Coordinate parallel AI agents in isolated git worktrees and merge the winning branch.
npx skills add https://github.com/Patasse97/claude-skills --skill agenthub
Please help me install this Agent Skill. Skill: agenthub Source: https://github.com/Patasse97/claude-skills/tree/main/engineering/agenthub Command: npx skills add https://github.com/Patasse97/claude-skills --skill agenthub
Orchestrates a multi-agent session where several AI agents operate in isolated git worktrees to generate diverse solutions and automatically select a winner.
Run /hub:init to create a session, then spawn agents, evaluate results, and merge the winner.
High-intent search queries and answers about installing and using this skill.
Coordinate parallel AI agents by spawning multiple subagents in isolated git worktrees to generate diverse solutions simultaneously. Use slash commands, templates, and a message board to orchestrate the parallel exploration and automatically evaluate results to select a winning branch.
The best way to run multi-agent orchestration for software engineering is spawning N parallel subagents in isolated git worktrees that compete on the same problem. Results are evaluated by a metric or LLM judge, and the winning branch is merged while preserving the git-DAG.
The git-DAG workflow evaluates parallel agent results by applying a metric or an LLM judge to the solutions generated in isolated worktrees. After evaluation, it automatically selects the winner and merges that specific branch while preserving the directed acyclic graph history.
Yes, you can use this multi-agent orchestration for research, content generation, and optimization tasks. The session applies slash commands and templates to spawn parallel agents that explore diverse solutions across these domains before evaluating and merging the winning branch.
No external dependencies are required to start an isolated worktree collaboration session. The Skill provides built-in scripts and references, allowing you to run the /hub:init command to create a session, spawn agents, and monitor results directly without external packages.
You should not use parallel agents in isolated worktrees for tasks requiring sequential execution or shared mutable state across steps. The workflow is designed for parallel exploration where agents compete independently, making it unsuitable for highly interdependent or strictly linear processes.
thkim2015Manage Git worktrees for isolated parallel SPEC development with automatic branch registration.
rocky2431Enforce one-task-one-branch-merge-delete cycles with main always deployable.
Ibrahim-3dCoordinate multi-agent workflows with a file-based message bus.
maystudiosCoordinate parallel agent execution with git worktree isolation and PASS/FAIL results.
