pith. sign in

arxiv: 2606.24893 · v1 · pith:I73DEXWXnew · submitted 2026-05-29 · 💻 cs.CL

AgentOdyssey: Open-Ended Long-Horizon Text Game Generation for Test-Time Continual Learning Agents

Zheyuan Zhang , Zehao Wen , Alvin Zhang , Andrew Wang , Jianwen Xie , Daniel Khashabi , Tianmin Shu This is my paper

Pith reviewed 2026-06-28 21:57 UTC · model grok-4.3

classification 💻 cs.CL
keywords test-time continual learningtext game generationprocedural generationlong-horizon tasksepisodic memoryexplorationagent evaluation
0
0 comments X

The pith

AgentOdyssey generates open-ended text games to evaluate agents that learn continuously at test time.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents AgentOdyssey as a framework that creates procedurally generated text games featuring rich entities, dynamics, and long-horizon tasks. It positions agents in a setting where learning and inference interleave throughout deployment, moving past the standard view that learning ends after initial training. Evaluation tracks game progress alongside diagnostics for world knowledge acquisition, episodic memory, object and action exploration, action diversity, and model cost. Experiments across agent types show that performance improves with stronger base models yet stays well below human levels, while short-term memory aids multiple paradigms and supports longer effective horizons.

Core claim

AgentOdyssey procedurally generates open-ended text games with rich entities, world dynamics, and long-horizon tasks to place agents in a continuous setting that interleaves learning and inference throughout deployment, enabling multifaceted evaluation of exploration, episodic memory, world knowledge acquisition, and planning abilities.

What carries the argument

The AgentOdyssey framework of procedurally generated open-ended text games equipped with diagnostic metrics that measure test-time continual learning abilities.

If this is right

  • Stronger base models improve agent performance yet leave substantial headroom relative to human levels.
  • Short-term memory improves results across multiple agent paradigms and extends meaningful horizon length.
  • Current agents exhibit critical limits in exploration, episodic memory, and long-horizon planning.
  • Factors such as memory mechanisms influence how far agents can sustain effective behavior.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Diagnostic metrics from the games could guide targeted improvements in agent memory and adaptation modules.
  • The framework may generalize to generate evaluation environments in other sequential decision domains.
  • Repeated runs with varied generation parameters could quantify how game complexity affects observed agent limits.

Load-bearing premise

The generated text games and their associated metrics accurately capture the core abilities required for real-world test-time continual learning.

What would settle it

An experiment showing that agents achieving high scores on AgentOdyssey games fail to exhibit corresponding gains in exploration, memory retention, or planning when deployed in non-game continual learning environments.

read the original abstract

For agents to learn continuously from interaction with the world at test time, they must be able to explore effectively, acquire new world knowledge and skills, retain relevant episodic experiences, and plan over long horizons. To evaluate these key abilities of test-time continual learning agents, we introduce AgentOdyssey, a novel evaluation framework that procedurally generates open-ended text games with rich entities, world dynamics, and long-horizon tasks. Critically, AgentOdyssey goes beyond the conventional machine learning assumption that learning does not occur at test time by placing agents in a continuous, long-horizon setting that interleaves learning and inference throughout deployment. We further propose a multifaceted evaluation methodology that measures not only game progress but also offers diagnostic tests on world knowledge acquisition, episodic memory, object and action exploration, action diversity, and model cost. We evaluate diverse agent paradigms in the generated games. Our experimental results reveal critical limits in agents' key abilities, as well as factors that influence their meaningful horizon. Although performance scales with stronger base models, even the top agent remains far below human performance, leaving substantial headroom for improvement. Among agent mechanisms, we find that short-term memory benefits multiple agent paradigms and is an important component of agent test-time training.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces AgentOdyssey, a framework that procedurally generates open-ended long-horizon text games with rich entities and dynamics to benchmark test-time continual learning agents. It positions the setting as one that interleaves learning and inference throughout deployment (unlike standard episodic RL), proposes diagnostic metrics for exploration, episodic memory, world knowledge, action diversity, and cost, and reports that stronger base models improve performance yet all agents remain far below human levels, with short-term memory providing benefits across paradigms.

