# ordinary-claude-skills: 600+ skill folders, two ways to load them, one manual trim step

> A dump of Claude skill folders from Anthropic, Composio and K-Dense AI, sorted into categories and cloned locally. Useful as a starting inventory, thin on licensing and on per-skill guarantees.

**Microck/ordinary-claude-skills** — An unappealing collection of Claude Skills and resources.

- Repository: https://github.com/Microck/ordinary-claude-skills
- Website: https://microck.github.io/ordinary-claude-skills/
- Stars: 402 · Forks: 55
- Language: Python
- License: NOASSERTION
- Published: 2026-09-15 · Updated: 2026-09-15 · Language: en
- Canonical page: https://hysenlabs.com/projects/microck-ordinary-claude-skills

## The clone, or the search index the docs folder builds

Microck/ordinary-claude-skills is a pile of Claude skill folders, not a library with an API. The repository's own description field reads "An unappealing collection of Claude Skills and resources.", and the README opens by calling it a massive local repository of official and community-built claude skills, organized by category. Two entry points sit at the top.

One is the static site under docs/, which indexes everything with search and categories so you can browse instead of digging through folders. The other is a clone, for anyone wiring the folders into their own MCP servers or agents.

```bash
git clone https://github.com/Microck/ordinary-claude-skills.git
cd ordinary-claude-skills
```

The local-first framing is the argument for cloning instead of leaning on a hosted index: nothing stops working when a third-party URL goes down. At 402 stars and 55 forks, the inventory is the project, not the code. GitHub reports Python as the primary language, which comes from the scripts individual skills ship rather than from anything the top level runs.

## skills_all and skills_categorized hold the same collection twice over

The layout is the difference between the two directories, not the contents. skills_all is everything in one flat place. skills_categorized is everything in its right place, spread across category branches. The tree in the README shows the shape, with backend/api-design-principles on one branch, web3-tools/solidity-security on another, and docs sitting beside both as the static website files.

```text
ordinary-claude-skills/
├── docs/                  # the static website files
├── skills_all/    		   # everything
├── skills_categorized/    # everything in its right place
│   ├── backend/
│   │   └── api-design-principles/
│   └── web3-tools/
│       └── solidity-security/
└── README.md
```

Two of the author's own claims describe what you are buying here. The selection is non-curated: everything got dumped in, and a skill that does not work may never get noticed, let alone fixed. Folder structures are kept somewhat consistent by hand, which is what makes skills_categorized browsable at all, and it is a promise about shape rather than about quality.

Seven categories get named. Science and academia covers protein folding, astronomy and lab automation. Software engineering covers api design, debugging and testing. Infrastructure covers kubernetes, docker and terraform. Data and ai covers vector dbs, llm evaluation and rag. Business covers marketing, finance and legal. Creative covers writing, art and philosophy. Web3 covers solidity, smart contracts and defi.

Past those seven, the collection runs past 600 entries. The README stopped listing them after a while because it made the readme scroll for eternity, so the inventory now lives on the documentation site.

## A skill is a prompt package that loads only when it is needed

Each folder holds instructions, and sometimes scripts. The README's definition: skills are prompt packages and scripts that teach claude how to do specific things without you having to explain the context every single time.

Load behaviour carries the cost. Skills load lazily, only when needed, which the author credits for saving context window space and for keeping claude from getting confused by instructions it does not need yet.

That reasoning only holds if you load narrowly, and the example prompts show the narrow version. A React component that will not render its list items gets debugging-strategies plus frontend-design. A landing page to analyse gets competitive-ads-extractor. A scanned spreadsheet pasted into a Word document and exported as a PDF gets the pdf skill. A Friday afternoon push to production gets webapp-testing run against localhost:3000, which starts playwright. Four scenarios, four small slices, each named as a single skill or a pair. Load all 600 at once and the context saving that motivates the format disappears.

## Pointing an MCP client at a skill directory

For API users and developers, the instruction is to point your mcp client or system prompt config at the relevant skill directory. The configuration sketched in the README runs the filesystem MCP server over the whole collection.

```json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/path/to/ordinary-claude-skills/skills_all"
      ]
    }
  }
}
```

The second option is the one to copy. Narrow that path to a category, or to a single skill, instead of skills_all.

```json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/path/to/ordinary-claude-skills/skills_categorized/[category]"
      ]
    }
  }
}
```

The difference between those two blocks is one path segment, and it is the difference between every skill in the collection being reachable in one session and only the ones you picked. Nothing in the format stops you pointing at skills_all on day one and regretting it in week two.

## Uploading one folder through custom skills on claude.ai

The claude.ai path skips MCP entirely. Go to your profile, hit custom skills, and upload the specific folder for the skill you want. There is no install step, because the upload is the install.

