A directory of thousands of OpenClaw skills, counted four ways
🧠Curated collection of 1209+ best OpenClaw skills — weekly updated by MyClaw.ai
At a glance
- What is it?
- A weekly-synced mirror of the OpenClaw skill archive with a curated index, where the description, the badge, the section heading and the release notes each claim a different total.
- Who is it for?
- This repository is a supply chain convenience, and that framing sets expectations for everything else about it. The value is that 2,000-plus skill directories arrive in one pull and a skill index tells you which one you want; the cost is that quality filtering is asserted rather than demonstrated, since the release notes say only that a quality rule was applied.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 80 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 20, 2026, and from our analysis. They are not legal advice.
Editorial analysis
A mirror of an archive, refreshed weekly
The release notes are short and consistent. Version 0.21.0, published 2026-07-20, adds 100 skills for a total of 2,409. Version 0.18.0 from 2026-06-29 adds 100 for a total of 2,109. Version 0.17.0 from 2026-06-22 adds 100 for a total of 2,009. Each names its source as the official mirror of the openclaw skills archive and says the additions were selected against quality rules, with details delegated to a RELEASES.md file in the repository.
So the mechanism is clear: a periodic sync that pulls a batch, filters, and publishes as a tagged release. That is a different design from a curated collection where a human decides what belongs. The human decision is compressed into the phrase quality rules, which are not written down in the README.
The repository structure supports the mirror reading. There is a `skills/` directory holding the payloads, a `skill_index.md` for searching them, `RELEASES.md` for the running log, `CHANGELOG.md`, a `.clawhub/` directory for registry metadata, a `scripts/` directory for the sync, and a `SKILL.md` at the root. Eight README translations sit alongside the English one, covering Chinese, French, German, Russian, Japanese, Italian and Spanish.
GitHub reports the language as Python, which comes from whatever sits in `scripts/`. The skills themselves are instruction text, so the declared language describes the sync tooling rather than the content.
Installing means copying a directory into a workspace
Two installation routes are given, and both are one command. The first goes through ClawHub, the registry referenced in the sibling marketing repository:
clawhub install openclaw-master-skillsThe second clones the repository and copies a single named skill directory into the agent's workspace:
git clone https://github.com/LeoYeAI/openclaw-master-skills.git
cp -r openclaw-master-skills/skills/<skill-name> ~/.openclaw/workspace/skills/The second route is the more realistic one for a collection this size. Cloning 2,000-plus directories to install one of them is a heavy operation, and the copy step makes the intent clear: you take the subdirectory you want and place it where your agent looks for skills.
One inconsistency sits in the comment above the first command, which says to install a single skill via ClawHub while the command it labels installs the whole collection by repository name. Read the command rather than the comment.
The destination path, `~/.openclaw/workspace/skills/`, is the one piece of the system this repository defines for you. Everything else about how a skill is loaded, what file inside a directory matters, and what format it takes is not documented here, which means the copy is a directory copy and not a specific file placement. Assume you need to read the target agent's own documentation to confirm that is sufficient.
Four different totals for one collection
The numbers in this repository do not agree, and the disagreement spans five sources.
The GitHub description says 1209+ of the best OpenClaw skills. The README badge is written as Skills 2409+, and its own alternative text says 1211+ Skills, so even a single badge carries two different figures. The README's index section is headed Skill Index (561 skills). The release notes say 2,409, 2,109 and 2,009 across three tags. The README section beneath the badge repeats the description of a weekly-updated collection.
Only one of these is a count of the index. A heading reading Skill Index (561 skills) above a table of individual entries is plausible as a count of the visible rows, since a README can only display so many before it becomes unusable, while the 2,409 figure describes the full `skills/` directory.
The description field at 1,209 is the harder one to explain. It is close to the badge's alternative text of 1,211 and far below the current release total of 2,409, which suggests the description was written early and never updated, while the badge's visible text has been updated and its alternative text has not.
The practical consequence is simple. If you need an exact figure, read the `skills/` directory or the RELEASES.md file. Any number quoted from this repository's metadata should be treated as a snapshot from an unknown date.
What the index actually lists, and where it thins out
The index groups skills by category and the first group, titled AI and LLM Tools with 50 entries, is the densest and most revealing.
