Create New Experiment

Sets up a new autoresearch experiment interactively or via command-line arguments, collecting domain, experiment name, target file, eval command, metric, direction, evaluator, and scope. It validates inputs, runs a setup script, and reports the experiment path and baseline metric.

✨ What it does

  • Collects required parameters interactively or from arguments.
  • Verifies target file existence.
  • Provides a list of built-in evaluators.
  • Supports project or user scope storage.
  • Runs setup_experiment.py to create the experiment.
  • Reports experiment path, branch name, and baseline metric.

🎯 When to use it

  • When the user runs /ar:setup to start a new autoresearch experiment.
  • When the user asks to begin optimizing a file with the autoresearch loop.
  • When the user needs to list existing experiments or available evaluators.

🚀 How to use

Trigger: /ar:setup

Trigger with the slash command /ar:setup or by asking to set up a new autoresearch experiment. Provide parameters as arguments or answer interactive prompts. Example prompts:

/ar:setup
/ar:setup engineering api-speed src/api.py "pytest bench.py" p50_ms lower

📄 Output: A configured experiment with a path, branch name, and baseline metric, ready for /ar:run or /ar:loop.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/autoresearch-agent/skills/setup .claude/skills/setup

Skill source: engineering/autoresearch-agent/skills/setup/SKILL.md

⚠️ Good to know

Requires the setup_experiment.py script and a valid target file; interactive mode asks for all parameters.

❓ FAQ

What does /ar:setup do?

It sets up a new autoresearch experiment by collecting configuration like domain, target file, eval command, metric, direction, and evaluator, then runs a setup script.

Can I provide parameters directly?

Yes, you can pass them as arguments, e.g., /ar:setup engineering api-speed src/api.py "pytest bench.py" p50_ms lower.

What are the built-in evaluators?

Built-in evaluators include benchmark_speed, benchmark_size, test_pass_rate, build_speed, memory_usage, llm_judge_content, llm_judge_prompt, and llm_judge_copy.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.