[{"data":1,"prerenderedAt":506},["ShallowReactive",2],{"project-47149586":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":29,"sentiment":30,"stance_mix":31,"scored_comments":35,"aspects":36,"strip":44,"post_text":46,"what":47,"said":78,"maker_in_thread":193,"shape":205,"grid":208,"attempts":462,"same_week":470,"maker_shows":40,"maker_hidden":498,"reader_version":499,"card":500},"47149586","LLM Skirmish","A real-time strategy game that AI agents can play","title","Show HN: A real-time strategy game that AI agents can play","https:\u002F\u002Fllmskirmish.com\u002F","llmskirmish.com","__cayenne__",1772013765,2026,220,78,245,4366,6,"notable","ai","llm game benchmark","web_app",[24,25,26,27,28],"llm benchmark","ai agents","real-time strategy","screeps","llm game","curious",0.141,{"positive":32,"neutral":33,"mixed":34,"critical":18},10,20,1,37,{"design":37,"idea":38,"execution":41,"performance":43},{"positive":34,"critical":34},{"positive":39,"critical":40},5,4,{"positive":42,"critical":34},0,{"positive":42,"critical":42},[34,45,39,34,42,42,42,42,42,42,42,34,45,34,42,39,45,45,39,39,34,42,45,39,42,45,45,45,45,45,39,39,34,39,42,42,39,39,45,42],3,"I've liked all the projects that put LLMs into game environments. It's been a weird juxtaposition, though: frontier LLMs can one-shot full coding projects, and those same models struggle to get out of Pokémon Red's Mt. Moon.\n\nBecause of this, I wanted to create a game environment that put this generation of frontier LLMs' top skill, coding, on full display.\n\nTen years ago, a team released a game called Screeps. It was described as an \"MMO RTS sandbox for programmers.\" The Screeps paradigm of writing code and having it executed in a real-time game environment is well suited to LLMs. Drawing on ",{"problem":48,"audience":54,"pricing":54,"motivation":55,"time_to_build":54,"stack":59,"openness":76,"stage":77,"team_size":76},{"value":49,"source":50,"evidence":51},"Shows how well frontier LLMs write code that plays a real-time strategy game against other LLMs.","maker",[52],{"id":53},"post",null,{"value":56,"source":50,"evidence":57},"Put LLM coding skill on display in a game environment, since LLMs struggle at games despite coding well.",[58],{"id":53},[60,64,68,72],{"value":61,"source":50,"evidence":62},"Google Cloud Run",[63],{"id":53},{"value":65,"source":50,"evidence":66},"isolated-vm",[67],{"id":53},{"value":69,"source":50,"evidence":70},"Cloudflare",[71],{"id":53},{"value":73,"source":50,"evidence":74},"Screeps open source API",[75],{"id":53},"unknown","launched",{"praise":79,"objections":103,"requests":132,"questions":151,"alternatives":164},[80,96],{"value":81,"source":82,"evidence":83},"Fun to watch and an interesting testbed for agent capability","commenter",[84,88,92],{"id":85,"by":86,"text":87},"47149738","hmontazeri","This is actually fun to watch :D",{"id":89,"by":90,"text":91},"47151014","david3289","This is a really interesting direction. RTS games are a much better testbed for agent capability than most static benchmarks because they combine partial observability, long-term…",{"id":93,"by":94,"text":95},"47150293","myky22","Love it! I have a similar inuitiom in my use of Gemini (3 and 3.1). Great at \"turn 1\" task but degrades faster than opus or gpt.",{"value":97,"source":82,"evidence":98},"Sandbox hardening story is interesting",[99],{"id":100,"by":101,"text":102},"47164695","jamiecode","The sandbox hardening story is the most interesting thing here. GPT trying to cheat by reading opponent strategies is a perfect illustration of a broader problem - the objective…",[104,111,118,125],{"value":105,"source":82,"evidence":106},"Not truly real-time strategy; more like a turn-based autobattler",[107],{"id":108,"by":109,"text":110},"47181651","383toast","RTS is misleading, this is a turn based autobattler at best.",{"value":112,"source":82,"evidence":113},"Visualization is hard to follow and units are unclear",[114],{"id":115,"by":116,"text":117},"47149790","wongarsu","I know visualization is far from the most important goal here, but it really gets me how there's fairly elaborately rendered terrain, and then the units are just unnamed roombas…",{"value":119,"source":82,"evidence":120},"Leaderboard incentives were gamed by resetting scores on upload",[121],{"id":122,"by":123,"text":124},"47153045","FusspawnUK","Took