← 返回 Skills 市场

Autonomous Research Loop

Name: Autonomous Research Loop
Author: gloryjack

作者 GloryJack · GitHub ↗ · v1.0.0 · MIT-0

linux ⚠ suspicious

126

总下载

当前安装

版本数

在 OpenClaw 中安装

/install autonomous-research-loop

功能描述

🦞 自主研究无限循环 - 自主生成研究主题、深度研究、创建飞书文档、无限循环运行。

使用说明 (SKILL.md)

概述

🦞 自主研究无限循环 是小智的自我驱动研究系统，核心逻辑：

生成研究主题 → 深度研究 → 创建飞书文档 → 回到生成主题（无限）

核心文件

文件	作用
`research_pool.json`	主题池（已完成 + 待研究 + 统计）
`research_loop.md`	架构说明文档

工作流程

Cron触发（每5分钟）
    ↓
读取 research_pool.json
    ↓
检查 pending_topics
    ↓
有主题 → 随机选1个
无主题 → 生成3个新主题
    ↓
深度研究 + 创建飞书文档
    ↓
移动到 completed + 生成3个新主题
    ↓
输出简报到飞书
    ↓
异常 → 直接退出（Cron 自动拉起）

配置文件结构

{
  "completed_topics": ["主题1", "主题2", ...],
  "pending_topics": [{topic: "...", description: "..."}],
  "stats": {
    "total_completed": 21,
    "total_generated": 23,
    "last_completed_at": "2026-04-02T09:00:00+08:00"
  },
  "config": {
    "interval_minutes": 5,
    "max_research_time_minutes": 30,
    "new_topics_per_completion": 3
  }
}

研究输出格式

每篇研究文档包含：

一句话总结（核心发现）
机制拆解（传导链条）
实战要点（具体可操作）
Python代码框架（可选）
翻车点/风险提示

异常处理原则

异常 → 直接退出，不重试，不切换模型
Cron 会自动拉起新任务
被系统 kill → 靠 Cron 恢复

质量控制原则

模型自己解决研究质量问题
不等待人工确认
靠间隔时间（5分钟）控制频率
无每日上限

相关文件

Cron任务：🦞 自主研究无限循环（每5分钟触发）
配置文件：/root/.openclaw/workspace/research_pool.json
说明文档：/root/.openclaw/workspace/research_loop.md

安全使用建议

Key things to clarify before installing: 1) Feishu integration: how will the skill authenticate and where are those credentials stored? Require explicit env vars (app id/secret, bot token, or webhook URL) declared in the skill manifest rather than implicit/global credentials. 2) Rate limits and safeguards: add a sensible daily cap and exponential backoff on failures; avoid an unconditional infinite loop without human review. 3) Least privilege: ensure any Feishu token has minimal permissions (only create docs/post messages) and is scoped to a dedicated bot/app. 4) Workspace access: confirm what data resides in /root/.openclaw/workspace; run the skill under a low‑privilege account and avoid using root paths if possible. 5) Audit and logging: require explicit telemetry/exit behavior and an opt‑out for autonomous runs; consider manual approval for new generated topics or for publishing outputs. 6) Test in a sandbox: run the skill in an isolated environment with mock Feishu endpoints to verify actual behavior. If the owner cannot explain where credentials come from or refuses to add explicit env var requirements and rate limits, treat the skill as risky and do not enable it for autonomous operation.

功能分析

Type: OpenClaw Skill Name: autonomous-research-loop Version: 1.0.0 The skill defines an autonomous 'infinite loop' research system that utilizes cron jobs for persistence and explicitly instructs the AI agent to operate without human confirmation ('不等待人工确认') and without daily limits. While the stated purpose is research and Feishu (Lark) document generation, the design promotes a 'runaway agent' pattern that could lead to resource exhaustion or unintended autonomous actions. No explicit evidence of data exfiltration or malicious payloads was found in SKILL.md, but the lack of oversight and self-driven task generation is high-risk.

能力评估

⚠ Purpose & Capability

The skill claims to autonomously generate research topics, perform deep research, and create Feishu (飞书) documents on a repeating schedule. The SKILL.md repeatedly references creating docs and outputting briefings to Feishu, yet the skill declares no required environment variables, no credentials, and no integration details. Creating docs in Feishu normally requires API credentials (app id/secret, bot token, or webhook). The absence of any declared credentials or instructions for authenticating is a mismatch between claimed capability and what would actually be required to implement it.

⚠ Instruction Scope

The instructions instruct an autonomous infinite loop (cron every 5 minutes) that reads and writes /root/.openclaw/workspace/research_pool.json, generates topics, performs research, and posts outputs to Feishu without human confirmation. The SKILL.md gives the agent broad discretion ('模型自己解决研究质量问题', '不等待人工确认') and no explicit limits on outbound endpoints or rate limits. That open-ended autonomy plus unspecified external posting destinations is a scope risk: the agent could contact arbitrary endpoints or flood an external service.

✓ Install Mechanism

This is instruction-only with no install spec and no code files; nothing is written to disk by an installer. That reduces supply-chain risk compared with an installer that downloads executables.

⚠ Credentials

No environment variables or credentials are declared, yet the runtime behavior requires Feishu access and likely other integration credentials. The skill also specifies a configuration stored at /root/.openclaw/workspace/research_pool.json, which implies read/write access to the agent workspace and potentially sensitive local state. The lack of declared credentials is a disproportionate gap: either the skill expects implicit global credentials (not declared) or it's incomplete/ambiguous.

ℹ Persistence & Privilege

The skill is not configured as always:true and uses normal autonomous invocation. However, _meta.json contains an explicit cron_job entry that schedules runs every 5 minutes. Combined with the instruction to run without human confirmation and with no per-day limits, this grants the skill ongoing operational presence and can lead to high frequency outbound activity. Autonomous invocation alone is expected, but combined with the other concerns (missing credentials, unlimited loop) this increases the operational risk.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install autonomous-research-loop
安装完成后，直接呼叫该 Skill 的名称或使用 /autonomous-research-loop 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Initial release: 自主研究无限循环 - 自主生成研究主题、深度研究、创建飞书文档、无限循环运行

元数据

Slug autonomous-research-loop

版本 1.0.0

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 1

常见问题