Spawns parallel candidate solutions, judges them against a rubric, and synthesizes the strongest result.
npx skills add https://github.com/mmdmcy/fluttAIrbar --skill arena-mmdmcy
Please help me install this Agent Skill. Skill: arena Source: https://github.com/mmdmcy/fluttAIrbar/tree/main/plugins/pstack/skills/arena Command: npx skills add https://github.com/mmdmcy/fluttAIrbar --skill arena-mmdmcy
High-intent search queries and answers about installing and using this skill.
Use the arena workflow: frame one shared prompt with a rubric, then spawn N background subagents in a single message, each writing to its own output path. After all finish, read every candidate, pick a base, and graft the best ideas from the losers.
Score each candidate criterion by criterion against a rubric of 3-6 concrete gradeable items rather than judging holistically. A readonly cross-judge subagent on a different model family provides an independent verdict to compare against your own reading.
Use it for non-trivial artifacts where one attempt would lock in the wrong shape, such as new designs or architecture decisions. Skip it for simple or well-understood tasks where parallel generation adds cost without reducing design risk.
The workflow proceeds with N-1 candidates and records the dropout in the synthesis note. The remaining candidates are still judged, grafted, and verified normally.
Each candidate writes to its own git worktree or a separate /tmp/arena directory. Sharing one path across N writers creates shared mutable state, which corrupts outputs and makes judging unreliable.
different-aiCreate task-specific Git worktrees, initialize submodules, commit, and push changes.
drvossReview code changes for quality, correctness, and security.
cursorSpawns parallel candidate solutions, judges them against a rubric, and synthesizes the strongest result.
the-inconvenience-storeOrchestrate parallel subagent execution for development workflows using Git worktrees.