a crack at this earlier. the leader board is a little weird. seems to be like 2 real dudes and the rest are fake profiles. a Scores resetting on each new upload also…",{"value":126,"source":82,"evidence":127},"Unclear whether LLMs learn across rounds",[128],{"id":129,"by":130,"text":131},"47159854","builder51216","But does LLM actually learn from each round? The chart does not show improvements in win rate across rounds... And what is the game state here exactly? Is LLM able to even…",[133,137,144],{"value":134,"source":50,"evidence":135},"Test newer models such as Claude 4.6 Opus and GPT 5.3 Codex",[136],{"id":53},{"value":138,"source":82,"evidence":139},"Add a light mode",[140],{"id":141,"by":142,"text":143},"47205506","julienreszka","I wish there was a light mode",{"value":145,"source":82,"evidence":146},"Let humans play",[147],{"id":148,"by":149,"text":150},"47156210","JoeDaDude","How about opening up the game for humans to play? Can you beat your AI?",[152,156,160],{"value":153,"source":82,"evidence":154},"Whether self-play training loops are supported",[155],{"id":89,"by":90,"text":91},{"value":157,"source":82,"evidence":158},"How isolated-vm enforces the sandbox boundary",[159],{"id":100,"by":101,"text":102},{"value":161,"source":82,"evidence":162},"Whether results have confidence intervals or multiple runs",[163],{"id":100,"by":101,"text":102},[165,172,179,186],{"value":166,"source":82,"evidence":167},"Screeps",[168],{"id":169,"by":170,"text":171},"47152454","giancarlostoro","Reminds me of Screeps, which I never took the time to fully play, but now I'm wondering if using Claude Code to play Screeps is cheating. Additionally, Screeps lets you host your…",{"value":173,"source":82,"evidence":174},"StarCraft AI competition",[175],{"id":176,"by":177,"text":178},"47150047","busfahrer","This reminds me of this yearly StarCraft AI competition (since 2010), however I think it uses a special API that makes it easy for bots to access the game Edit: Forgot link:…",{"value":180,"source":82,"evidence":181},"MicroRTS",[182],{"id":183,"by":184,"text":185},"47152711","jonbaer","Might be worth digging through MicroRTS too, https:\u002F\u002Fgithub.com\u002FFarama-Foundation\u002FMicroRTS (it's been abandoned), Python RL interface @…",{"value":187,"source":82,"evidence":188},"C++Robots",[189],{"id":190,"by":191,"text":192},"47160043","anotherevan","For some reason this reminds me strongly of an old play-by-email game called C++Robots[1]. I loved the idea, but the timeslice limitation[2] I found too annoying. I had youthful…",{"comments":194,"answered":39,"key":195},8,[196,199,202],{"id":197,"text":198},"47152695","funny you mention this… I have a new project that is going in this direction",{"id":200,"text":201},"47153074","Very interested in self-play training loops, but I do like codegen as an abstraction layer. I am planning to make it available as an RL environment at some point",{"id":203,"text":204},"47153225","Tweaking the leaderboard match assignment logic now to prevent these bad incentives - definitely want people to iterate! I had started with the Silicon Valley characters as a one…",{"commenters":206,"max_depth":40,"maker_replies":194,"first_comment_mins":33,"lifespan_hours":207},65,97,[209,211,215,219,223,228,230,234,237,238,239,243,247,250,252,255,258,260,263,265,268,271,273,275,278,281,283,286,289,291,293,297,300,304,306,307,311,313,317,321,325,329,331,333,335,337,341,345,349,353,357,361,364,368,370,373,377,381,383,386,388,393,396,399,401,405,407,411,413,417,420,424,426,430,434,438,442,447,450,452,454,459],[115,42,45,210,116,42],"project\u002Fmixed\u002Fdesign",[212,42,34,213,214,42],"47149813","project\u002Fcritical\u002Fdesign","embedding-shape",[216,42,45,217,218,42],"47152762","project\u002Fneutral\u002Fdesign","infecto",[220,34,45,221,222,42],"47150607","project\u002Fneutral\u002Fidea","arscan",[224,225,42,226,227,42],"47150194",2,"other\u002Fneutral\u002Fnone","EwanG",[89,45,39,229,90,42],"project\u002Fpositive\u002Fidea",[231,45,34,232,233,42],"47151136","project\u002Fcritical\u002Fidea","dmos62",[235,45,45,221,236,42],"47151950","degenerate",[197,45,42,226,11,34],[200,45,42,221,11,34],[240,45,42,241,242,42],"47158180","commenter\u002Fneutral\u002Fnone","drakinosh",[244,40,42,245,246,42],"47150264","other\u002Fcritical\u002Fnone","dakolli",[248,40,42,241,249,42],"47150282","p-e-w",[251,40,42,241,246,42],"47150492",[253,40,42,241,254,42],"47152424","hu3",[256,40,42,226,257,42],"47150299","LatencyKills",[259,40,42,226,246,42],"47150426",[261,40,42,241,262,42],"47150539","stalfie",[264,40,42,241,257,42],"47150840",[266,40,42,241,267,42],"47159166","MaybiusStrip",[269,40,42,241,270,42],"47150579","paganel",[272,40,42,241,257,42],"47150746",[274,40,42,241,254,42],"47152319",[276,40,34,232,277,42],"47151689","Applejinx",[279,39,42,226,280,42],"47149770","egeozcan",[282,18,45,221,233,42],"47151184",[284,18,42,226,285,42],"47152729","chasd00",[122,287,34,288,123,42],7,"project\u002Fcritical\u002Fexecution",[203,287,42,290,11,34],"maker\u002Fneutral\u002Fexecution",[292,287,42,290,11,34],"47156994",[294,194,39,295,296,42],"47150946","project\u002Fpositive\u002Fnone","mpeg",[298,194,42,241,299,42],"47151055","Razengan",[301,302,45,221,303,42],"47155907",9,"Ross00781",[305,302,42,221,11,34],"47156150",[176,32,45,221,177,42],[308,32,45,309,310,42],"47150290","project\u002Fneutral\u002Fexecution","KeplerBoy",[190,312,45,221,191,42],11,[314,315,39,229,316,42],"47150697",12,"mitchm",[318,319,39,295,320,42],"47155418",13,"yuppiepuppie",[322,323,42,226,324,42],"47150135",14,"ph4rsikal",[326,327,45,221,328,42],"47149919",15,"PeterUstinox",[183,330,42,226,184,42],16,[129,332,34,232,130,42],17,[334,332,42,290,11,34],"47160653",[93,336,42,226,94,42],18,[338,339,42,226,340,42],"47153100",19,"sails",[342,33,45,343,344,42],"47149964","project\u002Fneutral\u002Fperformance","cahaya",[346,347,45,221,348,42],"47170246",21,"Sophira",[350,351,39,229,352,42],"47149872",22,"datawars",[354,355,42,226,356,42],"47163728",23,"angusik",[358,359,42,226,360,42],"47175036",24,"nirav72",[362,359,42,226,363,42],"47175176","AuthAuth",[365,366,45,221,367,42],"47151216",25,"FrustratedMonky",[148,369,45,221,149,42],26,[371,369,42,226,372,42],"47156330","midiguy",[374,375,45,221,376,42],"47150664",27,"Lerc",[378,379,42,226,380,42],"47153940",28,"nickpsecurity",[169,382,45,221,170,42],29,[384,382,42,241,385,42],"47152664","jack6e",[387,382,42,241,170,42],"47153103",[389,390,45,391,392,42],"47153367",30,"project\u002Fneutral\u002Fnone","tantalor",[394,390,42,395,11,34],"47153413","maker\u002Fneutral\u002Fdesign",[397,390,39,398,392,42],"47154120","project\u002Fpositive\u002Fdesign",[85,400,39,295,86,42],31,[402,403,42,226,404,42],"47150859",32,"GlacierFox",[108,406,34,232,109,42],33,[408,409,45,221,410,42],"47152485",34,"jeffro_rh",[141,412,45,217,142,42],35,[414,415,39,295,416,42],"47161047",36,"medi_naseri",[418,35,42,226,419,42],"47160847","kookster310",[421,422,42,226,423,42],"47151227",38,"chimpanzee2",[425,422,42,226,254,42],"47152097",[427,428,42,226,429,42],"47155804",39,"cowboylowrez",[431,432,39,229,433,42],"47155852",40,"bombashell",[435,436,45,221,437,42],"47149890",41,"xanth",[439,440,42,226,441,42],"47155554",42,"burgerone",[443,444,39,445,446,42],"47151299",43,"maker\u002Fpositive\u002Fidea","5o1ecist",[448,444,42,226,449,42],"47153708","softfalcon",[100,451,45,309,101,42],44,[453,451,42,290,11,34],"47167318",[455,456,42,457,458,42],"47181317",45,"other\u002Fneutral\u002Fidea","agenthustler",[460,461,42,226,101,42],"47168983",46,[463,466,469],{"id":464,"t":465,"points":39,"comments":225},"46885863",1770213576,{"id":467,"t":468,"points":40,"comments":34},"47061773",1771427324,{"id":4,"t":12,"points":14,"comments":15},[471,476,481,485,489,494],{"id":472,"name":473,"points":474,"category":475},"47075124","Micasa",657,"productivity",{"id":477,"name":478,"points":479,"category":480},"47114530","CIA World Factbook Archive",496,"learning_science",{"id":482,"name":483,"points":484,"category":20},"47104667","ntransformer",395,{"id":486,"name":487,"points":488,"category":20},"47131225","Steerling-8B",328,{"id":490,"name":491,"points":492,"category":493},"47123631","PgDog",326,"infrastructure",{"id":495,"name":496,"points":497,"category":20},"47143755","Moonshine Open-Weights STT models",316,false,"v2",{"id":501,"name":5,"line":6,"ai":498,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":502,"top":18,"kw":503,"maker":11,"domain":10,"tier":19,"attempts":45,"strip":504,"url":9,"mix":505,"praise":81,"objection":105,"sub":21},47149586,0.14,[24,25,26,27,28],"1351000000013105335510350333335515005530",[32,33,34,18],1793991145522]