Significance. If the generated environments demonstrably require online adaptation rather than pre-trained inference, the framework supplies a reusable benchmark for test-time continual learning that directly targets the four core abilities listed in the abstract. The multifaceted diagnostics and the finding that short-term memory aids multiple agent classes are concrete contributions that could guide future memory-augmented architectures.

major comments (2)
  1. [Procedural Generation] The procedural generation section does not supply the concrete entity schemas, dynamics templates, horizon-length distribution, or novelty-injection rules. Without these, it is impossible to verify that the generated tasks cannot be solved by the base model’s prior knowledge or short-horizon search alone, which is the load-bearing assumption for the claim that the setting “interleaves learning and inference throughout deployment.”
  2. [Evaluation Methodology] The diagnostic metrics for episodic memory and world-knowledge acquisition are introduced without ablation against an inference-only baseline or against human performance on the same games. Consequently, it remains unclear whether measured gains truly reflect test-time updates rather than improved prompting or retrieval.
minor comments (2)
  1. [Results] Figure captions and axis labels in the results section use inconsistent terminology (“game progress” vs. “task completion rate”) that should be unified with the metric definitions given earlier.
  2. [Experiments] The abstract states that “performance scales with stronger base models” but the main text does not report the exact model sizes or parameter counts used in the scaling experiment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our contributions. We address each major point below and indicate the revisions we will incorporate.

read point-by-point responses

  1. Referee: [Procedural Generation] The procedural generation section does not supply the concrete entity schemas, dynamics templates, horizon-length distribution, or novelty-injection rules. Without these, it is impossible to verify that the generated tasks cannot be solved by the base model’s prior knowledge or short-horizon search alone, which is the load-bearing assumption for the claim that the setting “interleaves learning and inference throughout deployment.”

    Authors: We agree that the current description of procedural generation is high-level and lacks the requested implementation specifics. In the revised manuscript we will expand this section to include the entity schemas (with examples of attributes and relations), dynamics templates (including state-transition rules and interaction effects), the horizon-length distribution used during generation, and the novelty-injection mechanism (including how new entities and rules are sampled and integrated). These additions will make it possible to inspect whether tasks require ongoing adaptation beyond base-model priors or short-horizon search. revision: yes

  2. Referee: [Evaluation Methodology] The diagnostic metrics for episodic memory and world-knowledge acquisition are introduced without ablation against an inference-only baseline or against human performance on the same games. Consequently, it remains unclear whether measured gains truly reflect test-time updates rather than improved prompting or retrieval.

    Authors: The manuscript already reports that even the strongest agents remain substantially below human performance on the generated games; however, we acknowledge that the current evaluation does not contain an explicit ablation that isolates test-time updates from inference-only behavior. We will add this ablation (comparing agents with and without test-time memory or knowledge updates) and will also provide further detail on the human baseline collection protocol to confirm that the same game instances were used. These changes will strengthen the evidence that observed gains arise from test-time adaptation rather than prompting or retrieval alone. revision: partial

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper introduces a new procedural generation framework (AgentOdyssey) and associated diagnostic metrics for evaluating test-time continual learning. No equations, parameter fits, or self-citations appear in the provided abstract or description that would reduce any claimed result to its own inputs by construction. The central premise is the design choice of interleaving learning and inference via generated long-horizon games; this is an explicit methodological decision rather than a derived prediction. The work is therefore self-contained against external benchmarks with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; ledger is empty.

pith-pipeline@v0.9.1-grok · 5773 in / 985 out tokens · 20006 ms · 2026-06-28T21:57:42.590529+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 33 canonical work pages · 20 internal anchors

  1. [1]

    Do As I Can, Not As I Say: Grounding Language in Robotic Affordances

    Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Hausman, et al. Do as i can, not as i say: Grounding language in robotic affordances.arXiv preprint arXiv:2204.01691,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Erik Andersen, Eleanor O’rourke, Yun-En Liu, Rich Snider, Jeff Lowdermilk, David Truong, Seth Cooper, and Zoran Popovic

