Research and Academic Skills

Every skill in this guide is Claude Code only. For the ML engineering and data/plotting libraries these workflows build on, see machine learning.

Experiment workflow

Roughly in order: design it, plan the claims, run it, watch it, read it.

$experimental-design: choose a design, randomize, block, and lay out treatment combinations before any data is collected. The lead at the start; it is about the study, not the model.
$experiment-plan: turn a refined proposal or method idea into a claim-driven experiment roadmap. Follows $experimental-design once the method is settled — it plans what to demonstrate, not how to randomize.
$run-experiment: deploy and run ML experiments on local, remote, Vast.ai, or Modal GPU. The execution step; use $get-available-resources in machine learning first if capacity is unknown.
$monitor-experiment: check progress and collect output from a running experiment. Its narrow job is "is it done / how far along"; $analyze-results is what interprets the numbers.
$analyze-results: compute statistics and comparison tables over finished experiment results. Pick it for the experiment loop's output; for a one-off dataset survey use $exploratory-data-analysis in machine learning.
$arbor: autonomously improve an artifact (training recipe, harness, pipeline, prompt) against an objective and evaluator over many experiments, guarding against dev-set overfitting. Reach for it for a long optimization search, not a single planned run — the skills above own the one-run path.

Academic paper pipeline

Ships in the academic-research-skills plugin. These are heavier orchestrators than the single-purpose skills above — reach for one only when the deliverable is a full paper or a formal review, not a quick literature scan.

$academic-pipeline: the top orchestrator — research → write → integrity check → review → revise → finalize. Pick it when you want the whole paper produced end to end; use the narrower skills below when you only need one stage.
$academic-paper: the multi-mode writing pipeline (plan, outline, revision, abstract, lit-review, citation-check, and more) for drafting and revising a manuscript. This writes; $academic-pipeline sequences it with review.
$academic-paper-reviewer: a multi-seat simulated peer-review panel over an existing draft. The critique step, not the writing step.
$deep-research: a general multi-agent deep-research team for any topic, with modes from quick brief to full report. Choose it for broad rigorous research; the paper skills above own the manuscript itself.

Literature and review

$novelty-check: verify an idea's novelty against recent literature before committing to it.
$research-review: get a deep critical review of your research from an external reviewer backend. Pair it with $novelty-check — one checks whether the idea is new, the other whether the work holds up.

Statistics

$statistical-analysis: guided analysis of research data — test selection, assumption checks, effect sizes, power analysis, Bayesian alternatives, APA reporting. This decides and runs the right test; $statsmodels in machine learning is the library it may reach for.

Scientific communication

$scientific-writing: draft, revise, and audit manuscripts with evidence provenance, reporting-guideline coverage, and authorship accountability.
$scientific-slides: build decks for research talks, conference presentations, and seminars.
$scientific-visualization: create and audit truthful, accessible, publication-ready figures in Matplotlib, Seaborn, or Plotly. Use it over the raw $matplotlib/$seaborn skills when the figure is going in a paper and must clear a truthfulness and accessibility bar.
$scientific-schematics: generate publication-quality diagrams with an AI image backend and iterative review. Pick it for conceptual schematics; $scientific-visualization for figures drawn from data.