Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Hermes AGI指令迭代与执行监督

v1.2.0

AGI指令迭代与执行监督工具。解决AGI"假大空、不落地"问题。输入模糊指令 → 拆解为可执行任务 → 设定奖惩机制 → 执行校验 → 不合格指令重写,形成闭环。触发词:hermes指令、AGI监督、指令迭代、任务拆解、指令优化。

0· 68·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for mengxiangshu666/hermes-agi-supervisor.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Hermes AGI指令迭代与执行监督" (mengxiangshu666/hermes-agi-supervisor) from ClawHub.
Skill page: https://clawhub.ai/mengxiangshu666/hermes-agi-supervisor
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install hermes-agi-supervisor

ClawHub CLI

Package manager switcher

npx clawhub@latest install hermes-agi-supervisor
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
The skill claims to provide instruction decomposition, scoring, verification and rewriting; the included Python and shell scripts implement those features and store results under ~/.hermes, which is coherent. However, decompose.sh invokes node -e to update metadata even though the registry metadata lists no required binaries — Node.js is an undeclared runtime dependency and is disproportionate to what the skill description explicitly stated.
Instruction Scope
SKILL.md directs the agent to run local scripts (decompose.sh, score.sh) and describes writing/reading task JSON under ~/.hermes/tasks. The instructions do not ask the agent to read unrelated system files or exfiltrate data, and there are no external network endpoints referenced in the docs or scripts.
Install Mechanism
There is no install spec (instruction-only), which is lower risk. But the package includes executable scripts that will create and modify files under the user's home directory (~/.hermes). That persistent file-writing behavior is expected for this tool but should be considered by users before installation.
Credentials
The skill declares no required environment variables or credentials and the code does not access network endpoints or secret-bearing files. This is proportionate to the described functionality. The only surprising environment reliance is on $HOME and an implicit Node.js runtime.
Persistence & Privilege
The skill persistently writes task, index, counter, and metadata JSON files into ~/.hermes. It does not request 'always: true' or attempt to modify other skills or system-wide settings. Persistent storage in the user's home directory is expected for a task-tracking supervisor but is a nontrivial level of presence the user should accept explicitly.
What to consider before installing
This skill largely does what its description promises (breaking fuzzy instructions into subtasks, scoring, and re-writing), but before installing you should: 1) verify whether Node.js is available on the target environment — decompose.sh runs node -e but the registry metadata does not declare Node as a requirement; lack of Node could lead to failures or surprising behavior; 2) be aware the skill writes persistent files under ~/.hermes (tasks, index, counter.json, metadata.json) — inspect those files and their contents if privacy or disk writes are a concern; 3) review the included scripts (decompose.sh, hermes_supervisor.py, score.sh) yourself or run them in a sandbox/container — they don't call external networks, but they will execute code locally and modify the home directory; 4) note templates.json contains Python-like content rather than pure JSON (likely redundant with the Python template DB) — ask the author for clarification or a trusted source/homepage before trusting the package; and 5) if you want to allow autonomous agent invocation, consider restricting it initially and monitoring what the agent executes (or run the skill only on demand). If you need a safer test, run the scripts in an isolated environment (temporary user or VM) and confirm behavior before enabling in production.

Like a lobster shell, security has layers — review code before you run it.

latestvk9762d8hq6wezph1xqcmmrq6eh84tmzg
68downloads
0stars
3versions
Updated 1w ago
v1.2.0
MIT-0

Hermes AGI 指令迭代 & 执行监督 Skill

核心定位

解决AGI系统的三大顽疾:

  • 指令模糊 → 不可执行
  • 缺乏校验 → 不了了之
  • 无奖惩机制 → 没有反馈

本skill将任何模糊的AGI指令,变成可测量、可执行、可迭代的闭环系统。


目录结构

hermes-agi-supervisor/
├── SKILL.md          # 主说明文件(本文件)
├── scripts/
│   ├── decompose.sh  # 任务分解脚本
│   └── score.sh      # 积分计算脚本
└── _meta.json        # 元数据

快速开始

用户输入任务指令后,按以下四步执行:

第一步:任务分解(AI执行 + 脚本记录)

用户输入:

我要:提升客户满意度

AI执行完整分解,输出结构化任务清单后,使用脚本记录:

