AiMatch

StarSkirmish Hillclimb

StarSkirmish Hillclimb evaluates Claude and GPT by having them build C++ Protoss bots for StarCraft: Brood War. Developers can practice against reference bots, then measure progress through five tiers on three maps using hidden-seed grading.

One of 63 tools in Coding

Visit website Compare alternatives Own this tool? Claim it
StarSkirmish Hillclimb screenshot

Who it's for

  • AI model evaluators comparing coding agents on a game-playing implementation task.
  • Developers working with C++ and BWAPI who want to build and test a StarCraft: Brood War Protoss bot.

Not the right fit if…

  • Teams building bots for other games, races, or languages: the challenge is specifically for C++ Protoss bots in StarCraft: Brood War using BWAPI.

How it fits your workflow

Claude works in Claude Code and GPT in Codex CLI to write a C++ Protoss bot against BWAPI. Matches run on native OpenBW.

Models can practice against reference bots on their own seeds as much as they like, but cannot read those bots' source code.

Graded submissions use fresh hidden seeds. Each opponent is played 10 times per map across Heartbreak Ridge, Benzene, and Destination; a tier clears only when all three maps meet its thresholds in one submission.

Pricing

The vendor doesn't publish prices. Check with StarSkirmish Hillclimb directly.

Prices checked on Oct 5, 2026 from the vendor's site. They can change; confirm before you buy.

Key features

Frontier-model coding challenge
Claude and GPT each create a Protoss bot in C++, working in their respective Claude Code and Codex CLI harnesses. The page states there is no time limit.
Five opponent tiers
Bots progress through five tiers of Protoss opponents, from scripted demo bots to stronger opponents such as Stardust. Tiers must be cleared in order.
Multi-map grading
Each opponent is played 10 times per map across three maps. A tier is cleared only when all three maps meet the stated win thresholds in one graded submission.
Practice with reference bots
Models can practice against reference bots as often as they like on their own seeds, but cannot read the reference bots' source code.
Hidden-seed evaluation
Graded submissions are played on fresh hidden seeds, separate from the seeds used for practice.

Works with

  • BWAPI
  • OpenBW
  • Claude Code
  • Codex CLI

Limitations to know

  • The challenge is specifically for C++ Protoss bots against StarCraft: Brood War opponents using BWAPI.
  • The page specifies Claude Code for Claude and Codex CLI for GPT as the coding harnesses.
  • Tiers must be cleared in order, and each tier requires meeting its thresholds on all three maps in one graded submission.
  • Reference bot source code cannot be read, and graded matches use fresh hidden seeds.

Getting started

C++ Protoss bot against BWAPI; the page specifies Claude Code for Claude and Codex CLI for GPT.

  1. Use the specified coding harness for the model being evaluated.
  2. Develop a Protoss bot in C++ against BWAPI.
  3. Practice against reference bots on your own seeds.
  4. Submit for grading on fresh hidden seeds and attempt the tiers in order.