pith. machine review for the scientific record. sign in

arxiv: 2605.13391 · v1 · submitted 2026-05-13 · 💻 cs.AI

Recognition: no theorem link

RS-Claw: Progressive Active Tool Exploration via Hierarchical Skill Trees for Remote Sensing Agents

Chengfu Liu, Cheng Yang, Dongyang Hou, Haifeng Li, Hanwen Yu, Kai Ouyang, Liangtian Liu, Wentao Yang, Zeyuan Wang, Zichao Tang, Ziyu Li

Authors on Pith no claims yet

Pith reviewed 2026-05-14 19:29 UTC · model grok-4.3

classification 💻 cs.AI
keywords remote sensing agentstool explorationhierarchical skill treesactive explorationtoken compressionskill encapsulationEarth-Bench benchmarkmulti-modal LLMs
0
0 comments X

The pith

RS-Claw lets remote sensing agents actively explore tools via hierarchical skill trees, achieving up to 86% token compression while outperforming flat and RAG baselines.

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

The paper establishes that passive tool selection in remote sensing agents, whether through full registration or retrieval, fails to balance context load and completeness in large heterogeneous tool sets. RS-Claw introduces active exploration by hierarchically structuring tools into skill trees using encapsulation. This allows agents to select relevant branches from summaries first, then load details on demand for precise invocation. Sympathetic readers would care because this mechanism filters semantic noise, frees reasoning space, and improves performance on complex tasks in the Earth-Bench benchmark.

Core claim

RS-Claw redefines tool selection as active exploration in the tool space. By leveraging skill encapsulation to hierarchically structure tool descriptions, the agent executes on-demand sequential decision-making: first selecting relevant skill branches by reading only summaries, then dynamically loading detailed descriptions, and finally achieving precise invocation. This active paradigm liberates context space and ensures accurate hit rates of critical tools during long-horizon reasoning.

What carries the argument

Hierarchical skill trees constructed through skill encapsulation at the tool end, enabling progressive on-demand tool loading from summaries to details.

If this is right

  • RS-Claw achieves input token compression ratios of up to 86% by filtering irrelevant tool information.
  • It comprehensively outperforms existing Flat and RAG baselines on complex reasoning evaluations in the Earth-Bench benchmark.
  • The active exploration mechanism effectively filters semantic noise and frees up the agent's reasoning space.
  • Agents can maintain high tool hit rates without omissions in long-horizon tasks within massive RS tool ecosystems.

Where Pith is reading between the lines

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

  • This approach may scale to other domains with large numbers of heterogeneous tools, such as general AI agents or robotics.
  • Future work could test whether the hierarchical structure reduces errors in real-time remote sensing applications like disaster response.
  • Integrating this with other agent frameworks might allow dynamic tree updates based on new tools.

Load-bearing premise

That the hierarchical skill trees can be structured such that summary-level selections reliably guide the agent to the exact critical tools needed without omissions in sequential decisions.

What would settle it

Observing whether RS-Claw misses critical tools more often than RAG methods in long-horizon tasks on Earth-Bench or fails to achieve the reported token compression would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.13391 by Chengfu Liu, Cheng Yang, Dongyang Hou, Haifeng Li, Hanwen Yu, Kai Ouyang, Liangtian Liu, Wentao Yang, Zeyuan Wang, Zichao Tang, Ziyu Li.

Figure 1
Figure 1. Figure 1: Comparison of agent tool selection paradigms. (a) Passive paradigm: Existing methods define the agent as a passive tool recipient. Specifically, the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The overall framework of the progressive active tool exploration mechanism based on a hierarchical skill tree. The top panel illustrates the unified [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Accuracy and context overhead curves under same-domain tool [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Accuracy and context overhead comparison under cross-domain tool [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
read the original abstract

