con un clic
init-repo
// Initialize a new research repository with the recommended AIRS project structure. Use when starting a new project, setting up a repo, or user says "init repo", "scaffold project", "create project structure".
// Initialize a new research repository with the recommended AIRS project structure. Use when starting a new project, setting up a repo, or user says "init repo", "scaffold project", "create project structure".
| name | init-repo |
| description | Initialize a new research repository with the recommended AIRS project structure. Use when starting a new project, setting up a repo, or user says "init repo", "scaffold project", "create project structure". |
| argument-hint | ["project-name-or-description"] |
| allowed-tools | Bash(*), Read, Glob, Grep, Write, Agent |
Project: $ARGUMENTS
Scaffold a new research repository following the AIRS recommended structure. This creates a reproducible, well-organized project layout with pixi for environment management and clearly separated concerns.
<project-root>/
├── pixi.toml # Pixi environment config (conda/pip dependencies)
├── pixi.lock # Auto-generated lock file (do NOT create manually)
├── .gitignore # Ignore patterns for data, checkpoints, outputs
├── README.md # Project overview and getting started
├── resource/ # Related papers (PDFs) and reference codebases
│ ├── papers/ # PDFs of related work
│ └── codebase/ # Cloned or vendored reference implementations
├── papers/ # LaTeX paper sources
│ ├── main.tex # Main LaTeX entry point
│ ├── references.bib # BibTeX references
│ └── figures/ # Figures for the paper
├── scripts/ # Experiment-related scripts (train, eval, plot)
├── main/ # Main entry functions and core source code
│ └── __init__.py # Package init (empty)
└── outputs/ # Experiment outputs, logs, checkpoints (gitignored)
Create all directories:
mkdir -p resource/papers resource/codebase papers/figures scripts main outputs
Generate a pixi.toml with:
default environmentTemplate:
[project]
name = "<project-name>"
version = "0.1.0"
description = "<one-line description from ARGUMENTS or 'Research project'>"
channels = ["conda-forge"]
platforms = ["linux-64"]
[tasks]
train = "python main/train.py"
eval = "python main/eval.py"
[dependencies]
python = ">=3.10"
[pypi-dependencies]
Adjust platforms based on the target system if known (e.g., add osx-arm64 for Mac).
# Outputs and checkpoints
outputs/
checkpoints/
*.ckpt
*.pt
*.pth
# Pixi
.pixi/
pixi.lock
# Python
__pycache__/
*.pyc
*.egg-info/
dist/
build/
# Data
data/
*.h5
*.hdf5
*.parquet
# IDE
.vscode/
.idea/
# OS
.DS_Store
Thumbs.db
main/__init__.py — empty filepapers/main.tex — minimal LaTeX skeleton:\documentclass{article}
\usepackage{amsmath,graphicx}
\bibliographystyle{plain}
\title{<Project Name>}
\author{}
\date{\today}
\begin{document}
\maketitle
\begin{abstract}
TODO: Abstract
\end{abstract}
\section{Introduction}
TODO: Introduction
\section{Related Work}
TODO: Related Work
\section{Method}
TODO: Method
\section{Experiments}
TODO: Experiments
\section{Conclusion}
TODO: Conclusion
\bibliography{references}
\end{document}
papers/references.bib — empty file with a header commentREADME.md — brief project overview with the directory structure explainedscripts/.gitkeep — keep the directory in gitresource/papers/.gitkeep — keep the directory in gitresource/codebase/.gitkeep — keep the directory in gitgit init
git add -A
git commit -m "init: scaffold AIRS research repository structure"
If already inside a git repo, just stage and commit the new files.
If pixi is available on the system:
pixi install
If pixi is not installed, skip this step and note in the output that the user
should install pixi (curl -fsSL https://pixi.sh/install.sh | bash) and run
pixi install manually.
outputs/ directory must be gitignored.Autonomous multi-round research review loop. Supports two modes: (1) Plan-driven: takes an implementation plan file, executes TODO items respecting dependency DAG, uses Codex MCP to verify completion of each item. (2) Free-form: iterates review → fix → re-review until positive assessment. Use when user says 'auto review loop', 'review until it passes', or wants iterative improvement.
Show GPU availability across all SSH servers listed in this project's CLAUDE.md. Use when user says "check GPUs", "which GPUs are free", "gpu status", "GPU 状态", or needs to know where to run experiments.
"Register or update an ARIS run on the web dashboard. Internal utility used by other skills and CLAUDE.md auto-registration. Call at skill start to register, at skill end to report completion. If ~/.claude/aris-api.json is absent, does nothing (zero-impact)."
Sync local and remote project files bidirectionally. Handles code, resources, papers, and experiment outputs. Use when user says "sync", "push files", "pull files", "upload code", "download results".
Search and analyze research papers, find related work, summarize key ideas. Use when user says "find papers", "related work", "literature review", "what does this paper say", or needs to understand academic papers.
Set up a reproducible Python project environment on the remote server using pixi or uv. Use when executing a project environment setup node generated by the jumpstart mechanism.