Agent research papers and materials¶
简体中文 · English
This index is built from docs/research-manifest.json, generated from the shared paper specifications.
Maintain metadata in packaged paper.yaml files and per-domain catalog specifications, not generated pages.
Open a paper's write-up for background, architecture, equations, reported results, and local evidence.
Implemented papers and research materials¶
| Topic | Method | First author's institution and first publication | Author code | Local entry point |
|---|---|---|---|---|
| Harnesses and self-improvement | Strategist: Evolving Reasoning with Agents | Princeton University, 2026-11-02 | zlab-princeton/Strategist, public. The upstream initial blueprint library is empty and must be populated by real search; it does not ship all the paper's discovered strategies | strategist — Low-cost component only; full experiments not reproduced. |
| Unified evaluation and reliability | AgentHeal: Evaluating Goal-Preserving Resource Recovery of Web Agents under Access Barriers | University of Science and Technology of China, 2026-10-29 | Lzzzzzzzzzzy/AgentHeal, author code and data are public | agentheal — Low-cost component only; full experiments not reproduced. |
| Agent memory | Remembering Is Not Enough: Refining Memories and Learning When to Use Them in Multi-Agent Systems | East China Normal University, 2026-10-27 | ECNU-ICALK/DiscernMem, MIT, reviewed at 0d59f5fce155cbf5c695ada7f040c1d283e43ada. Local rules are independently implemented; no user data was copied. |
discernmem — Low-cost component only; full experiments not reproduced. |
| Agent self-improvement | When the Critic Falls Behind: Co-Evolving Rubrics and Harnesses for Continual Agent Self-Improvement | Harbin Institute of Technology, 2026-10-27 | No public author repository was found in the paper or author links as of 2026-10-29; no code license or revision was verified. | ruharness — Low-cost component only; full experiments not reproduced. |
| Self-improvement and workflows | Where Should the Evaluation Budget Go in LLM-Driven Agent Scaffold Evolution? | The University of Hong Kong, 2026-10-27 | No verifiable EvalAlloc repository found in the reviewed v1 sources. Referenced frameworks such as ShinkaEvolve are not EvalAlloc code. | evalalloc — Low-cost component only; full experiments not reproduced. |
| Self-improvement and workflows | EMMA: Structured Episodic Memory for Multi-Agent Workflow Synthesis | University of California San Diego, 2026-10-25 | No verifiable EMMA implementation repository found in the reviewed v1 sources. Referenced projects such as smolagents are not EMMA source code. | emma — Low-cost component only; full experiments not reproduced. |
| Agentic OPD / rollout budgeting | Frontier-OPD: Discovering the Edge of Agent Capability and Teaching Agents to Go Beyond | Tsinghua University, 2026-10-23 | The paper links jiangz20/FrontierOPD, but the repository and raw LICENSE were inaccessible on 2026-10-26. No public source, revision or license verified; paper CC BY-NC-ND 4.0. | frontier-opd — Low-cost component only; full experiments not reproduced. |
| Agent evaluation and self-improvement | HarnessMM: Recursive Self-Improvement for Harness Token Efficiency | Tsinghua University, 2026-10-23 | No verifiable method repository, revision or source license. The linked Codex, DeepSeek and Terminal-Bench baseline commits are not EDG/GDR/CCS source; the paper's arXiv license is not a code license. | Planned key harnessmm — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Agent evaluation and self-improvement | ImageRSI: Verification-Driven Recursive Self-Improvement for Agentic Image Generation | University of Science and Technology of China, 2026-10-23 | No verifiable author repository in the reviewed v1. No method source revision/license; paper and original figure CC BY-SA 4.0. | Planned key image-rsi — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Agent evaluation and self-improvement | RSI-Opt: Recursive Self-Improvement Optimizer Agents For Model Training | Purdue University, 2026-10-23 | No verified method source. OPAL is the underlying framework, not the cross-run RSI-Opt policy evolution code. Paper/figure CC BY 4.0. | Planned key rsi-opt — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Agent training and policy optimization | Asymmetric On-Policy Distillation Learns In-Context Exploration | University of Washington, 2026-10-21 | WEIRDLabUW/asteroid at 95eee5ee66488af3aaa0aa55a09ae163b1ca3778, linked from the author project. No LICENSE/COPYING or open license verified; independent local implementation. |
asteroid — Low-cost component only; full experiments not reproduced. |
| Public evaluation protocols | PLE: Physical-Engineering's Last Exam | Google Cloud AI Research, 2026-10-21 | google-research/ple, Apache-2.0 at 70ed96281d87380926a0d7c08e21799fd5edee63 (October 23 public-dev protocol). Paper CC BY 4.0; data license separate. |