The rise of multi-modal large language models (MLLMs) is shifting remote sensing (RS) intelligence from "see" to "action", as OpenClaw-style frameworks enable agents to autonomously operate massive RS image-processing tools for complex tasks. Existing RS agents adopt a passive selection paradigm for tool invocation, relying on either full tool registration (Flat) or retrieval-augmented generation (RAG). However, in the massive and multi-source heterogeneous RS tool ecosystem, such passive mechanisms struggle to dynamically balance "context load" and "toolset completeness" throughout task reasoning, thus exhibiting inherent limitations: full tool registration triggers context space deficits during long-horizon tasks, whereas RAG retrieval may omit critical tools in essential steps. To overcome these bottlenecks, this paper redefines tool selection by arguing that the agent should act as an active explorer within the tool space. Based on this perspective, we propose RS-Claw, a novel RS agent architecture. By leveraging Skill encapsulation technology at the tool end, this architecture hierarchically structures tool descriptions, enabling the agent to execute on-demand sequential decision-making: initially selecting relevant skill branches by reading only tool summaries, then dynamically loading detailed descriptions, and ultimately achieving precise invocation. This active paradigm not only significantly liberates the agent's context space but also effectively ensures the accurate hit rate of critical tools during long-horizon reasoning. Systematic experiments on the Earth-Bench benchmark demonstrate that RS-Claw's active exploration mechanism effectively filters semantic noise and substantially frees up reasoning space, achieving an input token compression ratio of up to 86%, and comprehensively outperforming existing Flat and RAG baselines across complex reasoning evaluations.

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

3 major / 1 minor

Summary. The paper claims that RS-Claw, by using hierarchical skill trees and skill encapsulation, enables remote sensing agents to perform active tool exploration: selecting branches from summaries then loading details on demand. This results in up to 86% input token compression and outperforms Flat and RAG baselines on complex reasoning tasks in the Earth-Bench benchmark.

Significance. Should the results be substantiated with detailed experiments, this work could have high significance for developing efficient agents in tool-heavy domains like remote sensing, where context management is critical for long-horizon tasks. It introduces a promising active paradigm that may reduce semantic noise and improve reasoning space.

major comments (3)
  1. [Abstract] The abstract asserts substantial outperformance and an 86% compression ratio on Earth-Bench but provides no quantitative metrics, error bars, ablation details, or experimental protocol, rendering the central performance claim unassessable.
  2. [Architecture Description] The hierarchical skill tree mechanism is presented as ensuring accurate tool hit rates without omissions, yet no details are given on the initial branch selection policy, decision criteria, or error-recovery strategies, leaving the single-point-of-failure risk unaddressed for long-horizon tasks in heterogeneous tool ecosystems.
  3. [Experiments] There is no reported measurement of omission rates or tool invocation accuracy on Earth-Bench long-horizon cases to support the claim that the active exploration avoids the omissions seen in RAG baselines.
minor comments (1)
  1. The term 'Skill encapsulation technology' is introduced without a clear definition or reference to prior work.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We address each major comment below and commit to revisions that strengthen the manuscript's clarity and substantiation without altering its core contributions.

read point-by-point responses

  1. Referee: [Abstract] The abstract asserts substantial outperformance and an 86% compression ratio on Earth-Bench but provides no quantitative metrics, error bars, ablation details, or experimental protocol, rendering the central performance claim unassessable.

    Authors: We agree that the abstract would benefit from additional quantitative anchors to improve assessability. In the revised manuscript we will expand the abstract to report the mean token compression ratio with standard deviation across Earth-Bench tasks, the average performance margin versus the strongest baseline, and a concise statement of the evaluation protocol (number of tasks, horizon lengths, and model backbone). Space constraints will limit the level of detail, but the added figures will directly support the headline claims. revision: yes

  2. Referee: [Architecture Description] The hierarchical skill tree mechanism is presented as ensuring accurate tool hit rates without omissions, yet no details are given on the initial branch selection policy, decision criteria, or error-recovery strategies, leaving the single-point-of-failure risk unaddressed for long-horizon tasks in heterogeneous tool ecosystems.

    Authors: We acknowledge that the current description is high-level and omits operational specifics. The revised architecture section will explicitly define: (i) the branch-selection policy (LLM-driven relevance scoring of summaries against the current reasoning state with a tunable threshold), (ii) the decision criteria (contextual utility, recency, and estimated token cost), and (iii) error-recovery mechanisms (progressive fallback to sibling branches, re-query with expanded context, or full-detail load on detected uncertainty). These additions will directly address the single-point-of-failure concern for long-horizon heterogeneous settings. revision: yes

  3. Referee: [Experiments] There is no reported measurement of omission rates or tool invocation accuracy on Earth-Bench long-horizon cases to support the claim that the active exploration avoids the omissions seen in RAG baselines.

