Model or dataset
tingaicompass/AI-Compass avatar
tingaicompass/AI-Compass

AI-Compass tells your coding agent to fetch install instructions from a raw URL

AI-Compass guides the community through the ocean of AI technology, helping both beginners and advanced developers find a path into any major AI direction. It aims to help developers systematically understand core AI concepts, mainstream techniques and cutting-edge trends, and master the whole process from theory to practice.

997 stars136 forksPythonLicense varies

At a glance

What is it?
A Chinese-language index of AI resources that doubles as an installable knowledge-base skill, where the mechanics worth reading are the unpinned install URL, directory numbers reused across the tree, and a weekly archive whose filenames carry no dates.
Who is it for?
This is a link index with an install path attached, and the two halves have different rules. As a curated map of AI tooling it is unusually broad and unusually honest about what it does not have yet, since two of its nine modules are described as partial work in progress.
Can I use it commercially?
Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
Is it still maintained?
Yes. The repository last received commits 9 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • tingaicompass/AI-Compass997 ★ · +56
Star history
Repository2026-09-102026-11-15
tingaicompass/AI-Compass941997

56 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The install instructions are a raw URL with no version in it

The documented way to install this repository is to hand one sentence to a coding agent:

text
如果本机还没有 AI-Compass,请先 clone https://github.com/tingaicompass/AI-Compass.git,然后按照 https://raw.githubusercontent.com/tingaicompass/AI-Compass/main/Install.md 把它安装为我本机 Codex 和 Claude Code 都能使用的本地知识库 Skill。完成后告诉我安装路径、验证结果,以及是否需要重启客户端。

That sentence names four hosts it is meant to work with, Claude Code, Codex, Cursor and Windsurf, and it splits the install into three fetches. Clone the repository, then retrieve a second document from `raw.githubusercontent.com` at the path `main/Install.md`, then run the install script that sits in the repository root. The agent is told to verify the symlinks it created and to report back with the install path, the verification result and whether the client needs restarting.

What is missing from that chain is a version. The URL points at the `main` branch, the repository has no GitHub releases, and there is no tag or commit referenced anywhere in the instruction. Two runs of the same sentence, a week apart, are therefore two runs of two different instruction sets, and the only way to tell them apart is to keep your own copy of Install.md.

The agent-facing surface is visible in the tree. `AGENTS.md`, `SKILL.md`, `agents/` and `install.sh` sit at the root next to the two READMEs, which is the shape of a repository that expects to be read by a tool rather than only by a person.

One install serves two clients, and symlinks are the check

The target is a single local knowledge base that both Codex and Claude Code can read, rather than a per-client install. That is what makes the symlink step load-bearing. A skill directory that two different clients reach through their own paths is normally a pair of links pointing at one copy of the content, and the verification the prompt asks for is a check that those links resolve.

The consequence for an operator is that an install touches directories belonging to two separate tools. The prompt itself acknowledges the boundary by asking whether a restart is needed, which implies the client reads its skill list at startup rather than on every request.

The path of least resistance here is delegation. An autonomous agent is asked to clone a repository it has never seen, fetch a second document whose contents are not visible at the time the instruction is written, and execute a script from that repository. The instruction says the script installs a skill; it does not say what the script writes, what it links, or what it changes outside the repository.

None of that makes the repository unusual. It makes the review step yours. Reading Install.md and install.sh before pasting the sentence costs a minute, and it is the only place where the actual write set is visible.

Four directory numbers are each used more than once

The tree is organised by numeric prefix, and the numbering is not unique. Four prefixes are reused. Two directories carry `6.`: `6.AI-LLM比赛/` and `6.AI课程/`. Six carry `9.`: `9.学习平台/`, `9.学术工具/`, `9.文章集/`, `9.社区论坛/`, `9.软件/` and `9.面试/`. Five carry `10.`: `10.Datawhale/`, `10.paddle/`, `10.华为开源/`, `10.腾讯/` and `10.阿里开源/`. Three carry `11.`: `11.Skills/`, `11.blog/` and `11.code/`.

That matters for anything that sorts by name. A directory listing ordered lexicographically interleaves `10.华为开源/` between `10.Datawhale/` and `10.paddle/`, and puts all six of the `9.` directories together while scattering the `9.` group against the other `9.`-prefixed siblings. There is no single ordering key in the tree itself, so any index built over it, including one an agent builds, has to reconstruct the intended order from somewhere else.

