Arena (Parallel Candidate Tournament)
Spawns N parallel candidates on the same task, cross-judges them against a rubric, picks a base, and grafts the strongest parts of the losers into it.
Dev & CodingAdvanced★ 3,867⑂ 267AI score 8/10Last updated: Nov 9, 2026
What it does
- Sends one identical prompt to N subagents in parallel so you get multiple competing versions of the same artifact.
- Runs a six-phase todolist: Frame → Fan out → Cross-judge → Pick → Graft → Verify.
- Derives a task-specific rubric (3–6 gradeable criteria) up front, then spawns a read-only judge from a different model family than the parent to score candidates.
- Picks one base (the one a maintainer can extend most easily), grafts only one or two worthwhile ideas per losing candidate by hand, then verifies the result.
- Includes detailed fallbacks for rejected model slugs, dropped seats, and candidates that produce no output.
Who it's for
- Developers working on artifacts where locking in the wrong shape early is costly: architecture, schemas, public APIs.
- Claude Code / Cursor users who run background subagents routinely.
- Users of the wider pstack skill set (poteto-mode, prove-it-works, etc.).
Example uses
- "Throw this refactor plan in the arena" → three implementations land in separate git worktrees, then get compared and merged.
- Designing a new CLI's command surface: run several distinct design directions and adopt the one with the cleanest boundary as the base.
- Writing a migration script: generation-bound work, so run the same model N times and graft only the error-handling logic from a loser.
· · · 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/backnotprop/pstack/HEAD/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.
Install in Claude Code
Let Claude do it — paste this into Claude Code
Install the skill I found on AI Skill Mart. Copy the skills/arena folder from the GitHub repo backnotprop/pstack into my ~/.claude/skills/backnotprop-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/backnotprop/pstack.git && mkdir -p ~/.claude/skills && cp -r pstack/skills/arena ~/.claude/skills/⚠ This is a third-party skill. Check the source repository before installing.
- Open a terminal.
- Clone the repo:
git clone https://github.com/backnotprop/pstack.git - Create the skills directory:
mkdir -p ~/.claude/skills - Copy the skill:
cp -r pstack/skills/arena ~/.claude/skills/ - (Recommended) Copy the skills it references too:
poteto-mode,prove-it-works,separate-before-serializing-shared-state,redesign-from-first-principles. - (Optional) Create
~/.claude/rules/pstack-models.mdwitharena runnersandarena cross-judge poollines listing the model slugs available in your harness; without it the skill uses its defaults. - Restart Claude Code and invoke it with
/arenaor "arena this".
View source on GitHub ↗License: MIT