    Authors: We accept that omission-rate and invocation-accuracy metrics were not reported. We will add a dedicated analysis subsection (and accompanying table) that measures tool-omission frequency and invocation precision on the long-horizon subset of Earth-Bench, with direct head-to-head comparison against the RAG baseline. These new results will be generated from the same experimental runs already described and will be presented with error bars. revision: yes

Circularity Check

0 steps flagged

No circularity: architecture and gains are externally benchmarked

full rationale

The paper presents RS-Claw as a new hierarchical skill-tree architecture for active tool exploration. Claims of up to 86% token compression and outperformance on Earth-Bench are supported solely by direct empirical comparisons against Flat and RAG baselines, with no equations, fitted parameters, or self-citations that reduce the mechanism to its own inputs by construction. The central design choice (on-demand branch loading from summaries) is described as an independent proposal rather than derived from prior self-referential results or uniqueness theorems. No load-bearing step matches any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the domain assumption that the RS tool ecosystem is too large for full registration and that RAG risks omissions, plus the new concepts of skill encapsulation and hierarchical trees introduced without external validation in the abstract.

axioms (1)
  • domain assumption Existing passive tool selection mechanisms (full registration or RAG) inherently struggle to balance context load and toolset completeness in massive heterogeneous RS tool ecosystems.
    Directly stated as the motivation and limitation of prior approaches.
invented entities (2)
  • Skill encapsulation technology no independent evidence
    purpose: To hierarchically structure tool descriptions for on-demand loading
    Introduced as the enabling technology at the tool end; no independent evidence provided.
  • Hierarchical Skill Trees no independent evidence
    purpose: To enable progressive active sequential decision-making for tool invocation
    Core architectural component of RS-Claw; no prior or external validation cited.

pith-pipeline@v0.9.0 · 5632 in / 1353 out tokens · 38470 ms · 2026-05-14T19:29:20.903863+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

45 extracted references · 15 canonical work pages · 4 internal anchors

  1. [1]

    Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models,

    L. Wang, W. Xu, Y . Lan, Z. Hu, Y . Lan, R. K.-W. Lee, and E.-P. Lim, “Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models,”arXiv preprint arXiv:2305.04091, 2023. 20

    work page arXiv 2023

  2. [2]

    Toolformer: Language models can teach themselves to use tools,

    T. Schick, J. Dwivedi-Yu, R. Dess `ı, R. Raileanu, M. Lomeli, E. Hambro, L. Zettlemoyer, N. Cancedda, and T. Scialom, “Toolformer: Language models can teach themselves to use tools,”Advances in Neural Infor- mation Processing Systems, vol. 36, pp. 68 539–68 551, 2023

    2023

  3. [3]

    HuggingGPT: Solving AI tasks with ChatGPT and its friends in Hugging Face,

    Y . Shen, K. Song, X. Tan, D. Li, W. Lu, and Y . Zhuang, “HuggingGPT: Solving AI tasks with ChatGPT and its friends in Hugging Face,”Ad- vances in Neural Information Processing Systems, vol. 36, pp. 38 154– 38 180, 2023

    2023

  4. [4]

    A survey on large language model based autonomous agents,

    L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, Z. Chen, J. Tang, X. Chen, Y . Linet al., “A survey on large language model based autonomous agents,”Frontiers of Computer Science, vol. 18, no. 6, p. 186345, 2024

    2024

  5. [5]

    OpenClaw: Open-source personal AI assistant,

    OpenClaw, “OpenClaw: Open-source personal AI assistant,” https: //github.com/openclaw/openclaw, 2026, version 2026.3.8, Accessed: 2026-03-09

    2026

  6. [6]

    RS-Agent: Automating remote sensing tasks through intelligent agent,

    W. Xu, Z. Yu, B. Mu, Z. Wei, Y . Zhang, G. Li, J. Wang, and M. Peng, “RS-Agent: Automating remote sensing tasks through intelligent agent,” arXiv preprint arXiv:2406.07089, 2024

    work page arXiv 2024

  7. [7]

    Earth-agent: Unlocking the full landscape of earth observation with agents,

    P. Feng, Z. Lv, J. Ye, X. Wang, X. Huo, J. Yu, W. Xu, W. Zhang, L. Bai, C. Heet al., “Earth-Agent: Unlocking the full landscape of earth observation with agents,”arXiv preprint arXiv:2509.23141, 2025

    work page arXiv 2025

  8. [8]

    Big data for remote sensing: Challenges and opportunities,

    M. Chi, A. Plaza, J. A. Benediktsson, Z. Sun, J. Shen, and Y . Zhu, “Big data for remote sensing: Challenges and opportunities,”Proceedings of the IEEE, vol. 104, no. 11, pp. 2207–2219, 2016

