auto-evolving-agent(智能体自主进化)
AdvisoryAudited by Static analysis on Apr 30, 2026.
Overview
No suspicious patterns detected.
Findings (0)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If used too broadly, the skill could change how the agent behaves in future sessions or alter the installed skill set.
The declared evolution scope includes modifying core agent instructions/configuration, creating new skills, updating the agent persona, and installing or removing capability modules.
1. **修改自己的配置** ... 2. **创建新的Skill** ... 4. **完善和优化Soul** ... 6. **能力模块管理** - 评估、搜索、安装、淘汰能力模块
Require explicit per-change approval, inspect diffs before applying changes, and avoid installing/removing modules unless the source and impact are clear.
The agent may repeatedly propose or perform self-modification workflows after tasks or on a schedule, increasing the chance of unwanted behavioral drift.
The skill defines autonomous or scheduled self-improvement triggers, which can keep changing behavior beyond a single user-requested task if enabled or interpreted broadly.
任务完成后自动反思 ... 连续3次类似任务后 ... 定期自动进化（可选）- 每天一次（快速反思） - 每周一次（深度进化）
Keep periodic evolution disabled by default, require a fresh user confirmation for every evolution run, and define clear stop conditions.
Private task details or bad instructions could be preserved and reused across future tasks unless the user reviews the memory entries.
The gene-pool template stores success cases and reusable prompt injections, meaning prior task content or flawed patterns can influence future agent behavior.
- **成功案例**: - [案例1] - [案例2] - **注入提示词**: [在进化提示中如何描述这个基因]
Require user approval before adding gene-pool entries, exclude secrets and private data, and provide a simple way to inspect, edit, expire, or delete stored patterns.