auto-ciUse as soon as a pull request is opened, or new commits are pushed to one, in a repository with GitHub Actions (a .github/workflows folder: Heimdall, Neptune, Poseidon, Shaggy, Albus, ...), whether you opened it or I did; also when I say a PR is up or ask whether the CI passed. Watches the checks in
auto-message-formatUse for every answer to me, in every session and project (the hook auto-message-format.py reminds you on each prompt), and when my message looks dictated by voice: long, no formatting, spoken style with fillers (donc, en fait, du coup, ok, so, basically). Gives the two answer formats, short and long
auto-updateUse when a session starts or resumes after a while in a folder of one of my projects or libraries (Heimdall, Neptune, albus, shaggy: the hook auto-update.py reminds you), when asked what changed in the libraries or the projects, before a task that needs the latest version of a library, and to write
cluster-luciaUse as soon as you work on Lucia, the CÉCI Tier-1 supercomputer of Cenaero (Wallonia) where Neptune runs (hostname frontal01 or frontal02, compute nodes cna, cni, cns), before any Slurm or dawgz job, choice of account, partition or GPU, BUh credit check, data transfer, storage or scratch operation,
cluster-remoteUse to start, check, reattach, stop or repair my Claude Code session kept alive in tmux on a cluster login node and reached through Remote Control (claude.ai/code, the Claude app), also when it is unreachable or says 'not running' because it runs on another login node; to install the setup on a new
cluster-torchUse as soon as you work on Torch (not PyTorch), the HPC cluster of New York University (hostname torch-*, or a compute node gh*, gr*, gl*), before any Slurm job, choice of account or GPU, storage or quota operation, data transfer or Apptainer environment. Also for the Open Storage Network (S3-compat
create-diagramUse before drawing a diagram or a step-by-step animation that explains how a mechanism, an architecture or a probabilistic concept works (attention, convolutions and receptive fields, graph networks, U-Net, autoencoders and VAEs, normalizing flows, diffusion models, stochastic interpolants, distribu
create-figureUse before drawing or fixing any figure meant to carry a scientific message (plots, charts, maps of results, data or diagnostics), for a report, a slide or on its own, in any project. Gives the rules to choose the plot, the colours, the scales, the captions, the uncertainty and the maps, the style o
create-messageUse when I want to write a message to share my progress, ask a question or ask for help with my team or a community (un message, un post, une update, partager mon avancement), usually with figures. Opens with a short list of proposed points and questions taken from the session, then writes one short
create-reportUse before writing an analysis report or a study summary to publish as an HTML artifact (results of an analysis, a diagnostic, a comparison of runs or designs, a summary of past work), in any project. Gives the fixed structure, the way of writing (the message first, details only when they serve), th
create-slidesUse when I ask for slides, a deck, a talk or a presentation (diapos, présentation, slides pour la réunion), for a progress meeting or a talk to people who do not know my work, built from the session or a project; also to change, rebuild or publish one of my talks. Opens with a fixed list of question
researchUse at the start of every session in one of my projects or libraries (Heimdall, Neptune, shaggy, albus), and whenever you need: what my projects study, who my supervisors are, which library to use before writing code (azula diffusion, zuko flows, shaggy training tools and autoencoders, albus metrics