I tried an Alibaba Cloud documentation Skill with Gemini CLI

[yanji-lxz]

I’ve been using Gemini CLI for some Alibaba Cloud coding and troubleshooting tasks.

One problem I kept running into was documentation accuracy. Gemini could usually produce a reasonable answer, but details such as endpoints, supported regions, available models, SDK parameters, and quotas were not always current.

There is also an additional complication: Alibaba Cloud has separate China and International documentation sites. An answer can look correct while referencing the wrong site.

I recently tried an Agent Skill called alibabacloud-help-doc-search to address this.

Installation

I installed it with:

npx skills add aliyun/alibabacloud-aiops-skills \
  --skill alibabacloud-help-doc-search

The Skill is available here:

https://github.com/aliyun/alibabacloud-aiops-skills/tree/master/skills/others/cseskillshub/alibabacloud-help-doc-search

How I use it

For Alibaba Cloud International, I explicitly include the site and language in the prompt:

Use alibabacloud-help-doc-search with site=intl and lang=en.

Find the current Model Studio OpenAI-compatible endpoint,
API key setup, and supported regions.

Answer with official documentation sources.

I also tried it with an OSS question:

Use alibabacloud-help-doc-search with site=intl and lang=en.

Find the current OSS bucket policy documentation and explain how
to grant read-only access to a specific bucket.

Cite the official sources.

The important part is:

site=intl
lang=en

Without that scope, it is easy for an agent to mix China-site and International-site documentation.

For China-site documentation, the default is:

site=cn
lang=zh

What I found useful

The main benefit was not simply getting an answer. It was getting the corresponding documentation URL and knowing which site and language the answer came from.

The Skill combines:

1. regular full-text documentation search;
2. product-level llms.txt index search;
3. reading the selected documentation page.

A result can include information like:

{
  "product": "Object Storage Service",
  "title": "Bucket Policy",
  "url": "https://www.alibabacloud.com/help/en/oss/user-guide/oss-bucket-policy/",
  "source": "both",
  "site": "intl",
  "lang": "en"
}

Here, source=both means the page was found through both full-text search and the llms.txt index.

This made it easier for me to verify the answer instead of relying only on Gemini’s existing knowledge.

A few observations

It also works better when the prompt is specific. For example, I got better results when I included:

• the product name;
• China or International;
• the documentation language;
• the exact fact I wanted to verify;
• a request for official citations.

A prompt like this worked better than a generic question:

Use the Alibaba Cloud International documentation.

Find the current OpenAI-compatible endpoint for Model Studio,
confirm the supported regions, and cite the official sources.

Final thoughts

For me, the most useful part is not that Gemini CLI can find another documentation page. It is that I can verify which documentation site the answer came from and check the official source before using the result in code.

This is particularly useful for Alibaba Cloud questions involving:

• endpoints and supported regions;
• Model Studio APIs and available models;
• ECS security groups;
• OSS bucket policies;
• RAM permissions;
• SDK parameters;
• quotas, pricing, and error codes;
• differences between the China and International sites.

If you use Gemini CLI for Alibaba Cloud development or troubleshooting, this Skill may be worth trying:

npx skills add aliyun/alibabacloud-aiops-skills \
  --skill alibabacloud-help-doc-search

I’d be interested to hear whether it improves documentation accuracy in your workflow.