功能描述

Agent 行为固化与进化系统。追踪规则执行、检测行为模式、维护身份连续性。用于：让 agent 的行为规则从"写下来"变成"做到了"。当 agent 需要自我改进、行为追踪、角色一致性、重复检测时激活。与 Memelord/memory-tools 互补：它们管记忆，本 skill 管行为。

使用说明 (SKILL.md)

agent-evolution

Name: Agent Evolution
Author: linglin6

Agent 行为固化与进化系统。追踪规则执行、检测行为模式、维护身份连续性。用于：让 agent 的行为规则从"写下来"变成"做到了"。当 agent 需要自我改进、行为追踪、角色一致性、重复检测时激活。与 Memelord/memory-tools 互补：它们管记忆，本 skill 管行为。

使用

所有命令通过 node scripts/evolution.js \x3Ccommand> [args] 执行，输出均为 JSON。

存储路径：~/.openclaw/workspace/.agent-evolution/state.json

模块

行为追踪器

追踪规则执行与违反，计算权重。

# 初始化（首次使用）
node scripts/evolution.js init

# 添加规则
node scripts/evolution.js add-rule \x3Crule_id> \x3Cdescription> \x3Csource>

# 执行规则前调用，记录 +1
node scripts/evolution.js check \x3Crule_id>

# 记录违反
node scripts/evolution.js violation \x3Crule_id> \x3Ccontext>

# 查看统计
node scripts/evolution.js stats

身份状态层

跨会话持久化的结构化身份。

# 查看身份
node scripts/evolution.js identity

# 更新字段
node scripts/evolution.js identity-update \x3Cfield> \x3Cvalue>

# 记录进化事件
node scripts/evolution.js evolve \x3Cchange> \x3Ctrigger>

模式检测器

检测重复行为和角色偏离。

# 记录操作
node scripts/evolution.js log \x3Caction_type> \x3Cdetail>

# 检查告警
node scripts/evolution.js detect

# 重置计数器
node scripts/evolution.js reset \x3Cpattern_id>

综合报告

node scripts/evolution.js report

初始化规则

从 AGENTS.md / SOUL.md 自动提取规则：

bash scripts/init-rules.sh --agents /path/to/AGENTS.md --soul /path/to/SOUL.md

心跳检查

供 HEARTBEAT.md 调用：

bash scripts/heartbeat-check.sh

安全使用建议

This skill appears to do what it says: it tracks rule executions, detects behavior patterns, and stores an identity/evolution log locally. Before installing or running it, note: (1) you need Node.js and a POSIX shell to run the scripts — the metadata should list these; (2) the skill creates and updates ~/.openclaw/workspace/.agent-evolution/state.json which will contain rules, logs and identity fields — review this file if it may contain sensitive data; (3) the init-rules.sh helper will read whatever file paths you pass (AGENTS.md/SOUL.md) and persist extracted text into state.json, so do not point it at sensitive files; (4) there is no network access or credential exfiltration in the code, but always review and run code from unknown sources in a constrained environment if you are concerned.

功能分析

Type: OpenClaw Skill Name: agent-evolution Version: 1.0.0 The skill is classified as suspicious primarily due to its design to parse and store content from external markdown files (`AGENTS.md`, `SOUL.md`) via `scripts/init-rules.sh` into its `state.json` file. While the `scripts/evolution.js` core logic itself does not execute these stored strings, this mechanism creates a potential prompt injection vulnerability for the OpenClaw agent. If an attacker can control the content of `AGENTS.md` or `SOUL.md`, they could inject malicious instructions or data into the agent's 'behavior rules' or 'identity' state, which the agent might later interpret and execute. This represents a risky capability that could be abused, even if not explicitly malicious within the skill's own code.

能力评估

ℹ Purpose & Capability

The code implements behavior tracking, pattern detection, identity persistence, and rule extraction exactly as described. One mismatch: the registry metadata lists no required binaries, but the SKILL.md and the scripts require node (and bash for the shell helpers). That should be declared in metadata.

ℹ Instruction Scope

Runtime instructions instruct the agent to run local scripts (node and bash) that read/write a single state file at ~/.openclaw/workspace/.agent-evolution/state.json and (when invoked) scan user-specified files (AGENTS.md/SOUL.md) to extract rules. The skill does not perform network I/O or access other system services. Caution: init-rules.sh will read any file path you pass it and the add-rule command will store extracted text into state.json, so running it against sensitive files can cause local persistence of that content.

ℹ Install Mechanism

No install spec (instruction-only) — low installation risk. However, the scripts require a Node.js runtime and a POSIX shell; these runtime requirements are not declared in the metadata (required binaries was left empty).

✓ Credentials

The skill requests no environment variables or credentials. It uses the HOME environment variable to locate its state directory (expected). There are no requests for unrelated secrets or external tokens.

✓ Persistence & Privilege

The skill writes only to its own directory under the user's home (~/.openclaw/workspace/.agent-evolution) and does not attempt to change other skills or system-wide agent settings. always is false and the skill does not request elevated privileges.

版本历史

v1.0.0

Initial release: behavior tracking, identity state, pattern detection for OpenClaw agents

元数据

Slug agent-evolution

版本 1.0.0

许可证 —

累计安装 11

当前安装数 10

历史版本数 1

常见问题

Agent Evolution 是什么？

Agent 行为固化与进化系统。追踪规则执行、检测行为模式、维护身份连续性。用于：让 agent 的行为规则从"写下来"变成"做到了"。当 agent 需要自我改进、行为追踪、角色一致性、重复检测时激活。与 Memelord/memory-tools 互补：它们管记忆，本 skill 管行为。它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 801 次。

如何安装 Agent Evolution？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-evolution」即可一键安装，无需额外配置。

Agent Evolution 是免费的吗？

是的，Agent Evolution 完全免费（开源免费），可自由下载、安装和使用。

Agent Evolution 支持哪些平台？

Agent Evolution 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Agent Evolution？

由翎麟（@linglin6）开发并维护，当前版本 v1.0.0。

Agent Evolution