Marimo Figures
Use when the user wants to build, refine, or "Tufte-check" a figure in a marimo notebook — phrases like "plot this", "make a figure", "visualize these results", "tufte review this plot", or any request to iterate on a chart inside a `.py` marimo notebook. Pairs an author pass with a reviewer subagent that scores the figure against Tufte's principles and returns concrete revisions. Do not invoke f…
Experiment Report
Use when the user asks to "write up", "summarize", or "report on" an experiment run — phrases like "draft a report for run X", "write up these results", "summarize what we found". Produces an IMRaD-style markdown report under `reports/<slug>.md` and runs a scientific-writing reviewer subagent over it before handing back. Do not invoke for blog posts, commit messages, or casual summaries.
Plan Pi Fleshout
Use when a grad student arrives with a fuzzy experimental idea and wants help turning it into a concrete `PLAN.md`. Phrases like "I have an idea but it's not concrete", "help me think through…", "let's plan this out", "flesh this out", "I'm not sure what I want yet". You are the PI; the user is a grad student. You have no idea what the user actually wants — your job is to find out, term by term,…
Podman Runner
Use when the user wants to run an experiment script, command, or pipeline step inside a reproducible container — phrases like "run this in a container", "containerize this run", "execute train.py under podman", "reproduce this run". Handles the Containerfile, build, GPU passthrough, log capture, and `runs/<id>/` provenance. Do not invoke for local-only dev loops or for production deployments.
Dataset Insights
Use when the user wants to explore, profile, or "find something interesting" in a dataset — phrases like "what's in this data", "give me insights on X", "profile this dataset", "anything weird here?". Produces a ranked list of insights under `analysis/<slug>/` plus a reviewer pass that scores each insight for novelty, actionability, and evidence. Do not invoke for pre-specified hypothesis tests o…
Experiment Structure
Use when the user starts a new experiment, asks "how should I organize this", "where should X live", "scaffold an experiment folder", or when other skills (podman-runner, experiment-report, marimo-figures) need a canonical layout to write into. Lays down a reproducible directory structure and the rules that keep it honest. Do not invoke to reorganize an existing, working project without the user'…