    2016

  9. [9]

    Google earth engine: Planetary-scale geospatial analysis for everyone,

    N. Gorelick, M. Hancher, M. Dixon, S. Ilyushchenko, D. Thau, and R. Moore, “Google earth engine: Planetary-scale geospatial analysis for everyone,”Remote sensing of Environment, vol. 202, pp. 18–27, 2017

    2017

  10. [10]

    Orfeo toolbox: open source processing of remote sensing images,

    M. Grizonnet, J. Michel, V . Poughon, J. Inglada, M. Savinaud, and R. Cresson, “Orfeo toolbox: open source processing of remote sensing images,”Open Geospatial Data, Software and Standards, vol. 2, no. 1, p. 15, 2017

    2017

  11. [11]

    ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs

    Y . Qin, S. Liang, Y . Ye, K. Zhu, L. Yan, Y . Lu, Y . Lin, X. Cong, X. Tang, B. Qianet al., “ToolLLM: Facilitating large language models to master 16000+ real-world APIs,”arXiv preprint arXiv:2307.16789, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  12. [12]

    Benchmarking single agent performance,

    LangChain Team, “Benchmarking single agent performance,” LangChain Blog. [Online]. Available: https://blog.langchain.com/ react-agent-benchmarking/, Feb. 2025, [Accessed: Apr. 24, 2026]

    2025

  13. [13]

    Longbench: A bilingual, multitask benchmark for long context understanding,

    Y . Bai, X. Lv, J. Zhang, H. Lyu, J. Tang, Z. Huang, Z. Du, X. Liu, A. Zeng, L. Houet al., “Longbench: A bilingual, multitask benchmark for long context understanding,” inProceedings of the 62nd annual meeting of the association for computational linguistics (volume 1: Long papers), 2024, pp. 3119–3137

    2024

  14. [14]

    ReAct: Synergizing reasoning and acting in language models,

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. R. Narasimhan, and Y . Cao, “ReAct: Synergizing reasoning and acting in language models,” inProc. 11th Int. Conf. Learn. Represent. (ICLR), 2023

    2023

  15. [15]

    Lost in the middle: How language models use long contexts,

    N. F. Liu, K. Lin, J. Hewitt, A. Paranjape, M. Bevilacqua, F. Petroni, and P. Liang, “Lost in the middle: How language models use long contexts,” Transactions of the Association for Computational Linguistics, vol. 12, pp. 157–173, 2024

    2024

  16. [16]

    Gorilla: Large language model connected with massive APIs,

    S. G. Patil, T. Zhang, X. Wang, and J. E. Gonzalez, “Gorilla: Large language model connected with massive APIs,”Advances in Neural Information Processing Systems, vol. 37, pp. 126 544–126 565, 2024

    2024

  17. [17]

    Retrieval- augmented generation for knowledge-intensive NLP tasks,

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. K ¨uttler, M. Lewis, W.-T. Yih, T. Rockt ¨aschelet al., “Retrieval- augmented generation for knowledge-intensive NLP tasks,”Advances in Neural Information Processing Systems, vol. 33, pp. 9459–9474, 2020

    2020

  18. [18]

    ToolReAGt: Tool retrieval for LLM-based complex task solution via retrieval augmented generation,

    N. Braunschweiler, R. Doddipatla, and T.-C. Zorila, “ToolReAGt: Tool retrieval for LLM-based complex task solution via retrieval augmented generation,” inProc. 3rd Workshop Towards Knowledgeable Foundation Models (KnowFM), 2025, pp. 75–83

    2025

  19. [19]

    Agent skills overview – Claude platform documentation,

    Anthropic, “Agent skills overview – Claude platform documentation,” [Online]. Available: https://platform.claude.com/docs/en/ agents-and-tools/agent-skills/overview, 2026

    2026

  20. [20]

    Reflex- ion: Language agents with verbal reinforcement learning,

    N. Shinn, F. Cassano, A. Gopinath, K. Narasimhan, and S. Yao, “Reflex- ion: Language agents with verbal reinforcement learning,”Advances in Neural Information Processing Systems, vol. 36, pp. 8634–8652, 2023

    2023

  21. [21]

    Tree of thoughts: Deliberate problem solving with large language models,

    S. Yao, D. Yu, J. Zhao, I. Shafran, T. Griffiths, Y . Cao, and K. Narasimhan, “Tree of thoughts: Deliberate problem solving with large language models,”Advances in neural information processing systems, vol. 36, pp. 11 809–11 822, 2023

