Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
This repository provides a suite of tools for machine learning researchers to streamline the literature review and experimentation process. It integrates capabilities for searching arXiv, retrieving LLM-optimized summaries via AlphaXiv, and computing statistics from experiment results to generate comparison tables.
The system extends beyond data retrieval by assisting in the preparation of paper submissions, offering specialized logic for planning ablation studies and generating technical Mermaid diagrams to visualize complex requirements.
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Follow the documented setup, then try a first task.
Claude Code selective paper-writing install
Installs: Installs the ARIS paper-writing skill into one project through a symlink, automatically including its hard dependencies, and configures the global Codex reviewer bridge. · Claude Code
Before you start
macOS
Homebrew installed
Claude Code installed
Codex CLI installed
Git available for cloning the repository
Bash available to run the installer
Python 3 available as python3 for the Codex MCP bridge
Platform: macOS local environment; the overall guide also covers an optional remote Linux GPU server.
Run from: ~/your-paper-project
In a terminal, verify that Claude Code is installed.
claude --version # verify installation
In a terminal, verify Codex CLI and authenticate it once. The login opens a browser; skip the login command if already authenticated.
In a terminal, install the LaTeX environment required for the paper-writing workflow and verify that latexmk and pdfinfo are available. The documented command installs MacTeX; the source notes that BasicTeX may be used instead by replacing the first installation command with the alternative shown in its comment.
In a terminal, clone ARIS to the stable local path shown. If choosing another clone location, use that same location in all later installer and MCP-server paths.
In a terminal, change to the project that should receive the skill. Replace ~/your-paper-project with the actual project directory.
cd ~/your-paper-project
In that project directory, run the selective installer for paper-writing. Its hard dependencies are included automatically. Review the displayed installation plan and type y when prompted to confirm.
bash ~/aris_repo/tools/install_aris.sh --skills paper-writing # by skill (hard deps auto-included)
In a terminal, remove any older codex registration and register ARIS's Codex bridge globally in Claude Code. If the repository was cloned somewhere other than ~/aris_repo, replace $HOME/aris_repo in the second command with the absolute clone path while retaining the registered name codex.
claude mcp remove codex -s user 2>/dev/null # drop an older `codex mcp-server` registration, if you have one
claude mcp add codex -s user -- python3 "$HOME/aris_repo/mcp-servers/codex-exec/server.py"
Restart Claude Code so it reloads the MCP configuration.
In a terminal after restarting Claude Code, verify that the codex MCP server is connected.
<results-path>: path to a directory or files containing JSON/CSV experiment results
<baseline-name>: configuration to use as the comparison baseline
Optional: primary metric and whether higher or lower is better
Check the repository for this skill’s setup.
Use the analyze-results skill to analyze the JSON or CSV files under <results-path>. Build a raw comparison table organized by model or configuration, compute each result's delta versus <baseline-name>, and report concise numbered findings. If multiple seeds are present, include mean and standard deviation. Flag suspicious values or outliers and suggest only the next experiments directly supported by the observed data. Do not modify project documentation; propose any useful note as text instead.
<experiment-identifier>: run, screen session, application, or job identifier
<project-notes-or-launch-summary>: path or text identifying the local, SSH, Vast.ai, or Modal backend and access details
<baseline>: optional compatible baseline and expected metric
Any required credentials or authenticated access for the actual backend
Check the repository for this skill’s setup.
Use the monitor-experiment skill to perform one read-only status check for <experiment-identifier>. Determine the actual backend from <project-notes-or-launch-summary>, inspect the available logs or result files, and summarize raw metrics, progress, status, and any NaN, divergence, or crash evidence. Compare only against <baseline> when the configuration is compatible. Do not send notifications, terminate resources, or begin recurring monitoring.
<section>: the section to structure, such as Introduction, Design, or Evaluation
<research-summary>: the problem, gap, system design, evidence, and intended contributions
Check the repository for this skill’s setup.
Use the writing-systems-papers skill to create a paragraph-level structural blueprint for the <section> of a paper targeting <venue>. Use <research-summary> to assign each paragraph a specific role and estimate a page budget consistent with a 10–12 page systems paper. Return an outline and writing guidance only; do not draft unsupported results, claims, or citations.
Claude
Codex
Cursor
Gemini CLI
OpenClaw
Copilot
Windsurf
Amp
Hermes
Antigravity
Goose
DeepSeek
Cline
OpenHands