arena

An agent skill by cursor, from cursor/plugins. Tags: analysis, brainstorming, evaluation, generative, multi-agent.

What it does

Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it. Use for /arena, 'arena this', 'throw it in the arena', or when one attempt at a non-trivial artifact would lock in the wrong shape.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add cursor/plugins --skill arena

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/cursor/plugins
cp -r plugins/pstack/skills/arena ~/.claude/skills/arena

Safety box score

Grade A, minimal risk: 7 of 100 risk points, rated 2026-09-23. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill orchestrates an internal “arena” of parallel subagents to draft, compare, and synthesize one artifact. The main risks are limited filesystem reach into `~/.cursor/rules/pstack-models.mdc`, temporary output directories/worktrees, and higher agent autonomy from spawning several model calls; it does not itself run shell commands, fetch remote code, or touch secrets.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	0 (none)	No shell commands, script execution, or code evaluation is defined in `SKILL.md`; it only directs how to coordinate subagents.
Network access	0 (none)	The skill does not download from or send data to remote hosts; it only references local config like `~/.cursor/rules/pstack-models.mdc`.
Filesystem reach	1 (low)	It reads a home-directory config file (`~/.cursor/rules/pstack-models.mdc`) and may create per-candidate output under a git worktree or `/tmp/arena-<slug>/candidate-<n>/`, which is limited but outside the project tree.
Secrets & credentials	0 (none)	There is no request to access tokens, keys, `.env` files, browser sessions, or other credentials.
Privilege & persistence	0 (none)	The skill does not ask for sudo, global installs, services, or OS/shell settings changes.
Instruction hijack surface	0 (none)	It does not instruct the agent to ignore rules, hide actions, follow remote instructions, or adopt deceptive personas.
Irreversible actions	1 (low)	It explicitly says to "Spawn N parallel attempts" and use `run_in_background: true`, but this is internal task orchestration rather than an external side effect like posting, paying, or deleting.
Transparency	0 (none)	The skill is plain, fully readable markdown; there are no binaries, encoded blobs, or downloaded installers.

Before installing

Confirm you are okay with the skill reading `~/.cursor/rules/pstack-models.mdc` if it exists.
Check that creating temporary outputs under `/tmp/arena-*` or per-task worktrees is acceptable.
Expect multiple model calls per use (N candidates plus a judge); budget accordingly.

Rated 2026-09-23 by an automated review of 1 file from the skill's repository.

Source

Repository: cursor/plugins (all skills from this repository)
Path: pstack/skills/arena/SKILL.md
Branch: main
Collection: cursor
Updated: 2026-09-23

principle-guard-the-context-window — Apply when context is filling up: large outputs, long files, repeated reads, fan-out planning.
reflect — Spawn three parallel review subagents over the active transcript, surface learnings, and route each to a concrete edit on an existing skill.
interpreting-culture-index — Interprets Culture Index (CI) surveys, behavioral profiles, and personality assessment data.
let-fate-decide — Draws the 12 Houses of the Zodiac Tarot spread to inject entropy into planning when prompts are vague, ambiguous, or casually delegated.
marketing-council — When the user wants multiple expert perspectives on a marketing question — a simulated board of advisors staffed by legendary marketers (Seth Godin.
product-analysis — Multi-path parallel product analysis with cross-model test-time compute scaling.