Verification takes one question. Ask claude `can you use the [skill name] skill now?` and a yes means the folder is attached.

A no has one documented cause: make sure you explicitly told claude the skill exists in the system prompt, or that the file was successfully attached to the project context. That is a failure in the handoff, not in the folder, which is why the fix touches the prompt and the attachment rather than the skill itself. A skill can sit on disk, be named in a config file, and still never reach the model.

Per-folder upload also bounds the damage when a skill misbehaves. One folder means one set of instructions in context, so a pdf extractor never ends up competing with an api design skill for the same attention.

## node_modules inside a skill folder, and the file too large error

Some skills ship heavy dependency folders, and that is what produces the file too large error. The fix given in the README is to ignore the node_modules inside skill folders: you only need the source scripts and the instructions.

That is a per-skill problem rather than a repository one, and it lands hardest on the upload path, where a folder goes to claude.ai in one piece and size is the limit. On the clone side nothing forces you to keep a dependency tree the instructions never reference, and removing node_modules leaves the scripts those instructions actually call.

No script is offered for doing this across all 600-plus folders at once. It stays a manual pass unless you write it yourself, which is the first piece of real work this repository hands you.

## creative-writing beside technical-documentation, and who disagrees

Skills contradict each other when their instructions overlap. The README names the pair: do not load creative-writing and technical-documentation at the same time, because claude gets confused about whether it should be shakespeare or a robot. Two skills that prescribe incompatible registers do not average out into something usable.

Runtime requirements vary per skill too. The repository has no dependencies of its own, but an active internet connection and a claude account or api key are listed as mandatory. Python 3.x is listed for data analysis skills, node.js for mcp builder and testing skills, and playwright for browser automation.

A category label tells you nothing about what a folder needs. The dependency line has to be read inside each skill, which is the tax the local-first layout charges you for not having a single manifest.

## A changelog file, no releases, and a license section that stops mid-sentence

The top level is short: .gitignore, CHANGELOG.md, LICENSE, README.md, docs/, skills_all/ and skills_categorized/. There is a CHANGELOG.md and there are no GitHub releases, so that file is the only version history on the repository itself. The last push landed on 6 September 2026.

Licensing is the open question. GitHub reports no asserted license for the repository. The README's license and credits section begins by conceding that the author did not write most of the skills, then stops mid-sentence, so the attribution scheme the collection was built on is not spelled out at the level a reader needs it.

That matters more here than in a normal dependency. The skills come from Anthropic, Composio, K-Dense AI and what the README calls random internet geniuses, and nothing at the top level settles which terms travel with which folder. Check a folder before shipping it.

## Conclusion

Adopt this if you want breadth fast and are willing to sort it yourself: clone it, point one client at one category, and read the dependency list inside each skill folder before counting on it. Pass on it if you need a supported inventory with one license and one release train, because this repository has neither releases nor an asserted license, and its own troubleshooting notes tell you to prune dependency folders by hand. Before wiring anything in, load a single skill and confirm the model actually reaches for it.

## FAQ

### What does Microck/ordinary-claude-skills contain?

Skill folders collected from Anthropic, Composio, K-Dense AI and what the README calls random internet geniuses. They sit in a flat skills_all directory and again in a skills_categorized tree, with a static site under docs for searching them.

### How do I install a skill from Microck/ordinary-claude-skills?

Clone the repository, then either upload one skill folder through custom skills in your claude.ai profile, or point your mcp client config at the skill directory. Verification is one prompt: can you use the [skill name] skill now?

### How many skills does Microck/ordinary-claude-skills hold?

The README refers to 600 or more. Seven categories are named: science and academia, software engineering, infrastructure, data and ai, business, creative, and web3. The complete inventory is on the documentation site rather than in the README.

### Does Microck/ordinary-claude-skills need Python, Node.js or Playwright?

The repository itself has no dependencies. An internet connection and a claude account or api key are listed as mandatory, with python 3.x for data analysis skills, node.js for mcp builder and testing skills, and playwright for browser automation.

### What license does Microck/ordinary-claude-skills use?

GitHub reports no asserted license for the repository, and a LICENSE file sits at the top level beside a CHANGELOG.md. The README's license and credits section stops mid-sentence after the author notes that he did not write most of the skills.

## Sources

- [Issues](https://github.com/Microck/ordinary-claude-skills/issues)
- [Microck/ordinary-claude-skills on GitHub](https://github.com/Microck/ordinary-claude-skills)
- [Project website](https://microck.github.io/ordinary-claude-skills/)
- [README](https://github.com/Microck/ordinary-claude-skills/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/microck-ordinary-claude-skills