physical-engineering-exam — Low-cost component only; full experiments not reproduced. |
| Agent runtime | Cache Engineering for Agents | Tsinghua University, 2026-10-20 | Author AKV repository, verified from the paper backlink and repository description. | Planned key cache-engineering — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Self-improvement | Evolving Agent Interfaces | University of California, Santa Barbara, 2026-10-20 | Agent-S / agent_s4, Apache-2.0, pinned 58187e364e3024076540892f5f69d1efc218b74f. |
agent-s4 — Low-cost component only; full experiments not reproduced. |
| Self-improvement | Object-Oriented Harness: Evolving Agent Harnesses through Object-Oriented Design | University of Illinois Urbana-Champaign, 2026-10-20 | No verified author implementation. Benchmark and baseline repositories are not OO-Harness source. | oo-harness — Low-cost component only; full experiments not reproduced. |
| Memory and long-term interaction | Remembering Has a Price: Learning Memory Worth under Resource Constraints | Nanjing University, 2026-10-19 | AIGNLAI/TokenWorth includes tokenworth_core and environments; the official tasks were not run locally. | tokenworth — Low-cost component only; full experiments not reproduced. |
| Memory and long-term interaction | MemPrice: Online Memory Allocation via Adaptive Pricing for Long-Horizon Agents | Hong Kong University of Science and Technology, 2026-10-18 | The paper lists an anonymous repository, which could not be accessed and verified in this review. The complete official implementation is not confirmed. | memprice — Low-cost component only; full experiments not reproduced. |
| Scientific research and closed-loop experiments | Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery | Google Research, 2026-10-17 | The paper lists third-party research tools and supplementary materials, but no complete author agent-harness repository was found. Open-source dependencies do not make the entire author system open source. | Planned key aptamer-lab-loop — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Retrieval and tools | Learning to Program Retrievals: Composing Search Primitives via Reinforcement Learning | KAIST AI / Meta Superintelligence Labs, 2026-10-17 | October 29 follow-up verified Meta's retrieval-programs repository at 992fda1f5f0eb01a8fb70202ff8f7a65aed48465. Original work is CC BY-NC 4.0; vendored components retain their individual licenses, as documented in LICENSE. This supersedes the earlier finding that no repository link had been located. |
rpl — Low-cost component only; full experiments not reproduced. |
| Planning, reasoning, and tool use | SCOPE: Selective Certification of Program Edits for Embodied Heuristic Learning | Peking University, 2026-10-16 | VDIGPKU/SCOPE provides code and scripts under Apache-2.0. Its single-task example is not the full paper matrix. | scope-replay — Low-cost component only; full experiments not reproduced. |
| Multi-agent collaboration | World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds | Massachusetts Institute of Technology, 2026-10-16 | The paper says its framework and traces are released through the project page. That page returned 404 in this review, so the repository and license could not be verified. This establishes neither that the code is unreleased nor that a runnable implementation has been verified. | Planned key world-of-swecraft — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Multi-agent collaboration | A team of Gemini 4 Argon agents beats Minecraft | Google DeepMind, 2026-10-15 | No complete experiment harness found. Mineflayer is an upstream game library, not the research source. | Planned key gemini-minecraft — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Public evaluation and protocols | AgentPerturb: Evaluating LLM Agent Robustness under Controlled Interaction Perturbations | The Hong Kong University of Science and Technology (Guangzhou), 2026-10-15 | The official xuyj233/AgentPerturb repository was still empty when checked on October 19, 2026. The method code and data had not been released. | agent-perturb — CPU evaluation interface; does not run a model. |
| Recursive self-improvement | Athanor: An Agent-Native System for Autonomous Post-Training Research | ByteDance / Duke University, 2026-10-15 | The paper describes the system as open. A usable complete official source repository and license were not verified during this review, so it is not marked as open source. | Planned key athanor — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Recursive self-improvement | AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning | Tsinghua University, 2026-10-15 | Thinking-Space/AutoSFT has only a README stating “Source code coming soon”; algorithm source has not been released. | Planned key auto-sft — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Expert model orchestration | Disagreement-Guided Weak-to-Strong Routing | University of Pennsylvania, 2026-10-15 | No runnable code released. The official repository had only a README marked Code coming soon. Appendix A.5 says configurations will be released after acceptance. | disagreement-routing — Low-cost component only; full experiments not reproduced. |