It runs from research tooling through browser automation to media generation. `academic-deep-research` is described as transparent research with full methodology rather than a black-box wrapper. `agent-browser` is a Rust-based headless browser CLI with a Node.js fallback. `browser-use` and `playwright` cover browser automation again from different angles, and `playwright-mcp` does it over MCP. `computer-use` describes full desktop control on headless Linux servers using Xvfb, XFCE and xdotool.
The media and audio entries are mostly thin wrappers around named services: `gemini` for the Gemini CLI, `perplexity` for grounded web answers with citations, `sag` for ElevenLabs text to speech, `openai-whisper` for local transcription with no API key, `nano-banana-pro` and `openai-image-gen` for image generation, `ltx-video` for video.
Several descriptions are empty or nearly so. The rows for `ai-humanizer`, `edge-tts`, `humanizer` and `ltx-video` show no description at all, and `humanize-ai-text` is described as a rewriter intended to bypass AI text detection, which is both a description of the tool and a flag worth pausing on.
Descriptions are also mixed-language. Rows like `ai-prompt-generator` and `image-generate` are described in Chinese while neighbouring rows are in English, and the skill index is not a place where consistency was enforced. A later group covering content and media includes `master-skills`, which is described as secure key management for AI agents covering private keys, API secrets and wallet credentials.
A maintained mirror attached to a commercial service
The promotion is at the top of the README and it is direct. MyClaw.ai is described as an AI personal assistant platform that gives every user a fully-featured agent running on a dedicated server, and this collection is framed as its curated weekly-updated set. A call to action to try the platform follows immediately.
That framing is worth separating from the technical facts. The repository itself is MIT licensed, has 2,144 stars, 325 forks and 31 open issues, and its last push was 2026-07-20, the same day as the 0.21.0 release. Weekly updates are real and verifiably happening.
What the repository does not provide is any evidence for the quality claim that the release notes lean on. There is no scoring rubric, no list of rejected skills, no popularity or download data, and no per-skill review. The words hand-picked in the platform description and quality rules in the release notes are the same assertion expressed twice.
For a mirror, that is a defensible position. You are not expected to trust the filtering when the alternative is not having the skills at all. But it does mean the index is a search aid rather than a recommendation, and that an empty description in the table means nobody wrote one rather than the skill being undescribed.
Editorial conclusion
This repository is a supply chain convenience, and that framing sets expectations for everything else about it. The value is that 2,000-plus skill directories arrive in one pull and a skill index tells you which one you want; the cost is that quality filtering is asserted rather than demonstrated, since the release notes say only that a quality rule was applied. The count discrepancy is the practical warning: nobody appears to be maintaining a single number. Use `skill_index.md` to find a skill, copy just that directory into your workspace, and treat the mirror as a search index rather than a dependency, because a weekly re-pull is exactly the kind of operation that should not sit in your build path.
Frequently asked questions
How many skills are in this OpenClaw collection?
Five sources give five different figures. The GitHub description says 1209+, the README badge displays 2409+ while its alternative text says 1211+, the index section is headed 561 skills, and the release notes claim totals of 2,409, 2,109 and 2,009 across three tags. Count the skills directory for an exact number.
What is an OpenClaw skill?
Based on this repository's structure, a skill is a directory containing instruction text that an agent loads. There is no code, build step or package manifest in the skills payload, and GitHub reports the language as null for the sibling marketing collection for the same reason. The installation step copies a whole skill directory into the agent's workspace.
How do I install a single skill from OpenClaw?
Clone the repository and copy one directory out of it into the agent workspace: git clone the project, then copy skills/<skill-name> to ~/.openclaw/workspace/skills/. Installing the whole collection with clawhub install also works, and the comment above that command claiming it installs one skill is inaccurate.
How do I master my skills?
This repository has no bearing on that question, since its contents are agent instruction files rather than a learning program. If you mean finding a specific agent skill, use the skill_index.md file at the repository root, which lists every directory in skills/ with a short description of what it does.
Are the skills in this mirror vetted for quality?
The release notes state that new skills were selected against quality rules sourced from the official openclaw skills-archive mirror, and delegate the details to RELEASES.md. No rubric, rejection log or per-skill review is published in the README, so the filtering is asserted rather than demonstrated.
Official sources
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.
[](https://hysenlabs.com/projects/leoyeai-openclaw-master-skills)