How AutoDev’s New Team Prompts Boost AI‑Assisted Development and Code Review

Overview

AutoDev 1.4.0 adds Team Prompts , a mechanism that lets developers store prompt files in the code repository. AutoDev reads these files at runtime, making AI assistance configurable at the team level without requiring individual IDE extensions.

Team Prompts

Team Prompts enable shared, reusable AI instructions across a whole team or sub‑team. The prompts are ordinary files (e.g., .vm) placed in a designated folder and can reference AutoDev configuration variables such as the current file’s code.

Example TDD Workflow Prompts

Tasking.vm

– splits a requirement into test cases. TDD-Red.vm – generates the first failing test. TDD-Green.vm – writes or refines the implementation to satisfy the test. TDD-Refactor.vm – performs code refactoring.

Prompt File Structure

Each prompt begins with a YAML front‑matter that configures menu priority and interaction mode. Example:

---
priority: 2023
interaction: ChatPanel
---
```user
You are a senior software engineer proficient in TDD. Based on the new test case, improve the existing implementation.
```
Original code: $context.underTestFileCode($methodName)
New test code:
${selection}
Please optimize the class under test and return the updated method implementation.
```

The priority field controls ordering in the UI. The interaction field selects the output mode, e.g., ChatPanel for a side‑panel chat window or AppendCursorStream for a streaming, typewriter‑style insertion.

Living Documentation

AutoDev can generate or refresh documentation (class, method, variable) via a context menu or Alt+Enter. If documentation already exists, the tool attempts to improve it based on the latest code.

A JSON‑style prompt can be used to produce structured documentation such as Swagger annotations:

{
  "documentations": [
    {
      "title": "Living Documentation",
      "prompt": "Write Living Documentation. Return in the following format:",
      "start": "",
      "end": "",
      "type": "annotated",
      "example": {
        "question": "...",
        "answer": "..."
      }
    }
  ]
}

Code Review Integration

AutoDev can be invoked from CI/CD pipelines to perform AI‑assisted code review. It groups related commits (e.g., all commits for a feature) and reviews them together. When a commit message contains a requirement identifier, such as feat(devops): init first review command #8, AutoDev fetches the corresponding ticket information and supplies it as business context to the LLM, improving the relevance of review suggestions.

Repository

The project is open source. Issues, contributions, and releases are managed at:

https://github.com/unit-mesh/auto-dev