Create New Session
Initializes a new AgentHub collaboration session, creating the .agenthub directory structure, generating a session ID, and configuring evaluation criteria. It supports both interactive and command-line modes, and optionally captures a baseline metric for later comparison.
✨ What it does
- Creates .agenthub directory structure and session ID.
- Accepts task, agent count, eval command, metric, direction, and base branch via CLI arguments or interactive prompts.
- Supports LLM judge mode when no eval command is provided.
- Captures a baseline metric by running the eval command and stores it in config.yaml.
- Displays session summary and next steps.
🎯 When to use it
- When the user runs /hub:hub-init or asks to start a multi-agent competition on a task.
- When setting up a new AgentHub session with a defined task, number of agents, and optional evaluation criteria.
- When you need to initialize a session before spawning agents with /hub:spawn.
🚀 How to use
Trigger: /hub:hub-init
Trigger by running the slash command /hub:hub-init or asking to start a multi-agent competition. Provide the task (required), agent count (default 3), and optionally an eval command, metric name, direction, and base branch. If no arguments are given, the skill prompts for each parameter interactively.
Example prompts:
/hub:hub-init --task "Optimize API" --agents 3 --eval "pytest bench.py" --metric p50_ms --direction lower
/hub:hub-init --task "Refactor auth" --agents 2
📄 Output: An initialized AgentHub session with a session ID, configuration, and optional baseline metric.
📦 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/agenthub/skills/hub-init .claude/skills/hub-initSkill source: engineering/agenthub/skills/hub-init/SKILL.md
⚠️ Good to know
Requires the skill's Python script to be available; baseline capture is optional and may fail without halting initialization.
❓ FAQ
What happens if I don't provide an eval command?
The session uses LLM judge mode, meaning no automated evaluation is set up; agents will be judged by an LLM instead.
How is the baseline metric captured?
After session creation, the skill runs the eval command in the current directory, extracts the metric value from stdout, and appends it to the session's config.yaml as 'baseline: {value}'.
What is the next step after initialization?
Run /hub:spawn to launch the agents, or /hub:spawn {session-id} if multiple sessions exist.
🤖 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.