| Tool use and reliability | Diviner: Agentic RCA with Human-Governed Adaptation for Large-Scale Infrastructure | Meta, 2026-10-15 | No public author repository for Diviner was found. Dependencies such as LangChain are not the paper's implementation. | diviner — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | GACC-Bench: Benchmarking Goal-Adaptive Context Compression Ability across Large Language Models | Fudan University, 2026-10-15 | Official code/data entry. The full reuse license was unconfirmed; source-data licensing must be preserved. The Hugging Face data entry was not readable in this review. | gacc-bench — CPU evaluation interface; does not run a model. |
| Public evaluation and protocols | Harness-Delta Attribution: Dissecting Gains From Harness Evolution | Amazon, 2026-10-15 | Wenwen-D/HarnessDeltaAttribution provides HDA skills, bootstrap, framework code, and artifacts. | harness-decomposition — CPU evaluation interface; does not run a model. |
| Long-term memory | Hy-Memory-30B-A3B: Attributed Reinforcement Learning for Long-Term Memory Writing | Tencent Hunyuan, 2026-10-15 | Hy-Memory framework, Apache-2.0; checkpoint entry. Full training materials have not been confirmed public; model terms require separate review. | Planned key hy-memory — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Tool use and reliability | JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMs | University of Illinois Urbana-Champaign, 2026-10-15 | JevAgentHarness/JevAgentHarness is publicly readable; the full reuse license was not verified in this review. | jev-agent-harness — Low-cost component only; full experiments not reproduced. |
| Expert model orchestration | LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use Agents | Simular, 2026-10-15 | No released author method repository was found. License unconfirmed; no third-party implementation was copied. | leash — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | Librarian Agents: Training LLMs to Curate Long-Context Workspaces that Anticipate Future Queries | Stanford University, 2026-10-15 | Official code, MIT. Appendix A.11 states CC-BY-4.0 for results and synthetic artifacts. | librarian-agents — CPU evaluation interface; does not run a model. |
| Memory and long-term interaction | MemoProAgent: Toward Memory-Driven Proactive Assistance | Michigan State University, 2026-10-15 | Meta's Memo-Pro-Agent repository provides a GPM inference/evaluation harness and GRPO reward and launcher entry points, reviewed at 8563e58d3106c8e0b3108bff64c04d00e1c8f095 (2026-10-22 00:48:15 UTC). CC-BY-NC-4.0 is a noncommercial source license, not OSI open source, and does not imply all training/evaluation assets are public. |
proper-agent — Low-cost component only; full experiments not reproduced. |
| Long-term memory | MemScope: Evolving Agent Memory Architectures through Localized Code Repair | Institute of Computing Technology, Chinese Academy of Sciences / University of Chinese Academy of Sciences, 2026-10-15 | No link to a public author implementation was found on the title page or in the method or appendix. | memscope — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | PROBE: Learning to Audit Policy Compliance in Tool-Using LLM Agents | Mohamed bin Zayed University of Artificial Intelligence, 2026-10-15 | Author implementation is publicly readable. Reuse license unconfirmed; the local code independently implements the paper. | probe — Low-cost component only; full experiments not reproduced. |
| Planning and search | Repairing LLM-Generated Specifications with Deterministic Diagnosis | University of Oxford, 2026-10-15 | The paper mentions released code, but no verifiable author URL was found. The corrected dataset was verified; data release is not code release. | drago — Low-cost component only; full experiments not reproduced. |
| RSI | RSIFlow: Adaptive Self-Improvement Workflows with an Evolving Meta-Agent | Beihang University, 2026-10-15 | Author repository, MIT. The released version is the skill-memory variant, with a fixed executable Meta program. | rsiflow — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | SciPIvot: Benchmarking Principal-Investigator Judgment at Pivotal Moments of Scientific AutoResearch Agents | University of Maryland, College Park / The Chinese University of Hong Kong (affiliation marks verified in the PDF), 2026-10-15 | Released data, prompts, and scoring code, CC-BY-4.0. Full project histories have not been released. | scipivot — CPU evaluation interface; does not run a model. |
