Arena

Give several T3 child candidates the same prompt, rank them against a rubric written before they run, pick a base and graft the best of the rest into it. Run with /arena when one attempt at a non-trivial artifact would lock in the wrong shape.

10 stars 1 forksAI, LLMs & Agents

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add thatmike1/claude-skills --skill arena
Manual install
git clone --depth 1 https://github.com/thatmike1/claude-skills.git /tmp/claude-skills
mkdir -p ~/.claude/skills
cp -r /tmp/claude-skills/arena ~/.claude/skills/arena

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# Arena

Ported from pstack's arena ([cursor/plugins](https://github.com/cursor/plugins), MIT, Lauren Tan) for T3, where every candidate is a child task Mike can open and steer.

The arena's shape is **one contract, many candidates, a rubric they never see**. Reach for a sibling when the shape differs:

- different pieces of work split across workers: `/orchestrate`
- a command that scores each try in seconds: `/swarm`
- the open question is which approach, before anything can be built: `/council`

## 1. Frame

Make a run directory, `/tmp/arena-<slug>/`, and write three files there before anything spawns.

1. `rubric.md`: the artifact in one line, then 3 to 6 criteria, each gradeable from the artifact alone (a check you can run, or a property you can point at in the diff). When the right answer is computable (expected outputs, a list the change must produce), compute it now into `ground-truth.*`. Candidates never see the rubric; it is the picker's tool, and showing it turns the run into teaching to the test.
2. `prompt.md`: the **contract**, sent verbatim to every candidate. It names the repo and files, the done condition and how to verify it, and ends with the return shape: where the work lives (`pwd` for a worktree, or the output path), the full diff or artifact, the verification command with its output verbatim, and three to five sentences on the alternatives it considered and rejected. Those rejected alternatives are graft material later.
Read the full skill on GitHub

Details

Skill name
arena
Path
arena/SKILL.md
Repo updated
18 November 2026

FAQ

What does the Arena skill do?

Give several T3 child candidates the same prompt, rank them against a rubric written before they run, pick a base and graft the best of the rest into it. Run with /arena when one attempt at a non-trivial artifact would lock in the wrong shape.

How do I install the Arena skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add thatmike1/claude-skills --skill arena. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Arena popular?

It lives in thatmike1/claude-skills, which has 10 GitHub stars and 1 forks.

Related skills

Afk Prompt

10

thatmike1/claude-skills

Write a copy-paste prompt that hands one or more tasks (usually beads issues) to a fresh Claude instance to run autonomously while the user is away — no questions, no waiting for input. Includes picking which tasks are safe to run unattended. Use when the user says they're going AFK (shower, lunch, errand, overnight), asks "which tasks can just run without me", or wants an autonomous/batch prompt. For attended sessions the user will steer, use the live-prompt skill instead.

AI, LLMs & Agents

AI CV Scanner

10

thatmike1/claude-skills

Mine Claude Code conversation history to answer questions about your AI experience. Scans session indexes, spawns parallel subagents to deep-dive relevant conversations, and drafts evidence-based answers. Use when user mentions CV, AI experience, questionnaire, scan my conversations, or needs to document their AI usage.

AI, LLMs & Agents

Attention Queue

10

thatmike1/claude-skills

Mike's attention queue, his day as a queue of one-thing blocks plus a dump shelf of raw thoughts, read and changed with the `aq` command. Use when asked what Mike is on right now or until when, to check or plan from his dump ("check the brain dump", "plan the day", "give it structure"), to work on what is next in his queue, to put work into his day or attach a report, link or note to a block, to reorder his blocks, or to read how a day went. Also whenever a thread does work that a block in today's queue stands for, so the block moves with it.

Productivity & Notes

Brief Output

9

lingengyuan/my-skills

精简回答、技术说明与工程文档,删除冗余信息与 AI 味表达,只保留完成任务所需的必要事实与技术决策。适用于精简输出、压缩改写、去除 AI 腔调以及全文一致性修订。

AI, LLMs & Agents

Cc Audit

10

thatmike1/claude-skills

Audits Claude Code setup and usage patterns, flagging anti-patterns (wrong launch dirs, context bloat, orphaned memories, missing CLAUDE.md) with severity-ranked fixes. Use when user says audit, health check, setup check, cc audit, or wants to improve their Claude Code workflow.

AI, LLMs & Agents

Gpt Pro

10

thatmike1/claude-skills

Hand a question to ChatGPT's Pro model or its image generator through the user's own chat window, off the Codex meter — the agent packs the prompt and context, the user pastes it and copies the answer back. Use when a plan or design wants a second opinion from another model family, for a large divergent idea run, for self-contained 3D or math-heavy code, for an image, or when the user says gpt-pro or asks to put something to ChatGPT.

AI, LLMs & Agents

Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.