# 在终端执行(由AI自动调用)
./scripts/decompose.sh "提升客户满意度"

生成结构化任务文件,记录到 ~/.hermes/tasks/


四大模块

模块一:指令拆解(Decompose)

输入: 一条模糊的AGI指令 输出: 拆解为3-7条具体、可测量、可验证的子任务

拆解标准:

  • 每条子任务必须可以回答"做到了"或"没做到"
  • 每条子任务必须有明确的完成标志
  • 每条子任务必须有时限或触发条件
输入示例:"提升客户满意度"
↓ 拆解输出
① 提取最近30条客户评价(完成标志:30条数据)
② 识别负向关键词频率TOP5(完成标志:词频表)
③ 生成改进建议清单(完成标志:清单≥5条)
④ 制定改进优先级(完成标志:带理由的排序)
⑤ 输出可执行的SOP(完成标志:含时间节点)

模块二:奖惩机制设定(Incentive Design)

输入: 拆解后的子任务列表 输出: 每条任务的奖惩规则(积分制)

奖惩规则:

  • 按时完成 +10分,超时1次 -5分
  • 完成质量达到标准 +20分,被打回重做 -10分
  • 连续3次达标触发"正向加速"(任务难度升级)
  • 连续2次未达标触发"强制降级"(简化任务粒度)

输出格式:

| 子任务 | 完成标志 | 奖励 | 惩罚 | 状态 |
|--------|----------|------|------|------|
| 任务1  |          |      |      | 待执行 |

模块三:执行校验(Verification)

输入: 子任务完成结果 输出: 合格/不合格/需重写 三个判定

判定规则:

  • 合格: 完成标志全部达成,内容无明显错误
  • 不合格: 完成标志部分达成,或存在事实性错误
  • 需重写: 完成标志全部达成,但不符合质量标准(需调整表述/结构/深度)

积分计算(自动):

./scripts/score.sh <task_id> <完成数> <失败数>
# 输出:{"task_id": "xxx", "base_score": 15, "bonus": 10, "total": 25, "status": "good"}

特殊处理:

  • 如果任务结果为"我不会/无法完成" → 触发指令降级
  • 如果任务结果为"已完成,但不确定质量" → 触发自检流程

模块四:指令重写(Rewrite)

输入: 不合格或需重写的任务结果 + 原始指令 输出: 优化后的新指令 + 改进说明

重写策略:

  • 增加具体限定词(时间/数量/标准/格式)
  • 拆解复合指令为串行步骤
  • 移除主观性词汇("做好一点"/"差不多就行")
  • 添加验证节点(中期检查点)
原始指令:"了解一下市场"
↓ 重写后
"在3个平台(知乎/抖音/得到)搜索'AI工具'关键词,
各抓取20条高赞内容,整理为包含:标题/平台/点赞数/核心观点
的Excel表,于24小时内完成。"

闭环流程

用户输入模糊指令
    ↓
【模块一】拆解 → 子任务清单
    ↓
【模块二】设定奖惩 → 积分表
    ↓
用户/系统执行子任务
    ↓
【模块三】校验结果
    ↓
    合格?→ 下一任务
    不合格?→ 【模块四】重写 → 重新执行
    ↓
全部完成 → 输出总结报告(含积分累计)

使用方式

标准流程(完整四步): 直接输入你的AGI指令,格式示例:

我要:提升团队执行力

单独调用某一模块:

  • "拆解 + 你的指令" → 只执行模块一
  • "校验 + 任务结果" → 只执行模块三
  • "重写 + 模糊指令" → 只执行模块四

脚本说明

脚本作用调用方式
decompose.sh创建任务文件,记录原始指令bash scripts/decompose.sh "<指令>"
score.sh计算任务积分和状态bash scripts/score.sh <task_id> <完成数> <失败数>

任务文件保存在 ~/.hermes/tasks/ 目录,支持历史查询。


注意事项

  • 本skill不替代执行,只负责监督和优化
  • 奖惩积分系统需用户自行维护(可用笔记或表格)
  • 连续3次无法完成的子任务,建议降级或拆分粒度
  • 任务文件采用JSON格式存储,支持程序化读取和分析

Comments

Loading comments...