The module descriptions in the README do not line up one to one with these directories either. Nine modules are described, and the numbered tree has its own groupings, with `3.0` through `3.7` covering MCP and A2A, RAG with workflow, Agent, DeepSearch, GraphRAG, NLP to SQL, popular AI frameworks and AI robots as eight separate directories under one numeric stem.

The weekly archive is numbered but not dated, and already has two entry fives

The rolling section is the part a knowledge-base user is told to read first. `weeklyHighlights/latest.md` and `weeklyHighlights/INDEX.md` are named as the two entry points, and the guidance for adding a report is to update both while following `TEMPLATE.md` to fill in metadata.

Below that sit the past issues as a numbered list. The visible run reaches 30 entries, and the list ends partway through the link for entry 30, leaving it unterminated. The collision is more interesting: there are two files numbered 5, `5.趣味应用.md` and `5.md`, so the numbering already has a duplicate in the same way the directory prefixes do.

The naming convention carries no date at all. A file called `27.md` says nothing about when it was written, and the freshness of an entry has to come from the metadata block that TEMPLATE.md defines, which means the index is only as current as the discipline behind that template.

The content of the archive is where the nature of this repository shows. Entries are dated external events: model releases, new agent frameworks, image and video generation tools, speech models, open-sourced repositories. Each is a link with a title that lists several products at once, which is efficient for scanning and means the archive is a pointer layer rather than a body of written guidance.

Two of the nine modules are described as work in progress

The project presents nine core modules, and the descriptions carry different levels of certainty.

Two are explicitly partial. The blog module is described as holding systematic technical articles, currently covering Python basics, algorithm solutions and programming ability improvement, with LLM guides, interview tips and interview questions, and walkthroughs of enterprise logic projects all marked as coming later. The code module is described as runnable AI practice code and demos, currently containing RAG engineering samples, with more project code to follow so it can be debugged, reused, and used for code-level question answering and project breakdown.

The other seven are aggregation layers rather than authored content. The basics module is an entry point for navigation tools, prompt engineering, LLM benchmark leaderboards and model collections. The frameworks module lists embedding models, training frameworks, inference and deployment, evaluation frameworks and RLHF. The application module covers RAG with workflow, Agent, GraphRAG and MCP with A2A. Products and tools, learning resources, the enterprise open-source collection and the community module are all described as curated sets of links.

So the honest summary is one authored directory of language basics and one of RAG samples, wrapped in a wide map of everything else. The audience list widens the same way, naming beginners, developers, product managers, researchers, enterprise teams and job seekers in six groups, which is a lot of reading paths for a repository whose only runnable code is described as RAG samples.

The submission policy says no paid ranking, in the same file as a coupon

Resource submissions go through GitHub issues with a fixed template: a title line, tool name, official link, category, one-line summary, core features listed as three bullets, intended audience and scenario, whether it is open source, free or paid, supplementary links, and a submitter note.

The policy attached to that template is specific. Commercial tools may be submitted, but the submitter has to state pricing, free quota or trial terms. The project states that it does not do paid ranking, and that maintainers organise entries by content quality, availability and community value.

The surrounding document is where the reading gets interesting. The same README carries a link to a coupon for a paid knowledge community, described as a knowledge planet, pointing at `picture/minor/KnowledgePlanet.md`, and names a Coze space as the official site, with that URL carrying a task identifier query parameter. The repository's own homepage field points somewhere else entirely, at a personal CSDN blog with a tracking query string attached.

None of that contradicts the stated policy on its face, since a coupon for the author's own community is not a paid listing for a submitted tool. It does mean the line between curation and promotion is drawn by the maintainer rather than by anything a submitter can verify, and that the submission template's pricing field is the only place the distinction is made explicit.

The language field says Python and the tree is markdown

The repository's primary language is recorded as Python, and almost nothing in the top-level tree is Python. What is there is a set of numbered markdown directories, `install.sh`, `AGENTS.md`, `SKILL.md`, `agents/`, `picture/` and `weeklyHighlights/`, plus a committed `.DS_Store`, which is a macOS Finder artefact that has no business in version control.

