pith. machine review for the scientific record. sign in

arxiv: 2605.09038 · v1 · submitted 2026-05-09 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

SearchSkill: Teaching LLMs to Use Search Tools with Evolving Skill Banks

Jinchao Hu, Kehai Chen, Meizhi Zhong, Min Zhang

Pith reviewed 2026-05-12 01:48 UTC · model grok-4.3

classification 💻 cs.AI
keywords search toolsquery planningskill banklanguage modelstool usequestion answeringsupervised fine-tuningfailure patterns
0
0 comments X

The pith

Language models achieve higher accuracy on knowledge-intensive questions by selecting reusable search skills from an evolving bank before generating queries.

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

The paper introduces SearchSkill to address poor query planning in language models that use search tools. Rather than issuing broad or repeated queries directly, the model selects a skill from a dynamic collection and then produces a search or answer step guided by that skill's description. The skill bank expands and refines itself by detecting repeated failure patterns and rebuilding affected training examples. A two-stage fine-tuning process then aligns the model to this select-then-execute routine at inference time. This yields measurable gains in exact-match accuracy and more efficient retrieval across open- and closed-source models.

Core claim

SearchSkill makes query planning explicit through reusable search skills. At each step the model first selects a skill, then generates a search or answer action conditioned on the selected skill card. The skill inventory is not fixed: the framework maintains an evolving SkillBank, expands or refines it from recurrent failure patterns, and reconstructs affected trajectories before supervised training. The resulting two-stage SFT recipe aligns training with the inference-time protocol of skill selection followed by skill-grounded execution, producing better exact match scores and improved retrieval behavior on knowledge-intensive QA tasks.

What carries the argument

The evolving SkillBank, a dynamic collection of reusable search skills that the model selects from before conditioning its next action on the skill card.

If this is right

  • Higher exact-match scores on knowledge-intensive QA benchmarks for both open-source and closed-source models.
  • Fewer copied first queries during search.
  • More atomic, hop-focused queries instead of broad ones.
  • More correct answers obtained within a fixed small search budget.

Where Pith is reading between the lines

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

  • The same skill-selection mechanism could be applied to other tool-using domains such as code execution or API calls.
  • An evolving skill bank might allow models to adapt to new knowledge domains without complete retraining.
  • Reduced waste from ineffective queries could lower the computational cost of retrieval-augmented generation pipelines.

Load-bearing premise

Recurrent failure patterns can be automatically identified and converted into refined skills that produce genuinely better model behavior after the two-stage supervised fine-tuning.

What would settle it

No improvement in exact-match scores or no reduction in copied first queries on knowledge-intensive QA benchmarks when the model is trained and evaluated with the SearchSkill procedure versus standard tool-use fine-tuning.

Figures

Figures reproduced from arXiv: 2605.09038 by Jinchao Hu, Kehai Chen, Meizhi Zhong, Min Zhang.

Figure 1
Figure 1. Figure 1: Overview of SearchSkill: evolve a reusable SkillBank, construct skill-guided trajectories, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of replacing the SkillBank with an empty bank. Left: EM under full [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: SkillBank controls under the same SFT policy. Left: removing selected card content. Right: [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Skill-category contribution after activation. Left: activation versus judged necessity. Right: [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Query-planning diagnostics on four multi-hop benchmarks. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Closed-source transfer with frozen B4. Bars show exact-match percentages; parentheses give gains over search-only prompting. The resulting drops show that activated skills contribute to answer construction, especially when the task requires structured decomposition or evidence grounding. 5.4 Query planning and evidence efficiency We further test whether SEARCHSKILL addresses the query-quality failure that … view at source ↗
Figure 7
Figure 7. Figure 7: GRPO training diagnostics for 7B models. Panels show train reward for 7B-Instruct and [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: RL execution diagnostics on 7B-Instruct examples corrected by GRPO. [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Training and validation loss curves for the two-stage SFT runs across four Qwen2.5 [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: GRPO training dynamics for the 7B-Instruct SFT-initialized policy. The plot shows train [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Cumulative GRPO training reward for the 7B-Base SFT-initialized policy. The plot shows [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
read the original abstract