    2023

  22. [22]

    RSGPT: A remote sensing vision language model and benchmark,

    Y . Hu, J. Yuan, C. Wen, X. Lu, Y . Liu, and X. Li, “RSGPT: A remote sensing vision language model and benchmark,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 224, pp. 272–286, 2025

    2025

  23. [23]

    GeoChat: Grounded large vision-language model for remote sensing,

    K. Kuckreja, M. S. Danish, M. Naseer, A. Das, S. Khan, and F. S. Khan, “GeoChat: Grounded large vision-language model for remote sensing,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2024, pp. 27 831–27 840

    2024

  24. [24]

    EarthGPT: A universal multimodal large language model for multisensor image comprehension in remote sensing domain,

    W. Zhang, M. Cai, T. Zhang, Y . Zhuang, and X. Mao, “EarthGPT: A universal multimodal large language model for multisensor image comprehension in remote sensing domain,”IEEE Transactions on Geo- science and Remote Sensing, vol. 62, pp. 1–20, 2024

    2024

  25. [25]

    EarthDial: Turning multi-sensory earth observations to interactive dialogues,

    S. Soni, A. Dudhane, H. Debary, M. Fiaz, M. A. Munir, M. S. Danish, P. Fraccaro, C. D. Watson, L. Klein, F. S. Khanet al., “EarthDial: Turning multi-sensory earth observations to interactive dialogues,” in Proc. Comput. Vis. Pattern Recognit. Conf. (CVPR), 2025, pp. 14 303– 14 313

    2025

  26. [26]

    Remoteclip: A vision language foundation model for remote sensing,

    F. Liu, D. Chen, Z. Guan, X. Zhou, J. Zhu, Q. Ye, L. Fu, and J. Zhou, “Remoteclip: A vision language foundation model for remote sensing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1– 16, 2024

    2024

  27. [27]

    Skysense: A multi-modal remote sensing foundation model towards universal interpretation for earth observation imagery,

    X. Guo, J. Lao, B. Dang, Y . Zhang, L. Yu, L. Ru, L. Zhong, Z. Huang, K. Wu, D. Huet al., “Skysense: A multi-modal remote sensing foundation model towards universal interpretation for earth observation imagery,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 27 672–27 683

    2024

  28. [28]

    Change-agent: Toward interactive comprehensive remote sensing change interpretation and analysis,

    C. Liu, K. Chen, H. Zhang, Z. Qi, Z. Zou, and Z. Shi, “Change-agent: Toward interactive comprehensive remote sensing change interpretation and analysis,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–16, 2024

    2024

  29. [29]

    Multi-agent geospatial copilots for remote sensing workflows,

    C. Lee, V . Paramanayakam, A. Karatzas, Y . Jian, M. Fore, H. Liao, F. Yu, R. Li, I. Anagnostopoulos, and D. Stamoulis, “Multi-agent geospatial copilots for remote sensing workflows,” inProc. IGARSS 2025, 2025, pp. 1084–1089

    2025

  30. [30]

    Evaluating tool-augmented agents in remote sensing platforms,

    S. Singh, M. Fore, and D. Stamoulis, “Evaluating tool-augmented agents in remote sensing platforms,”arXiv preprint arXiv:2405.00709, 2024

    work page arXiv 2024

  31. [31]

    ThinkGeo: Evaluating tool-augmented agents for remote sensing tasks,

    A. Shabbir, M. A. Munir, A. Dudhane, M. U. Sheikh, M. H. Khan, P. Fraccaro, J. B. Moreno, F. S. Khan, and S. Khan, “ThinkGeo: Evaluating tool-augmented agents for remote sensing tasks,”arXiv preprint arXiv:2505.23752, 2025

    work page arXiv 2025

  32. [32]

    Tool learning with large language models: A survey,

    C. Qu, S. Dai, X. Wei, H. Cai, S. Wang, D. Yin, J. Xu, and J.-R. Wen, “Tool learning with large language models: A survey,”Frontiers of Computer Science, vol. 19, no. 8, p. 198343, 2025

    2025

  33. [33]

    AgentBench: Evaluating LLMs as Agents

    X. Liu, H. Yu, H. Zhang, Y . Xu, X. Lei, H. Lai, Y . Gu, H. Ding, K. Men, K. Yanget al., “Agentbench: Evaluating llms as agents,”arXiv preprint arXiv:2308.03688, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  34. [34]

