT: 动态调整并发数
恢复流程:
- 错误检测
- 策略匹配
- 执行恢复
- 验证结果
运行记录格式
每次运行记录使用标准格式:
{
"timestamp": "2026-02-05T21:55:00Z",
"mode": "REPAIRING | OPTIMIZING | STABLE",
"action": "fix | refactor | validate | monitor",
"previous_state": "状态名称",
"current_state": "状态名称",
"details": "详细描述",
"results": {
"key1": true/false,
"key2": "value"
}
}
版本管理
系统自动管理版本:
版本格式: vN.M
升级规则:
- v1.0: 基础框架
- v2.0: 添加并发、测试、监控
- v3.0: 添加智能调度、学习引擎、错误预测
- v4.0: 添加恢复系统、完整生态
升级条件:
- 完成 N 轮优化
- 累计改进 10+ 项
- 持续稳定运行 24 小时
使用建议
何时使用:
- 需要持续改进的复杂系统
- 有明确性能指标的项目
- 需要自动化测试和验证的流程
- 多模块并行处理的任务
最佳实践:
- 初始化时定义明确的性能基线
- 每次优化后进行回归测试
- 定期检查错误预测和建议
- 保留优化历史用于分析
配置参数
可在配置文件中调整:
{
"optimization": {
"min_stable_rounds": 3,
"max_concurrent_tasks": 8,
"timeout_seconds": 5
},
"testing": {
"target_coverage": 80,
"critical_coverage": 100
},
"monitoring": {
"metrics_interval": 60,
"alert_thresholds": {
"cpu": 80,
"memory": 90
}
}
}
资源
Files
3 totalSelect a file
Select a file to preview.
Comments
Loading comments…