Teaching language models to use search tools is not only a question of whether they search, but also of whether they issue good queries. This is especially important in open-domain question answering, where broad or copied queries often waste retrieval budget and derail later reasoning. We propose \Ours, a framework that makes query planning explicit through reusable search skills. At each step, the model first selects a skill, then generates a search or answer action conditioned on the selected skill card. The skill inventory itself is not fixed: SearchSkill maintains an evolving SkillBank, expands or refines it from recurrent failure patterns, and reconstructs affected trajectories before supervised training. The resulting two-stage SFT recipe aligns training with the inference-time protocol of skill selection followed by skill-grounded execution. Across open-source and closed-source models, SearchSkill improves exact match on knowledge-intensive QA benchmarks and yields better retrieval behavior, including fewer copied first queries, more atomic hop-focused queries, and more correct answers within a small search budget. These results suggest that explicit skill-conditioned query planning is a lightweight alternative to treating search as an undifferentiated action.

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 proposes SearchSkill, a framework for improving LLM search tool use in open-domain QA by making query planning explicit: at each step the model selects a reusable skill from an evolving SkillBank and then generates a skill-conditioned search or answer action. The SkillBank is dynamically expanded or refined from automatically detected recurrent failure patterns, after which affected trajectories are reconstructed; a two-stage SFT procedure then aligns the training distribution with the inference-time protocol of skill selection followed by skill-grounded execution. The central empirical claim is that this yields higher exact-match scores on knowledge-intensive QA benchmarks for both open- and closed-source models, together with improved retrieval behavior (fewer copied first queries, more atomic hop-focused queries, and higher success rates within a small search budget).

Significance. If the reported gains are reproducible and the evolving-SkillBank component is shown to be responsible for them, the work would offer a lightweight, interpretable alternative to undifferentiated tool-use fine-tuning. Explicit skill cards and failure-driven evolution could reduce wasted retrieval budget and improve multi-hop reasoning efficiency; the two-stage SFT alignment is a sound design choice that matches training to inference. The absence of any quantitative results, baselines, or ablations in the provided description, however, prevents a firm assessment of whether these benefits materialize or whether they stem primarily from the skill-conditioned action format rather than the evolution mechanism.

major comments (2)
  1. [§4 and §3.2] §4 (Experiments) and §3.2 (SkillBank Evolution): the central attribution of gains to the evolving SkillBank is not supported by any ablation that compares the full method against a fixed initial SkillBank under the identical two-stage SFT protocol. Without this isolation, improvements could arise entirely from the skill-selection format and SFT alignment rather than from automatic failure-pattern refinement and trajectory reconstruction.
  2. [Results section] Results section: the abstract asserts exact-match improvements and better retrieval statistics (fewer copied queries, more atomic hops, higher success within budget) yet supplies no tables, numerical values, baseline comparisons (e.g., ReAct, standard tool-use SFT), number of runs, or statistical tests. This absence makes it impossible to judge whether the data support the stated claims.
minor comments (2)
  1. [§3] Notation for the SkillBank and skill cards is introduced without a clear formal definition or example card; a table or figure showing an initial SkillBank and an evolved entry would improve readability.
  2. [§3.2] The description of how recurrent failure patterns are automatically identified and turned into new or refined skills lacks pseudocode or a concrete example, leaving the evolution procedure underspecified.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We agree that stronger isolation of the evolving SkillBank's contribution and more explicit quantitative reporting are needed to support the central claims. We address each major comment below.

read point-by-point responses

  1. Referee: [§4 and §3.2] §4 (Experiments) and §3.2 (SkillBank Evolution): the central attribution of gains to the evolving SkillBank is not supported by any ablation that compares the full method against a fixed initial SkillBank under the identical two-stage SFT protocol. Without this isolation, improvements could arise entirely from the skill-selection format and SFT alignment rather than from automatic failure-pattern refinement and trajectory reconstruction.

    Authors: We agree that the manuscript does not contain an ablation isolating the evolving SkillBank against a fixed initial SkillBank under the identical two-stage SFT protocol. The reported experiments compare the full SearchSkill pipeline to ReAct and standard tool-use SFT baselines, but do not include this specific control. We will add the requested ablation in the revised version to demonstrate whether the automatic failure-pattern detection and trajectory reconstruction contribute beyond the skill-selection format and SFT alignment alone. revision: yes

  2. Referee: [Results section] Results section: the abstract asserts exact-match improvements and better retrieval statistics (fewer copied queries, more atomic hops, higher success within budget) yet supplies no tables, numerical values, baseline comparisons (e.g., ReAct, standard tool-use SFT), number of runs, or statistical tests. This absence makes it impossible to judge whether the data support the stated claims.

