feral-lab-os

Claude Code skills for health decision science, from FerAl Lab. The repository is a Claude Code plugin marketplace that currently offers one plugin, decision-modeling.

Skills in decision-modeling

Skill	Use it for
dm-calibration	Setting up, extending or debugging a model calibration in R: choosing the likelihood for each target (binomial, Poisson, negative binomial, multinomial, overdispersion, model discrepancy), running the PRE-CISE pre-calibration checks (coverage, sensitivity, collinearity) before spending compute, choosing a search method (random search, Nelder-Mead, IMIS, DEzs MCMC), diagnosing non-identifiability, and deciding whether parallelizing is worth it
model-code-review	Reviewing existing R and simulation-model code: correctness first (rate and probability conversion, transition matrices, time indexing, random numbers), then reproducibility, then computational efficiency
lab-conventions	Coding and naming conventions for decision models, based on the DARTH coding framework. For the project folder layout it points to FerAlLab-git/project-template

Claude uses each skill on its own when a request matches it. You can also call one directly, for example /decision-modeling:dm-calibration. dm-calibration is for building a new calibration and model-code-review is for reviewing code that already exists.

Install

You need Claude Code and git. In a Claude Code session in a terminal:

/plugin marketplace add FerAlLab-git/feral-lab-os
/plugin install decision-modeling@feral-lab-os

In Positron, VS Code or the desktop app, the chat panel does not run /plugin. Install from Terminal instead:

claude plugin marketplace add FerAlLab-git/feral-lab-os
claude plugin install decision-modeling@feral-lab-os

The same two commands work from any shell. The plugin is installed for your user, so it is then available in every Claude Code surface on that machine; start a new session to load it.

Check that it loaded:

claude plugin list

You should see decision-modeling@feral-lab-os with Status: enabled. Then start claude, type /, and look for dm-calibration, lab-conventions and model-code-review.

If your machine has no GitHub SSH key, set CLAUDE_CODE_PLUGIN_PREFER_HTTPS=1 in your environment so Claude Code clones over HTTPS. The repository is public, so no GitHub account is needed.

Update

A new release reaches you only when the plugin's version changes. To update, run these in Terminal (not in the chat panel of Positron, VS Code or the desktop app):

claude plugin marketplace update feral-lab-os
claude plugin update decision-modeling@feral-lab-os

Then start a new Claude Code session.

Turn on auto-update

Auto-update is off by default for marketplaces outside Anthropic's own. To turn it on, run /plugin, open Marketplaces, select feral-lab-os, and choose Enable auto-update.

With auto-update on, Claude Code checks for a new version shortly after the first message of a session (after a random delay of up to ten minutes) and installs it. You will see Plugin updated: decision-modeling · Run /reload-plugins to apply; the new version loads on your next launch either way.

Setting DISABLE_AUTOUPDATER=1, DISABLE_UPDATES=1 or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 turns plugin auto-update off, unless you also set FORCE_AUTOUPDATE_PLUGINS=1.

Get notified of new releases

Each version is published as a GitHub release, with notes taken from CHANGELOG.md. To get notified, open this repository on GitHub, click Watch, choose Custom, tick Releases, and click Apply.

Contributing

Anyone can propose a change to a skill. main is protected, so every change comes in as a pull request.

If you are not a member of FerAlLab-git, fork the repository first and open the pull request from your fork.

1. Create a branch:

   git checkout -b <short-description>

2. Edit the skill under plugins/decision-modeling/skills/<skill-name>/.
3. Add an entry to CHANGELOG.md under a ## [Unreleased] heading at the top (add the heading if it is not there). Say what changed and whether users need to do anything.
4. Validate from the repository root:

   claude plugin validate .

5. Push the branch and open a pull request.

Leave the version in plugin.json alone; a maintainer sets it before merging.

Releasing (maintainers)

For each Contributing PR you accept:

1. Bump the version on the pull request's branch, before merging. Raise version in plugins/decision-modeling/.claude-plugin/plugin.json using semantic versioning:

   • patch (0.1.0 to 0.1.1): wording fixes, clarifications
   • minor (0.1.0 to 0.2.0): new guidance, a new skill, changed behavior
   • major (0.x to 1.0.0): a renamed or removed skill

   In the same commit, rename ## [Unreleased] in CHANGELOG.md to ## [X.Y.Z] - YYYY-MM-DD. Then merge the pull request.

   If you can't push to the contributor's branch (for example, a fork that does not allow edits by maintainers), merge the pull request as it is and make these two changes in a small release pull request instead.

2. Tag the release from main once the version bump is merged:

   git checkout main && git pull
   claude plugin tag --push plugins/decision-modeling

   claude plugin tag reads the version from plugin.json, checks that it agrees with the marketplace entry, tags the current commit decision-modeling--vX.Y.Z, and with --push pushes that tag to origin. It stops if the working tree has uncommitted changes or the tag already exists.

3. Publish a GitHub release for the new tag. On GitHub, open Releases, choose Draft a new release, select the tag, and paste that version's entry from CHANGELOG.md as the notes.

4. Re-upload any skill you also keep on claude.ai. Skills uploaded to claude.ai do not update from this repo. If you changed one, zip it and upload it again in claude.ai skill settings, replacing the old copy:

   cd plugins/decision-modeling/skills
   zip -r ~/Desktop/<skill-name>.zip <skill-name>

   An out-of-date copy on claude.ai will also sync into Claude Code and run alongside the plugin version.

Credits

The worked examples (the Sick-Sicker and cancer relative survival models) come from the DARTH workgroup's teaching materials and from HRP 263 at Stanford University. The coding conventions follow Alarid-Escudero F, et al. "A need for change! A coding framework for improving transparency in decision modeling." PharmacoEconomics 2019;37(11):1329-1339, doi:10.1007/s40273-019-00837-x.

License

MIT