Spawns parallel candidate solutions, judges them against a rubric, and synthesizes the strongest result.
npx skills add https://github.com/lucasheriques/pstack --skill arena-lucasheriques
Please help me install this Agent Skill. Skill: arena Source: https://github.com/lucasheriques/pstack/tree/main/skills/arena Command: npx skills add https://github.com/lucasheriques/pstack --skill arena-lucasheriques
High-intent search queries and answers about installing and using this skill.
Use the arena workflow: frame a shared prompt and rubric, spawn N subagents in one message with run_in_background, each writing to its own isolated path. Then read every candidate, pick a base, and graft the best ideas from the others.
Derive 3-6 concrete gradeable criteria from the task, score each candidate criterion by criterion rather than on holistic feel, and cross-check with an independent judge model. Pick the base a future maintainer can extend most easily.
Use it when one attempt at a non-trivial artifact would lock in the wrong shape, such as architecture or API design decisions. Skip it for trivial tasks where the cost of N parallel runs exceeds the risk of a suboptimal first attempt.
The workflow proceeds with N-1 candidates and notes the dropout in the synthesis record. The remaining candidates are still judged, a base is picked, and grafts are applied normally.
Wild divergence means the framing phase was under-specified. The correct response is to reframe the task prompt and rubric and re-run the arena, rather than averaging incompatible designs into an incoherent result.
Leap0920Run Socratic one-question-at-a-time clarification with ambiguity scoring for product ideas.
semicolon-devteamVerify Kotlin projects via ktlint, compile, reactive, test and Jacoco checks before PR.
jnyrossFan out parallel cloud workers and aggregate their results into one report.
TalonT-OrgGenerate implementation plans by analyzing codebases and designing tests.
