Ersilia Skills

A guide to Ersilia's Claude Code skills repository: what it contains, how to install it, and how to contribute new skills.

ersilia-skills is Ersilia's curated library of Claude Code skills (reusable AI workflows built on top of Claude Code that encode our institutional knowledge into slash commands). Instead of re-explaining context to Claude each time you start a task, you invoke a skill and Claude already knows the process, the Ersilia conventions, and the expected outputs.

What's in this repository

• skills/ : The heart of the repository. Each subdirectory is one skill (a self-contained folder with a SKILL.md file that defines the workflow Claude follows when you invoke the skill). Most skills also include a references/ folder with supporting documents (guidelines, templates, real examples, troubleshooting notes) that Claude reads as context during execution.

• setup.sh: A one-time installation script. When you run it, it creates symlinks from every skill directory in this repository into ~/.claude/skills/, which is the folder Claude Code reads to discover locally installed skills. It also installs the post-merge git hook

• .git-hooks/: Contains a post-merge hook that runs automatically every time you do a git pull. It re-runs setup.sh in the background, so any new skills added to the repository since your last pull are immediately linked into your local Claude Code installation.

Setup

Prerequisites:

• Claude Code

• Git

1

Clone ersilia-skills repo

git clone https://github.com/ersilia-os/ersilia-skills.git
cd ersilia-skills

2

Run setup script

bash setup.sh

3

Verify installation

Verify the installation by opening a Claude Code session and running:

/test-skill

Claude should confirm that the symlink is in place and that the setup is working correctly.

Keeping skills up to date

Once setup is done, you only need to git pull to stay current. The post-merge hook that was installed by setup.sh runs setup.sh automatically after every pull, so new skills added to the repository will be linked into your ~/.claude/skills/ directory without any manual steps.

Using a skill

In any Claude Code session, type the skill name as a slash command:

/ersilia-model-test eos4ywv ~/models/eos4ywv
/molecule-auditing results.csv --context malaria
/newsletter-drafting

Claude will load the skill's workflow and follow it. Some skills accept arguments (like a model ID or a file path); the skill's description shown in Claude Code will tell you what to provide.

Skill Catalogue

Skill Catalogue

At Ersilia, we have identified three key goals we want to make progress on, and for each one we have mapped specific skills we believe can help us get there: better public reach, Ersilia Model Hub growth, and improving scientific literacy.

The skills are still being developed and tested by the Ersilia team; you can check the ones that are ready-to-use and available on GitHub.

1. Better public reach

Name	Skill
stylia-plotting	How to create Python plots using the stylia package — Ersilia's matplotlib wrapper for publication-ready figures.
event-discovery	Discover interesting events for Ersilia and write a summarised report. Classify between categories (local/global, science/philanthropy, etc.)
html-formatting	Style any HTML into the recognisable Ersilia look and feel, and improve its UX at the same time.

2. Ersilia Model Hub growth

Several of these skills are designed to be chained together as part of a larger workflow rather than run in isolation:

• Model incorporation pipeline — model-incorporation-request, model-incorporation-metadata, model-incorporation-code, and model-incorporation-reproduce cover the full lifecycle of bringing a new model into the Hub, from opening the initial request to verifying it reproduces the original paper's results. They are meant to be run in sequence.

• Hub maintenance workflow — model-discovery, ersilia-model-test, model-monitoring, model-fixing, and github-digest work together as a recurring maintenance loop: discovering new candidate models, testing them before incorporation, monitoring the state of models and stored data, fixing what fails, and digesting GitHub activity to keep track of it all. We recommend running these as a bundled workflow rather than as standalone skills.

Name	Skill
model-incorporation-request	Open a model request issue on ersilia-os/ersilia.
model-incorporation-metadata	Fill in metadata.yml from the paper and source repo.
model-incorporation-code	Wire the model code into the Ersilia template.
model-incorporation-reproduce	verifies model outcomes/performance matches the original work.
ersilia-model-test	Tests an Ersilia Model Hub model before hub incorporation.
model-fixing	When a model fails a test, reviews where it failed and fixes it automatically.
github-digest	Tracks open issues and produces summaries for tech-tracking meetings.
repository-auditing	Audits a repository to make sure it abides by Ersilia's standards.
model-monitoring	Track pending models, stored data, etc.

3. Improve scientific literacy

Name	Skill
literature-review	Given a topic, offer a structured review from the literature. This review will provide you with relevant research/review papers, alongside potential ML models and datasets that can be included in Ersilia.
literature-digest	Produces a weekly literature digest for Ersilia.
molecule-auditing	Audits small molecules suggested in Ersilia's screening and scores them according to parameters of interest.
paper-to-model-assesment	Summarize a given paper and put it in context of Ersilia's interests.
peer-reviewing	Emulate a peer review and suggest how to address changes.

Contributing

To add a new skill:

1. Create a branch from main.

2. Add a new directory under skills/ named after your skill (use lowercase and hyphens).

3. Write a SKILL.md file. The frontmatter must include at minimum name and description. Add argument-hint if the skill takes arguments, and allowed-tools to restrict which tools Claude can use.

To make the process easier, use Anthropic's built-in skill-creator skill. Once you have Claude Code set up, run /skill-creator in your session and Claude will guide you through designing and writing a new skill from scratch, helping you structure the workflow, write the frontmatter, and identify what reference documents to include. It is the recommended starting point for anyone contributing their first skill to this repository.

1. Add a references/ folder with any supporting documents Claude should read — guidelines, examples, templates, troubleshooting notes.

2. Open a pull request. The post-merge hook will ensure collaborators get the skill automatically after they pull