    Accessed: 2026-02-14. Erik Andersen, Eleanor O’rourke, Yun-En Liu, Rich Snider, Jeff Lowdermilk, David Truong, Seth Cooper, and Zoran Popovic. The impact of tutorials on games of varying complexity. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems, pages 59–68,

    2026

  3. [3]

    Qwen Technical Report

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. Qwen technical report.arXiv preprint arXiv:2309.16609,

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [4]

    Titans: Learning to Memorize at Test Time

    Ali Behrouz, Peilin Zhong, and Vahab Mirrokni. Titans: Learning to memorize at test time.arXiv preprint arXiv:2501.00663,

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [5]

    Nested learning: The illusion of deep learning architectures.arXiv preprint arXiv:2512.24695,

    Ali Behrouz, Meisam Razaviyayn, Peilin Zhong, and Vahab Mirrokni. Nested learning: The illusion of deep learning architectures.arXiv preprint arXiv:2512.24695,

    work page arXiv

  6. [6]

    arXiv preprint arXiv:2310.05915 , year=

    Baian Chen, Chang Shu, Ehsan Shareghi, Nigel Collier, Karthik Narasimhan, and Shunyu Yao. Fireact: Toward language agent fine-tuning.arXiv preprint arXiv:2310.05915,

    work page arXiv

  7. [7]

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhail Pavlov, Alethea Power, Lukasz Kaiser, Mohammad Bavarian...

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [8]

    Can vlms play action role-playing games? take black myth wukong as a study case.arXiv preprint arXiv:2409.12889,

    Peng Chen, Pi Bu, Jun Song, Yuan Gao, and Bo Zheng. Can vlms play action role-playing games? take black myth wukong as a study case.arXiv preprint arXiv:2409.12889,

    work page arXiv

  9. [9]

    Embodiedeval: Evaluate multimodal llms as embodied agents.arXiv preprint arXiv:2501.11858,

    Zhili Cheng, Yuge Tu, Ran Li, Shiqi Dai, Jinyi Hu, Shengding Hu, Jiahao Li, Yang Shi, Tianyu Yu, Weize Chen, et al. Embodiedeval: Evaluate multimodal llms as embodied agents.arXiv preprint arXiv:2501.11858,

    work page arXiv

  10. [10]

    Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory

    Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. Mem0: Building production-ready ai agents with scalable long-term memory.arXiv preprint arXiv:2504.19413,

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    Group-in-Group Policy Optimization for LLM Agent Training

    Lang Feng, Zhenghai Xue, Tingcong Liu, and Bo An. Group-in-group policy optimization for llm agent training.arXiv preprint arXiv:2505.10978,

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    The Llama 3 Herd of Models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models.arXiv preprint arXiv:2407.21783,

    work page internal anchor Pith review Pith/arXiv arXiv

  13. [13]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shi- rong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948,

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    Measuring Mathematical Problem Solving With the MATH Dataset

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset.arXiv preprint arXiv:2103.03874,

    work page internal anchor Pith review Pith/arXiv arXiv

  15. [15]

    lmgame-bench: How good are llms at playing games?arXiv preprint arXiv:2505.15146,

    Lanxiang Hu, Mingjia Huo, Yuxuan Zhang, Haoyang Yu, Eric P Xing, Ion Stoica, Tajana Rosing, Haojian Jin, and Hao Zhang. lmgame-bench: How good are llms at playing games?arXiv preprint arXiv:2505.15146,

    work page arXiv

  16. [16]

    Language understanding for text-based games using deep reinforcement learning

    Karthik Narasimhan, Tejas Kulkarni, and Regina Barzilay. Language understanding for text-based games using deep reinforcement learning. InProceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1–11,

    2015

  17. [17]

    A Generalist Agent

    Scott Reed, Konrad Zolna, Emilio Parisotto, Sergio Gomez Colmenarejo, Alexander Novikov, Gabriel Barth-Maron, Mai Gimenez, Yury Sulsky, Jackie Kay, Jost Tobias Springenberg, et al. A generalist agent.arXiv preprint arXiv:2205.06175,

    work page internal anchor Pith review Pith/arXiv arXiv

  18. [18]