    Api-bank: A comprehensive benchmark for tool-augmented llms,

    M. Li, Y . Zhao, B. Yu, F. Song, H. Li, H. Yu, Z. Li, F. Huang, and Y . Li, “Api-bank: A comprehensive benchmark for tool-augmented llms,” in Proceedings of the 2023 conference on empirical methods in natural language processing, 2023, pp. 3102–3116

    2023

  35. [35]

    The tool decathlon: Benchmarking language agents for diverse, realistic, and long-horizon task execution,

    J. Li, W. Zhao, J. Zhao, W. Zeng, H. Wu, X. Wang, R. Ge, Y . Cao, Y . Huang, W. Liuet al., “The tool decathlon: Benchmarking language agents for diverse, realistic, and long-horizon task execution,”arXiv preprint arXiv:2510.25726, 2025

    work page arXiv 2025

  36. [36]

    Taskmatrix. ai: Completing tasks by connecting foundation models with millions of apis,

    Y . Liang, C. Wu, T. Song, W. Wu, Y . Xia, Y . Liu, Y . Ou, S. Lu, L. Ji, S. Maoet al., “Taskmatrix. ai: Completing tasks by connecting foundation models with millions of apis,”Intelligent Computing, vol. 3, p. 0063, 2024

    2024

  37. [37]

    AnyTool: Self-reflective, hierarchical agents for large-scale API calls,

    Y . Du, F. Wei, and H. Zhang, “AnyTool: Self-reflective, hierarchical agents for large-scale API calls,”arXiv preprint arXiv:2402.04253, 2024

    work page arXiv 2024

  38. [38]

    Re-invoke: Tool invocation rewriting for zero- shot tool retrieval,

    Y . Chen, J. Yoon, D. S. Sachan, Q. Wang, V . Cohen-Addad, M. Bateni, C.-Y . Lee, and T. Pfister, “Re-invoke: Tool invocation rewriting for zero- shot tool retrieval,”arXiv preprint arXiv:2408.01875, 2024

    work page arXiv 2024

  39. [39]

    Tool- Gen: Unified tool retrieval and calling via generation,

    R. Wang, X. Han, L. Ji, S. Wang, T. Baldwin, and H. Li, “Tool- Gen: Unified tool retrieval and calling via generation,”arXiv preprint arXiv:2410.03439, 2024

    work page arXiv 2024

  40. [40]

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

    G. Wang, Y . Xie, Y . Jiang, A. Mandlekar, C. Xiao, Y . Zhu, L. Fan, and A. Anandkumar, “V oyager: An open-ended embodied agent with large language models,”arXiv preprint arXiv:2305.16291, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  41. [41]

    ToolNet: Connecting large language models with massive tools via tool graph,

    X. Liu, Z. Peng, X. Yi, X. Xie, L. Xiang, Y . Liu, and D. Xu, “ToolNet: Connecting large language models with massive tools via tool graph,” arXiv preprint arXiv:2403.00839, 2024

    work page arXiv 2024

  42. [42]

    Graph of Skills: Dependency-Aware Structural Retrieval for Massive Agent Skills

    D. Li, Z. Li, H. Du, X. Wu, S. Gui, Y . Kuang, and L. Sun, “Graph of skills: Dependency-aware structural retrieval for massive agent skills,” arXiv preprint arXiv:2604.05333, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  43. [43]

    SkillNet: Create, evaluate, and connect AI skills,

    Y . Liang, R. Zhong, H. Xu, C. Jiang, Y . Zhong, R. Fang, J.-C. Gu, S. Deng, Y . Yao, M. Wanget al., “SkillNet: Create, evaluate, and connect AI skills,”arXiv preprint arXiv:2603.04448, 2026

    work page arXiv 2026

  44. [44]

    Planning and acting in partially observable stochastic domains,

    L. P. Kaelbling, M. L. Littman, and A. R. Cassandra, “Planning and acting in partially observable stochastic domains,”Artificial intelligence, vol. 101, no. 1-2, pp. 99–134, 1998

    1998

  45. [45]

    Between mdps and semi-mdps: A framework for temporal abstraction in reinforcement learning,

    R. S. Sutton, D. Precup, and S. Singh, “Between mdps and semi-mdps: A framework for temporal abstraction in reinforcement learning,”Artificial intelligence, vol. 112, no. 1-2, pp. 181–211, 1999

    1999