Sensei Ewok skills
Reusable agent skills from the Sensei Ewok Research Lab, an independent open-source lab building small tools for cystic fibrosis (CF) research. Each skill is a folder with a SKILL.md that an AI agent reads when your task matches, plus scripts and references it can use. The format is the open Agent Skills format, so one folder works in many agents.
/------------------------------+
/ . . . o o . . . . . o o . . |
| . . o . . o . . . o . . o . |
| . o . . . . o . o . . . . o |
| o . . . . . . o . . . . . . - .
| . o . . . . o . o . . . . o | .
| . . o . . o . . . o . . o . |
| . . . o o . . . . . o o . . | .
+-------------------------------+
.
.
A punched card with a clipped corner, drawn in plain line characters; its holes trace a double helix across a field of unpunched dots, and a dotted thread leaves its right edge and drifts down and away.
A pattern punched once can be woven by hands that never met the one who punched it.
Nine skills so far, an install helper, an index of what each skill can do, and notes on where every major agent looks for skills. The catalog is CC0; each skill declares its own license. Research, not medical advice.
Using AI around cystic fibrosis care? Five skills here help patients, families and care teams use an AI chat app more safely. You need no install: Where to paste shows how to copy one into your chat app.
What this is
A skill is a folder: a SKILL.md that an agent reads when your task matches its description, plus the scripts, references and tests it may use. The lab writes skills for its own agents and publishes here the ones that are useful outside its workspace, in the open Agent Skills format, so one folder works in Claude Code, GitHub Copilot, Cursor and the other agents in the table under Use a skill.
cf-evidence-loop is the skill the lab's website uses for its evidence records. A researcher does not begin with an answer; they walk a fixed set of questions against the public record. This skill gives an agent that walk as a command line: 47 commands (counted from cli.py on 2026-10-25) across published literature, retractions, drug approvals and labels, clinical trials, preprints, variant classifications, research funding, citation graphs, and what the tool may reach at all. Every answer is an evidence record with its source, the date it was read, the exact fields relied on and a limitation that the code refuses to leave empty. It never calls a language model, and it reaches only the sources its catalog permits, under network rules that are code with tests: robots.txt obeyed, one request at a time, a budget, a circuit breaker, one honest user agent. site-seo-review is a careful search-and-sharing review for a small website, with rules for health and research sites; it is not specific to CF.
Five skills for people who use AI in cystic fibrosis care: cf-ai-safe-use, cf-plain-language-rewrite, cf-visit-prep, cf-answer-check and cf-ai-tool-review. They are for patients, families and care teams as well as builders. Each has a paste-ready version you copy into a chat assistant: no install and no scripts. If you are on a care team, use the AI tool your organization approves. If the question is about your own or your family's health, think about who can see your chats. A work account belongs to your workplace, so check its policy, or use a personal account. Where to paste says how, product by product, and skills by audience says who each skill is for. They give general information only, never medical advice, and every rule's evidence (or the lack of it) is in the skill's references/evidence.md.
These skills are "in review": the maintainer has not yet confirmed every license or read every file, and the install notes were read from vendors' pages, not tested in each client. Read a skill's SKILL.md and its scripts before you let an agent run them.
What this is not: a medical device, a clinical tool, or a claim about treatment. An approval date is not anyone's eligibility, a label's text is a public document and not advice, and a record is a thing to check, not a conclusion. The card at the top of this page is an emblem: its holes echo DNA, but it is not a diagram of a gene, a protein or a treatment.
Try it
git clone https://github.com/senseiewok/cf-skills.git
python cf-skills/scripts/install_skill.py --list
python cf-skills/scripts/install_skill.py --how cf-evidence-loop --agent claude
--list names the skills; --how says in plain words what a skill does (it makes network requests, it ships scripts an agent may run, it needs Python) and lists only the install routes that work on your machine. The script runs nothing from the skill. The per-agent folders, the plugin marketplace and the paste-in prompts are under Use a skill. The evidence tool's own tests touch no network: cd cf-skills/.claude/skills/cf-evidence-loop && python -m pytest -q tests (needs pytest, requests and PyYAML: pip install -r scripts/requirements.txt; tested on Python 3.13).
What's inside
What is in this repository below lists the repository layout. In short: the skills under .claude/skills/, each with its tests; scripts/ with the install helper, the index builder and the repository check; docs/ with how each agent finds skills, the paste-in prompts and recommended skills from other people; template/ and spec/ for writing a skill; skills-index.json for tools; .claude-plugin/ for Claude Code's marketplace.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