    Authors: We acknowledge that the submitted version does not present tables, numerical values, baseline comparisons, run counts, or statistical tests in the results section. The full manuscript will be revised to include comprehensive tables reporting exact-match scores, retrieval behavior metrics, direct comparisons to ReAct and standard tool-use SFT, experimental details on the number of runs, and any statistical tests. The abstract will also be updated to reference these concrete results. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical framework evaluated on external benchmarks

full rationale

The paper describes SearchSkill as an empirical training recipe: explicit skill selection followed by skill-grounded execution, with an evolving SkillBank that identifies recurrent failures, expands or refines skills, and reconstructs trajectories for a two-stage SFT procedure. All performance claims (exact-match gains, fewer copied queries, more atomic hops, higher success within budget) are presented as outcomes measured on held-out knowledge-intensive QA benchmarks across open- and closed-source models. No equations, uniqueness theorems, or self-citations are invoked to derive results by construction; the method is not self-definitional, and no fitted parameter is relabeled as a prediction. The derivation chain is therefore self-contained against external evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

The framework introduces the SkillBank as a central new component whose construction and evolution rest on author-defined procedures for detecting failure patterns; no external benchmarks or prior independent evidence for this component are cited in the abstract.

invented entities (1)
  • SkillBank no independent evidence
    purpose: Evolving inventory of reusable search skills that is expanded or refined from recurrent failure patterns
    Core invented construct of the proposed framework; the abstract provides no independent evidence or falsifiable handle outside the method itself.

pith-pipeline@v0.9.0 · 5498 in / 1202 out tokens · 59015 ms · 2026-05-12T01:48:04.426951+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 11 internal anchors

  1. [1]

    Self-rag: Learn- ing to retrieve, generate, and critique through self-reflection

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. Self-rag: Learn- ing to retrieve, generate, and critique through self-reflection. InThe Twelfth International Conference on Learning Representations, 2023

    work page 2023

  2. [2]

    Rae, Erich Elsen, and Laurent Sifre

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, Diego de Las Casas, Aurelia Guy, Jacob Menick, Roman Ring, Tom Hennigan, Saffron Huang, Loren Maggiore, Chris Jones, Albin Cassirer, Andy Brock, Michela Paganini, Geoffrey Irving, Ori...

    work page 2022

  3. [3]

    Unified active retrieval for retrieval augmented generation

    Qinyuan Cheng, Xiaonan Li, Shimin Li, Qin Zhu, Zhangyue Yin, Yunfan Shao, Linyang Li, Tianxiang Sun, Hang Yan, and Xipeng Qiu. Unified active retrieval for retrieval augmented generation. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 17153–17166, 2024

    work page 2024

  4. [4]

    Retrieval augmented language model pre-training

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. Retrieval augmented language model pre-training. InInternational conference on machine learning, pages 3929–3938. PMLR, 2020

    work page 2020

  5. [5]

    Constructing a multi-hop qa dataset for comprehensive evaluation of reasoning steps

    Xanh Ho, Anh-Khoa Duong Nguyen, Saku Sugawara, and Akiko Aizawa. Constructing a multi-hop qa dataset for comprehensive evaluation of reasoning steps. InProceedings of the 28th International Conference on Computational Linguistics, pages 6609–6625, 2020

    work page 2020

  6. [6]

    Cascade: Cumulative agentic skill creation through autonomous development and evolution

    Xu Huang, Junwu Chen, Yuxing Fei, Zhuohan Li, Philippe Schwaller, and Gerbrand Ceder. Cascade: Cumulative agentic skill creation through autonomous development and evolution. arXiv preprint arXiv:2512.23880, 2025

    work page arXiv 2025

  7. [7]

    Leveraging passage retrieval with generative models for open domain question answering

    Gautier Izacard and Edouard Grave. Leveraging passage retrieval with generative models for open domain question answering. InProceedings of the 16th conference of the european chapter of the association for computational linguistics: main volume, pages 874–880, 2021

    work page 2021

  8. [8]

    Atlas: Few-shot learning with retrieval augmented language models.Journal of Machine Learning Research, 24(251): 1–43, 2023

    Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. Atlas: Few-shot learning with retrieval augmented language models.Journal of Machine Learning Research, 24(251): 1–43, 2023

    work page 2023

  9. [9]

    Deepretrieval: Hacking real search engines and retrievers with large language models via reinforcement learning, 2025