The Python probably comes from the install script and whatever sits under `agents/`, but the point stands: anyone filtering this repository by language will see a Python project and find a link index. That is the same mismatch pattern as the package metadata problem in other AI-Compass-adjacent repositories, and it is the first thing to check before assuming a repository contains runnable code.

Distribution is similarly split across hosts. There is a Gitee mirror alongside the GitHub repository, `README.md` and `README-EN.md` as the two language versions, and profiles linked on Zhihu and Juejin. The README in hand is the Chinese one, and the English version exists as a separate file in the tree.

Two audiences are addressed in two places. Chinese readers get the full document, while the English file is linked from the same header line rather than from a translation service, so the two can drift.

No license file appears at the root of a repository that asks you to run its script

The repository metadata carries no license value, and no LICENSE file appears among the top-level entries, which are otherwise enumerated in fine detail, down to `picture/` and the individual numbered directories. The README does not state terms either.

That gap sits oddly next to what the project asks. Contributors are invited to submit resources through issues with a structured template, readers are asked to clone the repository and run a script from it, and the whole knowledge-base pitch depends on people pointing their agents at it. A curated link index under no license is a small problem. A repository whose root script writes into a coding agent's configuration directories, with no license and no release tag, is a larger one, because there is no stated basis on which to fork it, vendor it into an internal tool, or rely on it staying available.

What the project does state is its structure, which is the more useful part for anyone deciding whether to adopt it. Nine modules, a weekly archive with an index and a latest file, a submission process with a template, and an install path that produces a local skill for two clients. Everything an adopter needs to evaluate is in the README; what is missing is the legal and versioning layer that would make the install path safe to automate.

Editorial conclusion

This is a link index with an install path attached, and the two halves have different rules. As a curated map of AI tooling it is unusually broad and unusually honest about what it does not have yet, since two of its nine modules are described as partial work in progress. As an installable artifact it asks an autonomous coding agent to clone a repository, fetch a second file from a raw URL on the default branch and execute a script, with no release tag to pin any of it. Before running that prompt, read Install.md yourself and know what install.sh does. Note that the repository records no license and lists no license file, which matters if you plan to redistribute anything you build from it.

Frequently asked questions

What is tingaicompass AI-Compass?

It is a curated index of AI resources organised into nine modules, covering large language models, multimodal models, training frameworks, inference deployment, RAG, Agent and GraphRAG, plus learning platforms, interview material and enterprise open-source collections. The README in hand is the Chinese version, with an English one listed beside it as README-EN.md.

How do you install AI-Compass as a local knowledge base?

The documented route is a one-line instruction to a coding agent such as Claude Code, Codex, Cursor or Windsurf: clone the repository, follow the instructions at the raw Install.md URL on the main branch, and let the agent run the root install script and verify the symlinks. The agent is asked to report the install path, the verification result and whether the client needs a restart.

Does AI-Compass have versioned releases to pin?

No. The repository has no GitHub releases, and the install instructions point at Install.md on the main branch with no tag or commit in the URL, so the instructions an agent follows can change between runs.

What code actually ships in AI-Compass?

The code module is described as runnable AI practice code and demos, currently containing RAG engineering samples, with more project code promised later. The blog module is likewise described as currently holding Python basics, algorithm solutions and programming ability material, with LLM guides and interview content listed as future work.

How are resources submitted to AI-Compass, and is paid placement allowed?

Submissions go through a GitHub issue with a template covering the tool name, official link, category, summary, three core features, intended audience, whether it is open source, free or paid, and a submitter note. Commercial tools are accepted as long as pricing, free quota or trial terms are stated, and the project states that it does not do paid ranking.

What license does AI-Compass use?

None is stated. The repository metadata carries no license value, no LICENSE file appears among the top-level entries, and the README does not state terms, which matters for anyone planning to redistribute or vendor the repository or its install script.

Official sources

  1. Issues
  2. Project website
  3. README
  4. tingaicompass/AI-Compass on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/tingaicompass-ai-compass.svg)](https://hysenlabs.com/projects/tingaicompass-ai-compass)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/tingaicompass-ai-compass.svg)](https://hysenlabs.com/projects/tingaicompass-ai-compass)

Was this useful?