zhaoolee/ins: a link directory kept honest by a daily GitHub Actions ping

Every entry starts as a row in website_info.csv

The whole directory is one file at the repository root, website_info.csv. The owner, zhaoolee, adds a data source to that file. After the commit reaches GitHub, a GitHub Action runs a crawler written in Python and checks the URL.

The repository root also holds main.py, markdownLog.py, tagit.sh, Pipfile, Pipfile.lock and EditREADME.md. Reading the names, main.py is the entry script and markdownLog.py is a good place to start if you want to see how the table is produced. The README table itself is generated output: its header line states the number of entries, 143 at the last update on 2026-10-30, along with a timestamp.

Because the table is rebuilt from the CSV, editing the README by hand would be overwritten. Anyone contributing is expected to touch the data file, and the author invites replacement suggestions through the repository issues.

A green or red light is appended to each name, plus a millisecond count

The status rule is short. If the crawler receives a response from the URL, a green light is appended to the Name. If not, the name gets a red light. For green entries, the response time in milliseconds follows, and the author reads smaller values as a faster site.

The table shows the spread. SpeedRun is listed at 51ms and Hugging Face at 48ms, ChatGPT at 46ms, while the command-line IP lookup at cip.cc sits at 3288ms. Microsoft's wallpaper design site appears with a red light and no timing at all.

The limit is plain. A single number comes from wherever the Action runs, which is a GitHub runner, not your connection. A site that answers in 46ms there can be slow or blocked from where you sit, and the README itself tells readers to look elsewhere for help reaching sites that fail to open.

SSL expiry dates ride along with each green row

Rows that respond also carry a line giving the SSL certificate expiry date and the days remaining. AppleGamingWiki is shown with an expiry of 2026-12-15, 46 days out. Hugging Face shows 2027-02-27, 120 days out, and jellyfin.org shows 2026-12-29.

This is a useful side effect for a directory. A certificate that lapses is often the first sign that a small site has been abandoned, and the count of days left gives an early warning before the light turns red.

The red-light row for the Microsoft wallpaper site has no such line, so the certificate check only appears to run on sites that answered. A reader who wants to know why something is red gets no reason: the table does not separate a DNS failure from a timeout or a blocked connection.

Tombstones: a red light is an invitation to propose a replacement

The author calls red-light entries tombstones. A project with a red light may have been good but could not keep going. If you can build a replacement, the instruction is to leave a message in the repository issues, and the project says it will include it.

This turns the directory from a static bookmark list into something with a lifecycle. Dead entries are left visible, not deleted, and the name of the mechanism is deliberate: the stated goal is to collect inspiration, track the sites that are alive and bring back the ones that have fallen.

The cost is that dead rows stay in the table and in the category counts until someone acts. Nothing in the repository describes automatic removal after some number of failed checks, so a long-red row can sit there indefinitely.

Twenty-three category tags, with two broken ones in the index

Each row carries semicolon-separated tags, and the table of contents links to them with counts. 信息流 has 25 entries, 在线工具 22, 极客 22, 资源集市 20, 游戏 12, 摄影 12, 每日常逛 12, 课程 11, 设计 10, bbs 9 and 手机 7. Smaller tags include 壁纸 at 5, AI at 3, 电子书 at 3 and 健身, 白噪音 and 评测 at 2 each.

Two index entries look like typing mistakes. One reads "游戏 with a stray quote and a count of 1, and another is a bare quote character, also with a count of 1. The SpeedRun row shows where this comes from: its tag cell is written with a leading space and quotes, as ' "游戏;信息流;在线工具;'.

That means tag counts are not fully reliable, and a filter built on the raw CSV tags will treat the quoted one as separate from 游戏.

The data comes as a raw CSV, a CDN copy, or a shallow clone

The README offers the full CSV in two forms: the raw GitHub URL at raw.githubusercontent.com/zhaoolee/ins/main/website_info.csv, and a CDN copy through cdn.jsdelivr.net/gh/zhaoolee/ins/website_info.csv. The CDN route is the one to use if the raw GitHub domain is slow for you.

For a fork that collects something other than inspiration, the author invites you to clone and modify it:

git clone --depth=1 https://github.com/zhaoolee/ins

The clone brings the crawler along with the data, so a themed fork needs only its own CSV rows and a schedule. The Action runs daily at 6 o'clock according to the README, and the repository has a .github directory for the workflow.

Two gaps matter before you reuse the data. The repository metadata lists the license as unknown, and no releases are published, so the CSV has no version tag you can pin.

Descriptions are short Chinese notes, and some entries point at questionable sites

Each description is a one-line Chinese gloss. AppleGamingWiki is described as collecting macOS-playable games and their compatibility. Stable Diffusion web UI is described as an offline-deployable web program that makes AI images from text or images. The NanoPi R4S entry is the owner's own firmware for a small proxy box.

The directory is a mix of tools, communities, wallpapers and games, and it does not vet them. Among the rows are Byrut, which the entry describes as a Russian game site with a very large collection of cracked games, and SMS Activate, a paid virtual phone number service. Listing a site is not an endorsement, but the table gives no warnings either.

English readers should expect machine translation. Nothing in the repository offers an English edition, and the long ChatGPT row links out to a Chinese-language registration guide on mirror.xyz.

Editorial conclusion

Take INS if you read Chinese and want a curated, tagged list of sites with a daily indication of which ones are still up. Skip it if you need English descriptions, a license you can rely on, or response times measured from your own network. Before reusing the data, open website_info.csv yourself, since the repository metadata shows no license and the tables are regenerated into the README on every run.

Frequently asked questions

Official sources

1. Issues
2. README
3. zhaoolee/ins on GitHub