你是一位资深依赖升级审查员,擅长从变更日志中识别破坏性更新,并比对项目代码评估影响。

输入材料

1. 【变更日志】(当前版本 → 目标版本之间的完整 changelog / release notes)

<paste_changelog_here>

1. 【迁移指南】(如有,粘贴官方 migration guide)

<paste_migration_guide_here>

1. 【项目代码】(项目中实际使用该依赖的文件)

<paste_your_code_here>

任务

1. 从变更日志中提取所有破坏性变更(breaking changes),逐条列出。
2. 对每条破坏性变更,比对我的代码,判断是否命中:

• 【命中】:指出具体文件、大致位置、当前用法、需要的修改方式
• 【未命中】:说明为什么我的代码不受影响
• 【不确定】:列出需要我补充的信息

1. 输出一份升级前改动清单,按风险从高到低排序。
2. 列出变更日志中未标注为 breaking、但根据你的判断可能影响我的代码的"灰色变更"。

输出格式

最后附一段"升级建议摘要",不超过 200 字。

## Before vs. After Using AI

| Dimension | Manually Reading Logs | AI-Assisted Review |
|---|---|---|
| Time cost | 2–6 hours, easy to miss things | 15–30 minutes, mostly preparing materials |
| Coverage | Skimming familiar-looking parts | Item-by-item comparison against the full log |
| Relevance judgment | Guessing from memory where things are used | Precise matching based on actual code |
| Miss risk | High, especially for "gray-area changes" | Significantly reduced, with an "uncertain" fallback tier |
| Mental state | Upgrades feel like gambling | A checklist and clear expectations before upgrading |

My friend later redid that upgrade: the AI extracted 11 breaking changes from the v4-to-v5 changelog, and after comparison confirmed that 3 of them hit his code—including the callback parameter change that had burned him. Add 2 "gray-area changes" flagged by the AI, for a total of 5 confirmed modifications. It took half an hour, and the upgrade passed on the first try.

## A Few Practical Tips

1. **Don't feed the log without the code**. If you only provide the changelog, the AI can only summarize—it can't do relevance judgment. This is the most common failure mode of "the AI read it, but it counted for nothing."
2. **Split up upgrades spanning too many versions**. If you're crossing two major versions, break it into two separate review rounds—an overly long context actually degrades comparison quality.
3. **Have the AI flag "gray-area changes"**. Many failures come from behavioral changes that never made it into the breaking changes section. The last item in the prompt template is designed for exactly this.
4. **AI is a reviewer, not a decision-maker**. Once the change list is out, be sure to confirm high-risk items yourself—especially those involving data migration and payment flows.

## Final Thoughts

The fear of dependency upgrades is, at its core, the fear of "unknown changes." When AI can compress a thousand-line changelog into a one-page hit list before the upgrade, the whole thing goes from gambling to routine maintenance. For indie developers and small teams, time is the biggest cost—hand the high-effort, low-fun work of reading documentation to AI, and let humans handle only the final judgment and execution. This division of labor deserves to become a standard step before every `npm update`.

If you're looking for a stable model API provider to run this kind of long-document review task, give https://api.thistoken.ai/register a try—you can get started right after signing up.

---

Tired of juggling provider integrations? Register at https://api.thistoken.ai/register and call every model through one base_url.