daivehhm

Agent Engineering Toolkit

Name: Agent Engineering Toolkit
Author: daivehhm

AI community

SOURCE https://github.com/daivehhm/Agent-Engineering-Toolkit

Description

Agent Engineering Toolkit 是一套面向本地 AI 编程 Agent 的轻量工程执行标准与配置工具包。它通过少量全局不变量、按需 Skills、项目级工程合同和独立 Review 机制，提高 Codex / Claude Code / Antigravity 在复杂软件开发中的执行一致性、证据真实性和一次完成率。它不是 Agent 平台，也不负责安装模型、管理账号或替代项目本身的设计。

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

Agent Engineering Toolkit v1.5

Agent Engineering Toolkit（AET）是一套面向 Codex / Claude Code / Antigravity 等本地 AI Coding Agent 的轻量工程执行、证据约束与独立 Review Toolkit。

Agent Engineering Toolkit (AET) is a lightweight engineering execution, evidence-control, and independent-review toolkit for local AI coding Agents such as Codex, Claude Code, and Antigravity.

核心问题不是“Agent 会不会写代码”，而是：

The core question is not whether an Agent can write code, but whether:

执行是否真实
改动是否受控
证据是否可信
完成声明是否可验证
不同 Agent 是否遵守同一项目合同

execution is real
changes are controlled
evidence is trustworthy
completion claims are verifiable
different Agents follow the same project contract

通俗理解 / Plain Explanation

这东西是给那些“雇了 AI 程序员给自己打工”的人用的。比如，你是一个程序员，现在你让 Claude、Codex 这些 AI 来帮你写代码、改代码。这时候，你就是老板，AI 就是你的“码农实习生”。这个工具包，就是给老板（程序员）和 AI 实习生之间定的一套“工作纪律手册”。

This is for people who have effectively hired AI programmers to work for them. For example, if you are a developer asking Claude, Codex, or another coding Agent to write and modify code, you are the boss and the AI is your coding intern. This toolkit is the work-discipline manual between the developer and that AI intern.

AET 主要解决三个烦心事：AI 给你“造假账”，说测试通过但实际没跑；AI 给你“乱拆家”，修一个小 bug 却偷偷改了底层结构；AI “自己说了算”，改完只说搞定了，过程不可查。

AET mainly addresses three pain points: the AI fakes the books by claiming tests passed without running them; the AI changes the house by touching architecture or unrelated code while fixing a small bug; and the AI acts as its own judge by saying “done” without leaving a traceable process.

所以，AET 要求 AI 拿出“实锤证据”，例如 exit code、运行记录和测试结果；要求 AI 遵守“施工红线”，只改允许改的地方；还要求 AI 留下“流水账”，让关键操作、失败、重跑和结论都能被复核。

That is why AET requires hard evidence such as exit codes, run logs, and test results; clear work boundaries so the Agent touches only what is allowed; and an audit trail so key commands, failures, reruns, and conclusions can be reviewed.

总结成一句话：这个工具包，就是让你当 AI 的“严苛监工”。它确保你雇的 AI 程序员不撒谎、不乱改、不留坑，所有操作都有据可查。

In one sentence: this toolkit helps