kangarooking-skills: a personal Agent Skills collection with 16 workflows and no declared licence
My custom AI Agent skills
At a glance
- What is it?
- A single repository holding sixteen Agent Skills for Claude Code, Codex and other agents that follow the Agent Skills standard. The workflows are real and specific; the packaging around them is not, and the licence is unstated.
- Who is it for?
- Adopt individual skill directories if you already run Claude Code or Codex and want a working starting point for a task like video transcription, title generation or project scaffolding. Do not clone the whole repository expecting a coherent product, and do not treat any single skill as supported software.
- 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 September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What problem this repository solves, and for whom
Most people who use Claude Code or Codex end up rewriting the same instructions for the same recurring jobs. A prompt that reliably produces thirty candidate headlines for a WeChat article is not the same prompt that reliably downloads a Douyin video and separates the platform caption from the spoken transcript. kangarooking-skills is one author's answer to that: sixteen directories, each holding a structured workflow an agent can read and follow. The README describes them as "Agent Skill" already run through in practice, collected and open sourced so others can take what they need.
The audience is narrow and specific. You need an agent that supports the Agent Skills open standard, which the README links to at agentskills.io, and the badges name Claude Code and Codex explicitly. If you are running a different agent framework, or you want a library you import into Python code, this is the wrong shape. The skills are not a package. They are directories of markdown plus optional scripts and reference material, and the agent does the work by reading them.
The directory layout is the interface
Every skill directory contains a SKILL.md, which the README calls the skill definition and execution flow. Complex skills add three optional subdirectories: references/ for material the agent loads on demand, scripts/ for automation that needs to behave the same way every time, and assets/ for templates and delivery resources. That split matters more than it looks. Anything in scripts/ is deterministic code the agent invokes; anything in references/ is context the agent reads and reasons over. A skill that puts its logic in prose is asking the model to re-derive the same procedure on every run.
The breadth across the sixteen directories is unusual. Image generation appears three times with different scopes: apimart-image-gen wraps a single APIMart call to GPT-Image-2, multi-agent-image coordinates design analysis and reference selection across several agents for series consistency, and cover-skill builds a personal visual signature from a creator's portrait and a kangaroo mascot before producing four cover directions and adapting them to five platform ratios. Content work covers topic discovery, title generation and full article production. There is a video downloader, a book illustration workflow, and two agent-engineering skills. The last entry, reshape-your-life, is a personal-direction exercise built on NLP logical levels, which tells you the author treats the repository as a personal toolbox rather than a product line.
Installation and where the files land
The README offers two paths. The conversational one is to tell your agent to install a specific skill by URL, substituting the directory name:
帮我安装这个 skill：https://github.com/kangarooking/kangarooking-skills/tree/main/<skill-name>
The manual path clones the repository and copies one directory into the agent's skill folder, with separate targets for Codex and Claude Code:
git clone https://github.com/kangarooking/kangarooking-skills.git mkdir -p ~/.codex/skills cp -R kangarooking-skills/<skill-name> ~/.codex/skills/ mkdir -p ~/.claude/skills cp -R kangarooking-skills/<skill-name> ~/.claude/skills/
The copy is per-skill, not whole-repository, and that is the right way to read it. If your agent does not support skill installation at all, the README suggests placing the SKILL.md and the resources it references into your project so the agent can follow the procedure from there. Note what is absent: no package manifest, no version pinning, no upgrade command. You are copying files, and you will be copying them again when you want the author's later changes.
What the skills actually call out to
Several skills are thin wrappers over third-party services, and the README is direct about which. apimart-image-gen reaches GPT-Image-2 through APIMart and handles generation, polling and download, with resolution options of 1k, 2k and 4k and support for URL and local reference images. hy-3d-gen produces 3D models through Tencent Hunyuan, with text-to-3D, image-to-3D and multi-view modes, PBR and white-model output, and a default route through a TokenHub OpenAI-compatible interface that can fall back to the Tencent Cloud SDK. scroll-promo-site-builder generates its visual sequence with image2.0 and Seedance 2.0, then emits a Vite plus React site with a mobile fallback and a local preview bundle.
Two details in that list deserve attention. First, apimart-image-gen reads its API key only from an environment variable and the README states it is never written into code or commit history. That is the correct pattern and worth preserving if you fork the skill. Second, scroll-promo-site-builder credits oso95/scroll-world for the scroll-video and boundary-frame method and the README says it retains a full MIT attribution notice. That is the only licence information anywhere in the material, and it applies to a third-party project the skill borrows from, not to this repository.
Where the repository is thin
The licence field is unknown. There is no LICENSE file described in the README, no licence badge, and no statement about terms. Under default copyright, a public repository without a licence grants nobody the right to copy, modify or redistribute it, and every install instruction above is a copy. For personal use on your own machine that distinction rarely bites. If you plan to fork a skill, ship it inside a product, or bundle it into a company-wide agent setup, it does. I am not giving legal advice here; the point is that the material provides no permission and no answer, so the question has to go to the author.
Beyond licensing, the material is silent on maintenance. There are no releases, no changelog and no stated compatibility range for the Agent Skills standard itself. The repository was pushed recently, so it is active, but activity is not the same as a support commitment. The README's contribution section asks for SKILL.md, references/, scripts/ and assets/, which describes structure but not review, testing or versioning. Treat each skill as a snapshot of one person's working setup rather than a maintained dependency.
A concrete alternative, and where the approaches diverge
The README itself points at kangarooking/cangjie-skill, a separate project by the same author that distills a book's methodology into triggerable, composable, testable AI Skills. The division of labour is stated plainly: cangjie-skill produces skills from books, this repository collects and open sources skills that already work.
That is a real difference in kind, not just in content. cangjie-skill is a generator with an emphasis on composability and testing; kangarooking-skills is a curated shelf of finished artifacts with no generation step and no test harness described. If your problem is "I have a methodology and I need it turned into skills," the generator is the closer fit. If your problem is "I need a working procedure for downloading a Bilibili video and transcribing it today," the shelf is. The trade-off is that a shelf ages: nothing in the repository tells you when a platform changes its page structure and a scraping skill stops returning results.
Who should take it, and what to check first
Take individual directories if you already run Claude Code or Codex, you have a recurring task that matches one of the sixteen, and you are willing to read the skill before running it. The video-downloader is a reasonable example of the shape: it covers Douyin, Bilibili, YouTube and Xiaohongshu, keeps the platform's original caption separate from the ASR transcript, and uses a WeChat mini program called kg百宝箱 as a download prerequisite for Channels video, which is a manual step the README acknowledges rather than hides.
Skip the repository if you need a licence grant, a versioned dependency, or a single coherent tool. Also skip it if you object to skills that depend on paid external APIs, because apimart-image-gen, hy-3d-gen and scroll-promo-site-builder all route through third-party services whose terms and costs live outside this repository.
The first thing to verify is not a feature. Open the SKILL.md of the one skill you care about and read every script it invokes, because that is where credentials, network calls and file writes will be. The second thing is the licence question, which the repository does not answer.
Editorial conclusion
Adopt individual skill directories if you already run Claude Code or Codex and want a working starting point for a task like video transcription, title generation or project scaffolding. Do not clone the whole repository expecting a coherent product, and do not treat any single skill as supported software. Before copying anything into ~/.claude/skills or ~/.codex/skills, open that skill's SKILL.md and read every script it calls, because the repository states no licence and several skills touch paid third-party APIs and platform credentials. If you need a licence grant, ask the author first; the repository itself provides none.
Community notes