    Pengcheng Jiang, Jiacheng Lin, Lang Cao, Runchu Tian, SeongKu Kang, Zifeng Wang, Jimeng Sun, and Jiawei Han. Deepretrieval: Hacking real search engines and retrievers with large language models via reinforcement learning, 2025. URL https://arxiv.org/abs/2503. 00223

    work page 2025

  10. [10]

    Active retrieval augmented generation

    Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. Active retrieval augmented generation. InProceedings of the 2023 conference on empirical methods in natural language processing, pages 7969–7992, 2023

    work page 2023

  11. [11]

    Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

    Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan Arik, Dong Wang, Hamed Za- mani, and Jiawei Han. Search-r1: Training llms to reason and leverage search engines with reinforcement learning, 2025. URLhttps://arxiv.org/abs/2503.09516

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  12. [12]

    Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension

    Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. InProceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1601–1611, 2017. 10

    work page 2017

  13. [13]

    Dense passage retrieval for open-domain question answering

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. InProceedings of the 2020 conference on empirical methods in natural language processing (EMNLP), pages 6769–6781, 2020

    work page 2020

  14. [14]

    Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. Natural questions: A benchmark for question answering research.Transact...

    work page 2019

  15. [15]

    Retrieval-augmented generation for knowledge-intensive nlp tasks, 2020

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. Retrieval-augmented generation for knowledge-intensive nlp tasks, 2020

    work page 2020

  16. [16]

    arXiv preprint arXiv:2603.02176 , year=

    Hao Li, Chunjiang Mu, Jianhao Chen, Siyue Ren, Zhiyao Cui, Yiqun Zhang, Lei Bai, and Shuyue Hu. Organizing, orchestrating, and benchmarking agent skills at ecosystem scale, 2026. URLhttps://arxiv.org/abs/2603.02176

    work page arXiv 2026

  17. [17]

    Search-o1: Agentic Search-Enhanced Large Reasoning Models

    Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, and Zhicheng Dou. Search-o1: Agentic search-enhanced large reasoning models, 2025. URL https://arxiv.org/abs/2501.05366

    work page internal anchor Pith review arXiv 2025

  18. [18]

    arXiv preprint arXiv:2602.08004 , year=

    George Ling, Shanshan Zhong, and Richard Huang. Agent skills: A data-driven analysis of claude skills for extending large language model functionality, 2026. URL https://arxiv. org/abs/2602.08004

    work page arXiv 2026

  19. [19]

    Agent skills in the wild: An empirical study of security vulnerabilities at scale,

    Yi Liu, Weizhe Wang, Ruitao Feng, Yao Zhang, Guangquan Xu, Gelei Deng, Yuekang Li, and Leo Zhang. Agent skills in the wild: An empirical study of security vulnerabilities at scale,

    work page

  20. [20]

    URLhttps://arxiv.org/abs/2601.10338

    work page arXiv

  21. [21]

    When not to trust language models: Investigating effectiveness of parametric and non-parametric memories

    Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Ha- jishirzi. When not to trust language models: Investigating effectiveness of parametric and non-parametric memories. InProceedings of the 61st annual meeting of the association for computational linguistics (volume 1: Long papers), pages 9802–9822, 2023

    work page 2023

  22. [22]

    WebGPT: Browser-assisted question-answering with human feedback

    Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christo- pher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, Xu Jiang, Karl Cobbe, Tyna Eloundou, Gretchen Krueger, Kevin Button, Matthew Knight, Benjamin Chess, and John Schulman. Webgpt: Browser-assisted question-answering with human feedback, 2022. URL https:/...

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  23. [23]

    Measuring and narrowing the compositionality gap in language models

    Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah A Smith, and Mike Lewis. Measuring and narrowing the compositionality gap in language models. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 5687–5711, 2023

    work page 2023

  24. [24]

    arXiv preprint arXiv.2304.08354,

    Yujia Qin, Shengding Hu, Yankai Lin, Weize Chen, Ning Ding, Ganqu Cui, Zheni Zeng, Yufei Huang, Chaojun Xiao, Chi Han, Yi Ren Fung, Yusheng Su, Huadong Wang, Cheng Qian, Runchu Tian, Kunlun Zhu, Shihao Liang, Xingyu Shen, Bokai Xu, Zhen Zhang, Yining Ye, Bowen Li, Ziwei Tang, Jing Yi, Yuzhang Zhu, Zhenning Dai, Lan Yan, Xin Cong, Yaxi Lu, Weilin Zhao, Yux...

