@10xmylife: Reply to the questions in the comment section - How do you read the game state? We didn't actually have Jev look at scr…

X AI KOLs Timeline News
ai-gaming slay-the-spire python api-integration csharp-mod experiment cost-analysis

Summary

A developer explains using an AI agent named Jev to play Slay the Spire 2, employing a C# Mod to extract game data and a Python API for decision-making, while noting high token costs and mediocre performance.

Reply to the questions in the comment section - How do you read the game state? We didn't actually have Jev look at screenshots or use OCR. We used this C# Mod called STS2MCP to directly read hand cards, energy, health, enemy intentions, relics, and other data from inside the game, organize it into JSON, and pass it to the Python control program via a local HTTP interface. What gets sent to the model is structured data, ensuring Jev can respond. - How do you make Jev operate? Python sends the current game situation and a list of legal actions to TypeSafe's Jev API, like "play the nth card, attack which enemy, use a potion, or end the turn." Jev returns a structured choice, the Mod validates the state and executes it immediately, then reads the next step after settlement. The entire loop runs continuously via the program, with Codex/Kimi only handling setup and exceptions, not participating in per-card decisions. - How about the consumption? Played for about 20 minutes, only spent $0.0422, but codex's token usage was shockingly high, so we switched to kimi later. - How's the effect? Pretty mediocre, feels like it doesn't have enough info to make decisions, or rather, it doesn't understand the game at all.
Original Article
View Cached Full Text

Cached at: 09/20/26, 01:15 PM

Reply to the questions in the comment section

  • How do you read the game state? We didn’t actually have Jev look at screenshots or use OCR. We used this C# Mod called STS2MCP to directly read hand cards, energy, health, enemy intentions, relics, and other data from inside the game, organize it into JSON, and pass it to the Python control program via a local HTTP interface. What gets sent to the model is structured data, ensuring Jev can respond.

  • How do you make Jev operate? Python sends the current game situation and a list of legal actions to TypeSafe’s Jev API, like “play the nth card, attack which enemy, use a potion, or end the turn.” Jev returns a structured choice, the Mod validates the state and executes it immediately, then reads the next step after settlement. The entire loop runs continuously via the program, with Codex/Kimi only handling setup and exceptions, not participating in per-card decisions.

  • How about the consumption? Played for about 20 minutes, only spent $0.0422, but codex’s token usage was shockingly high, so we switched to kimi later.

  • How’s the effect? Pretty mediocre, feels like it doesn’t have enough info to make decisions, or rather, it doesn’t understand the game at all.

海明Dev (@10xmylife): 我正在小红书直播 Codex +Jev 挑战杀戮尖塔 2

速度飞快，我甚至看不清楚游戏画面… 就是玩得有点菜

Similar Articles

@10xmylife: cool 啊

X AI KOLs Following

An AI system named 'jev' is showcased playing Super Smash Bros. against itself, controlling multiple characters and making rapid decisions within fractions of a second, using over 22 million tokens.

@0xAikoDai: https://x.com/0xAikoDai/status/2057317742248931363

X AI KOLs Timeline

The author reflects on his experience with a new AI native RPG beta from the makers of AI Dungeon, identifying design challenges where AI-generated immersion breaks down during combat and system legibility, and calls for new design principles for AI-native games.