    Simworld: An open-ended realistic simulator for autonomous agents in physical and social worlds.arXiv preprint arXiv:2512.01078,

    Jiawei Ren, Yan Zhuang, Xiaokang Ye, Lingjun Mao, Xuhong He, Jianzhi Shen, Mrinaal Dogra, Yiming Liang, Ruixuan Zhang, Tianai Yue, et al. Simworld: An open-ended realistic simulator for autonomous agents in physical and social worlds.arXiv preprint arXiv:2512.01078,

    work page arXiv

  19. [19]

    Code Llama: Open Foundation Models for Code

    19 Baptiste Roziere, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Romain Sauvestre, Tal Remez, et al. Code llama: Open foundation models for code. arXiv preprint arXiv:2308.12950,

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    Statistical learning by 8-month-old infants

    Jenny R Saffran, Richard N Aslin, and Elissa L Newport. Statistical learning by 8-month-old infants. science, 274(5294):1926–1928,

    1926

  21. [21]

    Proximal Policy Optimization Algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347,

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    ALFWorld: Aligning Text and Embodied Environments for Interactive Learning

    Mohit Shridhar, Xingdi Yuan, Marc-Alexandre C ˆot´e, Yonatan Bisk, Adam Trischler, and Matthew Hausknecht. Alfworld: Aligning text and embodied environments for interactive learning.arXiv preprint arXiv:2010.03768,

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  23. [23]

    OpenAI GPT-5 System Card

    Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, et al. Openai gpt-5 system card.arXiv preprint arXiv:2601.03267,

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    In: IEEE International Conference on Robotics and Automation, ICRA 2023, London, UK, May 29 - June 2, 2023

    doi: 10.1109/ICRA48891.2023.10161317. Elizabeth S Spelke and Katherine D Kinzler. Core knowledge.Developmental science, 10(1):89–96,

    work page doi:10.1109/icra48891.2023.10161317 2023

  25. [25]

    Learning to (Learn at Test Time): RNNs with Expressive Hidden States

    Yu Sun, Xinhao Li, Karan Dalal, Jiarui Xu, Arjun Vikram, Genghan Zhang, Yann Dubois, Xinlei Chen, Xiaolong Wang, Sanmi Koyejo, et al. Learning to (learn at test time): Rnns with expressive hidden states.arXiv preprint arXiv:2407.04620,

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    Cradle: Empowering foundation agents towards general computer control.arXiv preprint arXiv:2403.03186,

    Weihao Tan, Wentao Zhang, Xinrun Xu, Haochong Xia, Ziluo Ding, Boyu Li, Bohan Zhou, Junpeng Yue, Jiechuan Jiang, Yewen Li, et al. Cradle: Empowering foundation agents towards general computer control.arXiv preprint arXiv:2403.03186,

    work page arXiv

  27. [27]

    End-to-end test-time training for long context.arXiv preprint arXiv:2512.23675,

    Arnuv Tandon, Karan Dalal, Xinhao Li, Daniel Koceja, Marcel Rød, Sam Buchanan, Xiaolong Wang, Jure Leskovec, Sanmi Koyejo, Tatsunori Hashimoto, et al. End-to-end test-time training for long context.arXiv preprint arXiv:2512.23675,

    work page arXiv

  28. [28]

    Voyager: An Open-Ended Embodied Agent with Large Language Models

    20 Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Voyager: An open-ended embodied agent with large language models.arXiv preprint arXiv:2305.16291, 2023a. Ruoyao Wang, Graham Todd, Xingdi Yuan, Ziang Xiao, Marc-Alexandre C ˆot´e, and Peter Jansen. Bytesized32: A corpus and challenge task for ...