    work page arXiv 2024

  25. [25]

    Qwen2.5 Technical Report

    Qwen Team. Qwen2.5 technical report, 2025. URL https://arxiv.org/abs/2412.15115

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  26. [26]

    Toolformer: Language models can teach themselves to use tools.Advances in neural information processing systems, 36: 68539–68551, 2023

    Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach themselves to use tools.Advances in neural information processing systems, 36: 68539–68551, 2023

    work page 2023

  27. [27]

    R1-Searcher: Incentivizing the Search Capability in LLMs via Reinforcement Learning

    Huatong Song, Jinhao Jiang, Yingqian Min, Jie Chen, Zhipeng Chen, Wayne Xin Zhao, Lei Fang, and Ji-Rong Wen. R1-searcher: Incentivizing the search capability in llms via reinforcement learning.arXiv preprint arXiv:2503.05592, 2025. 11

    work page internal anchor Pith review arXiv 2025

  28. [28]

    Zerosearch: Incentivize the search capability of llms without searching.arXiv preprint arXiv:2505.04588, 2025

    Hao Sun, Zile Qiao, Jiayan Guo, Xuanbo Fan, Yingyan Hou, Yong Jiang, Pengjun Xie, Yan Zhang, Fei Huang, and Jingren Zhou. Zerosearch: Incentivize the search capability of llms without searching, 2025. URLhttps://arxiv.org/abs/2505.04588

    work page arXiv 2025

  29. [29]

    MuSiQue: Multihop questions via single-hop question composition.Transactions of the Association for Computational Linguistics, 10:539–554, 2022

    Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. MuSiQue: Multihop questions via single-hop question composition.Transactions of the Association for Computational Linguistics, 10:539–554, 2022

    work page 2022

  30. [30]

    Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions

    Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions. In Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers), pages 10014–10037, 2023

    work page 2023

  31. [31]

    Reinforcement learning for self-improving agent with skill library, 2025

    Jiongxiao Wang, Qiaojing Yan, Yawei Wang, Yijun Tian, Soumya Smruti Mishra, Zhichao Xu, Megha Gandhi, Panpan Xu, and Lin Lee Cheong. Reinforcement learning for self-improving agent with skill library.arXiv preprint arXiv:2512.17102, 2025

    work page arXiv 2025

  32. [32]

    Text Embeddings by Weakly-Supervised Contrastive Pre-training

    Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. Text embeddings by weakly-supervised contrastive pre-training. arXiv preprint arXiv:2212.03533, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  33. [33]

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models, 2022. URLhttps://arxiv.org/abs/2201.11903

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  34. [34]

    SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning

    Peng Xia, Jianwen Chen, Hanyang Wang, Jiaqi Liu, Kaide Zeng, Yu Wang, Siwei Han, Yiyang Zhou, Xujiang Zhao, Haifeng Chen, Zeyu Zheng, Cihang Xie, and Huaxiu Yao. Skillrl: Evolving agents via recursive skill-augmented reinforcement learning, 2026. URL https: //arxiv.org/abs/2602.08234

    work page internal anchor Pith review arXiv 2026

  35. [35]

    Hotpotqa: A dataset for diverse, explainable multi-hop question answering

    Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D Manning. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. InProceedings of the 2018 conference on empirical methods in natural language processing, pages 2369–2380, 2018

    work page 2018

  36. [36]

    ReAct: Synergizing Reasoning and Acting in Language Models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models.arXiv preprint arXiv:2210.03629, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  37. [37]

    MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents

    Haozhen Zhang, Quanyu Long, Jianzhu Bao, Tao Feng, Weizhi Zhang, Haodong Yue, and Wenya Wang. Memskill: Learning and evolving memory skills for self-evolving agents.arXiv preprint arXiv:2602.02474, 2026

    work page internal anchor Pith review arXiv 2026

  38. [38]

    SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills

    Boyuan Zheng, Michael Y . Fatemi, Xiaolong Jin, Zora Zhiruo Wang, Apurva Gandhi, Yueqi Song, Yu Gu, Jayanth Srinivasa, Gaowen Liu, Graham Neubig, and Yu Su. Skillweaver: Web agents can self-improve by discovering and honing skills, 2025. URL https://arxiv.org/ abs/2504.07079. A Experimental setups A.1 Data process We construct the training pool with a cov...

    work page internal anchor Pith review arXiv 2025