lucid-coder
A Claude Code plugin for readable, reproducible R and Python in computational
biology. It gives Claude one house style for analysis code and figures, 21
tested pipeline skills for single-cell and statistics work, and rules learned
from your own corrections.
Docs: https://mdmanurung.github.io/lucid-coder/
Install
In Claude Code:
/plugin marketplace add mdmanurung/lucid-coder
/plugin install lucid-coder@lucid-coder
Start a new session. The rules load at session start and into every subagent.
What is inside
- Persona and rules (
rules/, skills/lucid-coder/): headers, knobs at
the top, checks where data enters, seeded steps, donors as the unit.
- Learned rules (
/lucid-coder:learn, /lucid-coder:rules): a lasting
correction becomes a rule after you approve it.
- Figure style (
skills/figure-style/, templates/): one look in ggplot2,
ComplexHeatmap and plotnine with plotnine-extra.
- Pipeline skills (
skills/<name>/): sc-qc, sc-normalize, sc-embed,
sc-annotate, sc-markers, pseudobulk-de, sc-de, variance-partition,
mashr-sharing, composition, milo-da, enrichment, signature-scoring,
tf-activity, gene-programs, ccc, trajectory, velocity, mofa, group-tests and
longitudinal. Each is a tested script with no analysis defaults, fixed
errors and a fixed output tree.
Commands
| Command | Effect |
|---|
/lucid-coder on, off, status | Turn the persona on or off |
/lucid-coder lang r (python, both, auto) | Choose the language rules |
/lucid-coder:review | Review code against the house style |
/lucid-coder:align | Install the house configs into a repo |
Develop
npm test # hooks, rules, templates
quarto preview site # docs site
Each pipeline skill's test command is in its references/maintainer.md. See
Develop for the full
guide.
