# agentara skills: the recommended one-liner auto-confirms a blob URL, six of twenty-three skills are Creative Commons, and only one category prefixes its names > agentara/skills is a personal library of prompt files that extend a coding assistant, grouped into five categories and installed one at a time into the assistant's skills directory. The engineering half is adapted from a public engineering practices guide under a Creative Commons licence; the rest is MIT. It is a prompt library rather than software, which is why the interesting problems are all about how the files are named, licensed, and installed. **agentara/skills** — Original and practical skills for AI builders. - Repository: https://github.com/agentara/skills - Stars: 600 · Forks: 42 - Language: HTML - License: MIT - Published: 2026-09-20 · Updated: 2026-09-20 · Language: en - Canonical page: https://hysenlabs.com/projects/agentara-skills ## The recommended install hands a web page address to a runner with no prompt The recommended way to install one skill is a single line: ```bash npx skills add https://github.com/agentara/skills/blob/main/skills/productivity/doctor-strange/ -y -g ``` Three things about that line are worth pausing on. The address is a repository web page, the kind a browser opens, rather than a raw content address or a clone URL. It is being passed to a package runner, which will fetch and resolve it. And the flags pre-set both decisions that a person would normally make: the first suppresses the confirmation, the second installs globally rather than into one project. So the recommended path is an unconfirmed, global, remote install triggered by a page URL, with no checksum and no output the reader is asked to read. Nothing here suggests it is malicious. The point is that the documented happy path removes every checkpoint a cautious installer would rely on. ## Two skill roots, two symlinks, and no rule for keeping them in step The manual route says to clone the repository and then symlink or copy the skill directory you want into two places: an agent-agnostic skills folder and the one specific assistant uses. The worked example does both: ```bash ln -s /path/to/skills/skills/productivity/doctor-strange ~/.agents/skills/doctor-strange ln -s /path/to/skills/skills/productivity/doctor-strange ~/.claude/skills/doctor-strange ``` So one installed skill becomes two directory entries, one of them a symlink into a clone the user chose the location of. The page says this is so the assistant and other agents can discover it, which is a reasonable goal, and then says nothing about which of the two is authoritative, what happens if they diverge, or whether the one-line route writes to both. With symlinks, a stale path is the obvious failure: move or delete the clone and both entries become dangling, with no error from the assistant beyond the skill not being found. ## Two licences in one repository, split by directory The licensing section has two paragraphs and the first one is qualified. Most original content in the repository is MIT. The engineering skills are adapted from a public engineering practices guide under a Creative Commons attribution licence, they carry source attribution in each skill file, and their wording and structure were changed for the library. So six of twenty-three files are under one licence and the rest under another, and the split is by directory rather than by file. That is workable, and attribution in each file is the right way to do it. The awkward part is the word most. It implies content that is neither original nor covered by the engineering exception, and the page does not say what that content is licensed under. Anyone vendoring a single skill out of this repository needs to know which half it came from, and the answer is only available by looking at which directory it sat in. ## Three skills give health guidance and one keeps records about a child The health category holds three skills. One triages urgent and semi-urgent symptoms, narrows possible causes, and prepares department and examination suggestions. One interprets medical lab reports from images, PDFs, or text and explains abnormal values with next steps. One is a family-doctor-style parenting assistant that maintains a directory in your home folder, builds child profiles, runs structured visits, creates plans, generates plan images, and archives cases. The first two are reference tools a person consults. The third is different in kind: it builds a persistent, growing record of a child's data on disk, outside the repository, with no stated retention or deletion path. None of the three carries a scope limit on the page, and none of the three says what it is not. ## Only one category prefixes its directory names The reference list gives every skill a name and a path, and the paths are not uniform. The AIGC skills are unprefixed, so a presentation design skill sits in a directory named for a presentation design skill. The entertainment and health and productivity skills are unprefixed too. All six engineering skills carry the category name twice: the category directory, and then a file whose name begins with the category. So a reader scanning the engineering section sees a doubled name and a reader scanning the others does not, with no stated reason. This matters more than it looks, because the install instructions use the directory name as the on-disk skill name. The name a skill ends up with in the assistant's skills folder therefore depends on which category it came from, for no reason a user could predict. ## A third of the library is art direction, not engineering The repository describes itself as original and practical skills for AI builders, and the categories are generated content, entertainment, health, engineering, and productivity. Counting the entries gives twenty-three. Eight of those are generated content: a presentation design board, a model-kit photography poster built from a guided interview, a photo-to-illustration transform, a torn-paper collage poster, character design specs, a video plan, poster key art, and a storyboard board. That is more than a third of the library, and all eight are image-prompt and layout work rather than engineering. Engineering holds six, productivity five, health three, and entertainment one. So the largest category is art direction, the second is software practice, and the category a visitor is most likely to arrive for is third. ## One skill deploys, one reads a betting market, and one writes durable priors Three entries reach outside the assistant's own context, and the page does not flag any of them. One turns a topic into a citation-backed interactive website and deploys it with a cloud command line tool. One predicts a football tournament through parallel subagents that read live news, weather, injuries, markets, a prediction market, tactics, and tournament context, updating a real-time dashboard. One runs scenario simulations through parallel universe subagents and then stores and recalls the projections as soft priors. The last is the most interesting mechanism in the library: a prompt file that creates durable state which then influences later answers, with no stated provenance, expiry, or way to inspect what it has concluded. None of the three is described as more privileged than the other twenty. ## Conclusion Read this as a curated prompt collection rather than a product, and pick skills individually, because the differences between them matter more than the collection. Start with the engineering ones: they are the most concrete, they are the ones with a second author and a second licence, and they are the ones whose provenance is documented inside each file. Treat the health ones as reference material rather than as anything to follow, since nothing on the page bounds what they will advise on a symptom. And if you install anything, install it from a clone you have read rather than from the one-line runner, because that command hands a web page address to a package runner with confirmation and scope both pre-set. ## FAQ ### How do I install one skill from agentara/skills? The recommended route is a one-line runner command that takes the skill's repository page URL and passes flags for skipping confirmation and installing globally. The manual route is to clone the repository and symlink or copy the skill directory into both an agent-agnostic skills folder and the one specific assistant reads. ### How many skills does agentara/skills contain? Twenty-three across five categories: eight in generated content, six in engineering, five in productivity, three in health, and one in entertainment. The largest category is image prompt and layout work rather than software engineering. ### What licence applies to the agentara/skills engineering skills? They are adapted from a public engineering practices guide under a Creative Commons attribution licence, with source attribution inside each skill file and wording changed for the library. Most other original content in the repository is MIT. ### Which agentara/skills entries reach outside the assistant's context? One deploys a generated website with a cloud command line tool, one forecasts a football tournament using parallel subagents that read live news, weather, markets, and a prediction market, and one stores scenario simulations as durable priors that are recalled in later answers. ## Sources - [agentara/skills on GitHub](https://github.com/agentara/skills) - [Issues](https://github.com/agentara/skills/issues) - [License: MIT](https://github.com/agentara/skills/blob/main/LICENSE) - [README](https://github.com/agentara/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/agentara-skills