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  29. [29]

    M+: Extending memoryllm with scalable long-term memory

    Yu Wang, Dmitry Krotov, Yuanzhe Hu, Yifan Gao, Wangchunshu Zhou, Julian McAuley, Dan Gutfre- und, Rogerio Feris, and Zexue He. M+: Extending memoryllm with scalable long-term memory. arXiv preprint arXiv:2502.00592, 2025b. Zihan Wang, Kangrui Wang, Qineng Wang, Pingyue Zhang, Linjie Li, Zhengyuan Yang, Xing Jin, Kefan Yu, Minh Nhat Nguyen, Licheng Liu, et...

    work page arXiv

  30. [30]

    Qwen3 Technical Report

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388,

    work page internal anchor Pith review Pith/arXiv arXiv

  31. [31]

    MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent

    Shunyu Yao, Howard Chen, John Yang, and Karthik Narasimhan. Webshop: Towards scalable real- world web interaction with grounded language agents.Advances in Neural Information Processing Systems, 35:20744–20757, 2022a. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in la...

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    Memgen: Weaving generative latent memory for self-evolving agents.arXiv preprint arXiv:2509.24704, 2025a

    Guibin Zhang, Muxin Fu, and Shuicheng Yan. Memgen: Weaving generative latent memory for self-evolving agents.arXiv preprint arXiv:2509.24704, 2025a. Hongxin Zhang, Weihua Du, Jiaming Shan, Qinhong Zhou, Yilun Du, Joshua B Tenenbaum, Tianmin Shu, and Chuang Gan. Building cooperative embodied agents modularly with large language models. InThe Twelfth Intern...

    work page arXiv

  33. [33]

    Test-Time Training Done Right

    Tianyuan Zhang, Sai Bi, Yicong Hong, Kai Zhang, Fujun Luan, Songlin Yang, Kalyan Sunkavalli, William T Freeman, and Hao Tan. Test-time training done right.arXiv preprint arXiv:2505.23884, 2025b. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, et al. Qwen3 embedding: Advancing tex...

    work page internal anchor Pith review Pith/arXiv arXiv

  34. [34]

    Story2game: Generat- ing (almost) everything in an interactive fiction game.arXiv preprint arXiv:2505.03547, 2025a

    Eric Zhou, Shreyas Basavatia, Moontashir Siam, Zexin Chen, and Mark O Riedl. Story2game: Generat- ing (almost) everything in an interactive fiction game.arXiv preprint arXiv:2505.03547, 2025a. Qinhong Zhou, Hongxin Zhang, Xiangye Lin, Zheyuan Zhang, Yutian Chen, Wenjun Liu, Zunzhe Zhang, Sunli Chen, Lixing Fang, Qiushi Lyu, et al. Virtual community: An op...

    work page arXiv

  35. [35]

    MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents

    Zijian Zhou, Ao Qu, Zhaoxuan Wu, Sunghwan Kim, Alok Prakash, Daniela Rus, Jinhua Zhao, Bryan Kian Hsiang Low, and Paul Pu Liang. Mem1: Learning to synergize memory and reasoning for efficient long-horizon agents.arXiv preprint arXiv:2506.15841, 2025c. Yan Zhuang, Jiawei Ren, Xiaokang Ye, Jianzhi Shen, Ruixuan Zhang, Tianai Yue, Muhammad Faayez, Xuhong He,...

    work page internal anchor Pith review Pith/arXiv arXiv

  36. [36]

    r e a s o n i n g

    22 Appendix A1. More Results for Experiment 1 and Experiment 2 TableA1:Experiment 1 results with proprietary LLM backbones. ↑ and ↓ indicate that higher and lower values are better, respectively. Best results in each column are highlighted in bold.Qdenotes the main quest progress reward.SQ,E,C, andDdenote supplementary rewards for side quests, area explor...

    work page arXiv 2051

  37. [37]

    env " , coin_id , bonus_coins , res . tloc (

    ) + b o n u s _ c o i n s res . t r a c k _ s p a w n (" env " , coin_id , bonus_coins , res . tloc (" area " , area_id ) ) res . a d d _ f e e d b a c k (...) res . events . append ( Event ( type =" c r a f t _ s t r e a k _ b o n u s " ... ) ) env . c u r r _ a g e n t s _ s t a t e [" c r a f t _ s t r e a k "][ agent . id ] = { " l a s t _ s t e p ": ...

    2025