# embedded-software-engineer-agent: A Controlled and Auditable Rule Base for Embedded Software Development Agents

> A rule repository for embedded software engineer agents targeting Codex, ChatGPT, and Claude. Through strict prompt constraints and workflow design, it ensures AI does not generate arbitrary code when information is incomplete, but instead advances development in a modular and verifiable manner.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-04-20T16:14:26.000Z
- 最近活动: 2026-04-20T16:25:12.821Z
- 热度: 143.8
- 关键词: 嵌入式开发, AI Agent, Codex, ChatGPT, Claude, MCU, STM32, 驱动开发, 工程规范
- 页面链接: https://www.zingnex.cn/en/forum/thread/embedded-software-engineer-agent-agent
- Canonical: https://www.zingnex.cn/forum/thread/embedded-software-engineer-agent-agent
- Markdown 来源: floors_fallback

---

## Embedded Software Development Agent Rule Base: A Controlled and Auditable AI-Assisted Solution

embedded-software-engineer-agent is a rule repository for embedded software engineer agents targeting Codex, ChatGPT, and Claude. It aims to solve the problems of insufficient accuracy and easy generation of hallucinated code in AI-assisted embedded development. The core concepts of the project are **Controlled Development** and **Audit Priority**. Through strict prompt constraints and workflow design, AI is allowed to advance development in stages and in a verifiable way instead of generating a complete project at once, ensuring no arbitrary code is written when information is incomplete.

## Core Challenges of AI-Assisted Embedded Development

Embedded development has extremely high requirements for precision. Register configuration errors may cause hardware damage, and timing issues can easily lead to system instability. AI tends to 'hallucinate' wrong code when information is incomplete. Traditional AI code generation tools often output a complete project at once, lacking a verification link, making it difficult to meet production-level development needs. These problems have prompted the project to explore a controlled and auditable AI-assisted development model.

## Core Methods of the Project: Rule System and Workflow Design

The project provides four types of core deliverables to build a complete rule system:
1. **Prompt**: Clarify AI's role, boundaries, and forbidden behaviors (e.g., prohibit writing code when information is incomplete);
2. **Workflow**: Split development into seven stages: data review, key questioning, module breakdown, single module development/verification, system integration, regression testing;
3. **Templates**: Structured templates (such as project information collection, hardware review, module plan, etc.) ensure traceable output;
4. **Examples**: Real-scenario examples (e.g., STM32 peripheral development) demonstrate the application of rules.

## Applicable Scenarios, Usage Methods, and Minimal Example

**Applicable Scenarios**: V1 supports single MCU/SoC projects (e.g., STM32), common peripherals (UART/SPI/I2C, etc.), and bare-metal/lightweight RTOS; it does not support Linux BSP, complex GUI, multi-core heterogeneous, and other scenarios.
**Usage Methods**:
- Most reliable: Provide the entire repository + project information;
- Lightest: Copy system.md + rules.md + output_contract.md;
- Most engineering-oriented: Continuously fill in templates to accumulate team documents.
**Minimal Example**: After the user provides STM32F407 information and requirements, the Agent first reviews the information, raises key questions, breaks down modules, and then develops and verifies in stages.

## Comparison with Traditional Tools and Summary of Project Value

| Dimension | Traditional AI Code Generator | embedded-software-engineer-agent |
|-----------|-------------------------------|----------------------------------|
| Development Mode | One-time generation | Phased, verifiable |
| Information Requirement | Optional | Must be reviewed |
| Code Source | AI 'imagination' | Official priority |
| Quality Assurance | None | Single module verification + integration testing |
| Auditability | Low | High (complete template records) |
| Applicable Scenario | Prototype verification | Production-level development |

Project Value: Position AI as a 'technical audit + module development assistant', leveraging AI's information processing capabilities while retaining human engineers' decision-making power. It provides predictable and verifiable development assistance, suitable for production-grade embedded projects.

## Open Source Collaboration Philosophy and Engineering Recommendations

**Collaboration Guidelines**: Focus on rules, templates, and examples; new rules should be merged into existing files first; submit changes with an explanation of the engineering problem solved; adhere to the principles of 'official priority, stage control, verification loop'.
**Engineering Recommendations**: Recommend adopting the most engineering-oriented mode. Continuously fill in template documents during project advancement to transform Agent output into reusable engineering assets for the team, ensuring the development process is traceable and transferable.