Arena (Parallel Candidate Tournament)

Fan out N parallel subagents on the same task, cross-judge them against a rubric, pick a base, and graft the strongest parts of the losers into one synthesized artifact.

AutomationAdvanced★ 1,816⑂ 195AI score 8/10Last updated: Oct 11, 2026

What it does

Spawns N parallel candidates on an identical prompt (mixed models, or the same model N times for generation-bound work).
Forces you to derive a 3–6 criteria rubric up front, so selection is criterion-by-criterion rather than vibes.
After candidates finish, spawns a read-only cross-judge on a different model family to score and recommend a base independently.
Picks the base on maintainability/extensibility, then hand-grafts the one or two genuinely valuable ideas from each losing candidate.
Verifies the synthesis and writes a synthesis note recording base, grafts, rejections, dropouts, and verification result.
Convergence across candidates is treated as a ship signal; wild divergence is treated as evidence the framing phase failed, triggering a re-frame instead of averaging.

Who it's for

Engineers facing decisions where a single first attempt locks in the wrong shape (API surfaces, architecture, big refactors).
Anyone running multi-agent workflows who wants structured model-vs-model comparison.
Teams that need a written rationale for design decisions.

Examples

"Refactor this auth module — arena this" → four models build in separate git worktrees → judge scores them → cleanest module boundary becomes the base, another candidate's error handling is grafted in.
Designing a new CLI's command surface: three of four candidates converge on the same shape → ship the consensus, note the convergence, skip grafting.
Performance-optimization candidates come back wildly different → diagnosed as under-specified framing, so the rubric is rewritten and the arena re-run.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/michael-denyer/pstack-claude/HEAD/plugins/pstack/skills/arena/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)

Download the ZIP with the button below.
In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.

↓ Download ZIP

Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the plugins/pstack/skills/arena folder from the GitHub repo michael-denyer/pstack-claude into my ~/.claude/skills/arena/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/michael-denyer/pstack-claude.git /tmp/pstack-claude && mkdir -p ~/.claude/skills && cp -r /tmp/pstack-claude/plugins/pstack/skills/* ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

Clone the repo: git clone https://github.com/michael-denyer/pstack-claude.git
Create the skills directory: mkdir -p ~/.claude/skills
This skill references sibling pstack files (poteto-mode/references/*, the principle skills), so copy the whole bundle rather than arena alone: cp -r pstack-claude/plugins/pstack/skills/* ~/.claude/skills/
Optional: run /setup-pstack in Claude Code to generate the pstack-models.md override sheet. Without it, arena falls back to the defaults (opus, fable, sonnet, haiku).
Restart Claude Code, then invoke /arena or say "throw it in the arena".
Note: running N background subagents multiplies token and model cost — reserve it for high-stakes artifacts.

View source on GitHub ↗License: MIT