Health Reasoning — Development skill for Claude Code

health-reasoning

Published at https://github.com/speckofthecosmos/health-reasoning (MIT).

An agent skill that makes the assistant check its own causal and evidential claims before it says them in a health conversation: weighing a treatment, reading a study or a lab trend, preparing for an appointment, or drafting an appeal or prior-authorization letter. It runs seven questions (compared with what, in whom, which endpoint, mechanism or result, how big in absolute terms, who is claiming it, and would I say this if it cut against the user), and when a claim fails it states the weaker true claim instead of refusing or hedging. `references/fallacies.md` lists about 40 healthcare reasoning fallacies, each with the question that catches it.

It is patient-side and is not medical advice, a clinician tool, or a disclaimer generator.

Install (Claude Code)

Copy the directory into your personal skills folder:

cp -r health-reasoning ~/.claude/skills/health-reasoning

or symlink it so updates follow the repo:

ln -s "$(pwd)/health-reasoning" ~/.claude/skills/health-reasoning

Format

The skill uses the [Agent Skills](https://agentskills.io) format: a `SKILL.md` with YAML frontmatter (`name`, `description`) plus a `references/` folder loaded on demand. Any agent that reads that format can use it.

**Trigger tuning:** the description is deliberately broad (any conversation about health, medical care or coverage of care). If it fires on work that is not about a patient's care, such as claim-submission mechanics or payer-portal scripts, add an exclusion line to the description.

License

The skill is released under the MIT License (see `LICENSE`), except for material adapted from CC BY 4.0 sources, which is included under the terms of [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).

Attribution

Changes were made to all borrowed material: it was paraphrased, rewritten for a patient-side conversation, and reorganised around the check q