devfun-org/poker-arena-starter-kit
Developer tools이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기
Agent-driven poker bot starter kit for dev.fun Arena. Three paths (learn/build/iterate), four-stage curriculum, sandbox-safe.
개요
A skill that turns any coding agent into a poker bot builder for dev.fun Arena's Poker Eval benchmark. Then say "go". The agent reads SKILL.md, asks you 1-2 strategy questions, writes your decide() function, runs local validation + an Arena preview, iterates on failures, and submits the final 500-hand match when you approve. Total ~30-60 min, mostly autonomous. Works with (most use today), plus Claude Code, Codex CLI, Cursor, Gemini CLI, Copilot, OpenHands, and others (Aider, Windsurf, Continue). The product is . The CLI binary stays pokerkit (so ./pokerkit run still works). "PokerKit" by itself is the name of the upstream Python poker engine (prinai/pokerkit) that we depend on — not this product. Poker Eval: https://arena.dev.fun/poker-eval ⚠️ Disconnecting mid-match or resuming later can cause timeouts and invalidate your leaderboard result. Run the full 500 hands in one go (~15 min) for an accurate score. Competition IDs live in .env.
README
원문 README의 일부입니다. 원문 보기.
Arena Starter Kit
A skill that turns any coding agent into a poker bot builder for dev.fun Arena’s Poker Eval benchmark.
Give it to your agent
Paste this URL into your agent’s chat:
https://github.com/devfun-org/poker-arena-starter-kit/blob/main/SKILL.md
Or install via the skills CLI:
npx skills add devfun-org/poker-arena-starter-kit
Then say “go”. The agent reads SKILL.md, asks you 1-2 strategy
questions, writes your decide() function, runs local validation + an
Arena preview, iterates on failures, and submits the final 500-hand
match when you approve. Total ~30-60 min, mostly autonomous.
Works with Hermes, OpenClaw, and Zo Computer (most use today), plus Claude Code, Codex CLI, Cursor, Gemini CLI, Copilot, OpenHands, and others (Aider, Windsurf, Continue).
Naming. The product is Arena Starter Kit. The CLI binary stays
pokerkit(so./pokerkit runstill works). “PokerKit” by itself is the name of the upstream Python poker engine (prinai/pokerkit) that we depend on — not this product.
Where your agent plays
Poker Eval: https://arena.dev.fun/poker-eval
Competitions (PVE vs 5 reference bots):
| Season | Hands | Wall time | Use |
|---|---|---|---|
| S1 | 500 | ~15 min | Direction-check each iteration |
⚠️ S1 is a single continuous 500-hand match. Disconnecting mid-match or resuming later can cause timeouts and invalidate your leaderboard result. Run the full 500 hands in one go (~15 min) for an accurate score.
Competition IDs live in
.env.exampleand are also discoverable at runtime viaGET /api/arena/competition/list-active.
Two paths — pick the right one for the job
| Local dev loop | Arena Evaluation | |
|---|---|---|
| Purpose | Fast iteration on decide() while developing |
Real benchmark — scores against Arena’s reference panel |
| Speed | 50 ms (unit tests) — 1 s per 200 hands (self-play) | ~15 min (S1 / 500 hands) |
| Network | None | Live Arena API |
| Opponent | Simple heuristic bots (tight/loose/random) | 5 server-side reference bots from dev.fun |
| When to use | Every time you edit decide(). Cheap, fast, no API limits. |
When you want a real bb/100 score on the leaderboard. |
| Commands | pokerkit test, pokerkit selfplay, pokerkit run --dry-run |
pokerkit run |
Develop locally, evaluate on Arena. Final 500-hand runs always go through Arena — that’s the only place the reference panel exists.
Quick start
추천 도구
다른 키워드를 입력하거나 필터를 제거해 보세요.
설치
npx skillfish add devfun-org/poker-arena-starter-kit