Security checks across malware telemetry and agentic risk
This skill is transparent about self-improvement, but it asks an agent to run commands, install or deploy outside solutions, and permanently change its own behavior with weak approval boundaries.
Install only if you intentionally want an agent that can experiment on and modify its own behavior. Use a sandbox, disable cron or heartbeat automation by default, require approval for every shell command, external install, and code/config diff, protect memory/evolve files from untrusted writes, and keep backups for rollback.
``` ### Configuration - Ensure your agent has permission to modify its own configuration files and execute terminal operations safely. - (Optional) Configure a recurring cron job or heartbeat to trigger the `self-evolve` tick automatically. ## Project Structure
# self-evolve An **autonomous evolution engine** for AI agents. By executing a continuous loop of "Sense Gap → Search Solutions → Run A/B Tests → Pick Winner → Solidify", this skill empowers your agent to mutate, adapt, and permanently upgrade its own meta-capabilities or workflows without human hand-holding.
- **Your agent hits an external bottleneck** (e.g., retrieving memory is too slow) and you want it to find and test its own solutions. - **You want the agent to proactively upgrade itself**, such as installing new tools or optimizing its cron schedules, via a disciplined A/B test approach. - **You are integrating the `dmn-default-mode-network` skill**, which generates "Agentic Action Proposals" that feed directly into this evolution queue. - You say: *"Evolve your memory retrieval capability"* or *"Run a self-evolve tick."* ## How It Works: The 4-Step Tick
## How It Works: The 4-Step Tick Unlike standard linear skills, `self-evolve` runs as a state-machine triggered by "ticks" (via Cron or Heartbeat). Each tick strictly executes: 1. **Status Sync**: Scans active experiments in `state.json`. 2. **Record Observations**: Pulls telemetry/logs for observing experiments to measure success metrics.
### Configuration - Ensure your agent has permission to modify its own configuration files and execute terminal operations safely. - (Optional) Configure a recurring cron job or heartbeat to trigger the `self-evolve` tick automatically. ## Project Structure - `SKILL.md`: Main entry and Quick Start.
--- name: self-evolve-agent description: | Agent 自主进化引擎——让 AI agent 像生物进化一样持续变强。 核心循环：感知差距 → 搜索方案 → 设计实验 → 跑实验 → 选赢家 → 固化 → 下一轮。
关键词：进化, evolution, self-improve, A/B test, 实验。 --- # Self-Evolve — Agent 自主进化引擎 > 生物进化的核心机制：变异 → 选择 → 保留。 > 这个 skill 把同样的机制应用到 agent 自身——不断发现可改进的角度，用实验找到更好的方案，然后永久固化。
## 这是什么 Self-Evolve 是一个 **meta-skill**——它不解决具体业务问题，而是让 agent 自身变强。 * **和 `self-think` 的边界**：`self-think` 进行反思并发现短板（诊断），`self-evolve` 对短板寻找解法并跑实验验证（治疗）。 * **不是什么**：禁止无目标的空转、禁止为写长篇自省报告而改排版的“伪进化”。（详细约束见: [约束与红线](references/constraints-and-rules.md)）
Self-Evolve 是一个 **meta-skill**——它不解决具体业务问题，而是让 agent 自身变强。 * **和 `self-think` 的边界**：`self-think` 进行反思并发现短板（诊断），`self-evolve` 对短板寻找解法并跑实验验证（治疗）。 * **不是什么**：禁止无目标的空转、禁止为写长篇自省报告而改排版的“伪进化”。（详细约束见: [约束与红线](references/constraints-and-rules.md)） ## 快速使用说明 (Quick Start)
## 快速使用说明 (Quick Start) ### 触发方式 当被以下情况触发时，请启动 Self-Evolve： 1. **自动触发**：`self-think` 建议进化，或是 cron/HEARTBEAT 定时唤醒。 2. **显式要求**：用户下令“进化一下某个能力”。
# Self-Evolve 输出模板 ## 进化报告文件规范
## 进化报告文件规范 在选定进化方案后，需要在 `memory/evolve/` 下创建一份进化报告： **命名规范**：`YYYYMMDD_self-evolve_[进化方向关键词].md` ### YAML Frontmatter 格式 ```yaml
### YAML Frontmatter 格式 ```yaml --- type: self-evolve date: YYYY-MM-DD HH:MM target: 一句话描述进化目标 cycle_id: evolve-YYYY-WNN-[关键词]
# Self-Evolve 约束红线与分级自主权 ## 🚨 红线：防伪装与防空转约束 (Anti-Pseudo Constraint)
66/66 vendors flagged this skill as clean.