| Public evaluation and protocols | Shortcut Sensitivity in Agent-Safety Evaluations | Kontext, 2026-10-15 | Public research code, but LICENSES.md explicitly states there is no separate reuse license. That is not a standard open-source license. Local code is independent and does not copy code or original records. |
shortcut-sensitivity — Low-cost component only; full experiments not reproduced. |
| Tool use and routing | SSR: Structural Signal Retriever for Scalable Agent Discovery in LLM-Orchestrated Multi-Agent Systems | State University of Campinas, 2026-10-15 | The authors have a public repository, but no license was found during this review, so it cannot be assumed to be permissively licensed under MIT or similar terms. The local implementation follows the paper independently and does not copy its code. | structural-signal-retriever — Low-cost component only; full experiments not reproduced. |
| RSI | Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine | Tencent Hy Frontier / National University of Singapore, 2026-10-15 | The project page still said Code soon / Model soon. No verifiable method code or license was released. | terminal-rsi — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | ToolBottleneckBench: Measuring Tool-Agent Performance under Controlled Pressures and Fixed Harness Budgets | University of Osaka, 2026-10-15 | Official repository, MIT. No upstream scripts were run. | tool-bottleneck-bench — CPU evaluation interface; does not run a model. |
| tool-retrieval | TRACER: Trajectory-Aware Tool Retrieval for LLM Agents | Microsoft, 2026-10-15 | Not released. Appendix A.13 explicitly says “We release no artifacts here.” No author repository for the method or synthetic data was found. The cited tau/BFCL repositories are not TRACER code. | tracer — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks | Hong Kong Baptist University, 2026-10-14 | tmlr-group/AlphaDiana provides the framework, configurations, scripts, and license. | alpha-diana — CPU evaluation interface; does not run a model. |
| Multi-agent collaboration | ARIA: Runtime Inspection and Amendment for Multi-Agent Systems | Florida State University, 2026-10-14 | LabRAI/ARIA, MIT; training and serving need CUDA, and frameworks and data have separate dependencies or access terms. | aria — Low-cost component only; full experiments not reproduced. |
| Public evaluation and protocols | Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents | Alibaba Token Hub, Alibaba Group, 2026-10-14 | The paper says the rewriting suite and per-episode results are released; no verifiable author repository URL was found in this review. | coda — CPU evaluation interface; does not run a model. |
| Agent memory and context | Maximising Transparency in Context Compaction Improves Retention of Facts and Commitments for Lightweight Coding Agents | Hokkaido University, 2026-10-14 | transparent-compact-pi and the experiment package, both MIT. The paper pins plugin commit 6b970d3. |
transparent-compact — Low-cost component only; full experiments not reproduced. |
| Verifiable capability composition | Recursive Formation and First-Class Reuse of Compositional Capabilities | not listed in paper, 2026-10-14 | Public f-system/V9.4 implementation; no clear reuse license was found, so no code was vendored. | formcraft — Low-cost component only; full experiments not reproduced. |
| RSI | The World Won't Wait: Evaluating and Teaching Agents Under a Live Clock | Peking University / Samsung Research, 2026-10-14 | The paper lists underlying game repositories and revisions. No unified RealtimeBench/TORSI author implementation repository link was found. | torsi — CPU evaluation interface; does not run a model. |
| Public evaluation and protocols | What Coding Agents Leave Unsaid: Measuring Task-State Gaps Between Execution and Reports | Tencent, 2026-10-14 | CTXGAP release; code MIT, original benchmark data and prompts CC BY 4.0. Some private evidence will never be released. | ctxgap — CPU evaluation interface; does not run a model. |
| Public evaluation and protocols | CLASHarness: Benchmarking Multi-Agent Conflict in Computer Use via a Controllable and Verifiable Generation Pipeline | Google / Emory University (joint affiliation), 2026-10-13 | Google Research / CLASHarness release; Apache 2.0 code, CC BY 4.0 benchmark data | clasharness — CPU evaluation interface; does not run a model. |
| agentic-rl | RSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-Evolution | Chinese University of Hong Kong, 2026-10-11 | Released: author repository (read-only review, not downloaded) | Planned key rsipo — Not reproduced for now due to cost; validation pending, no executable entry. |
| Self-improvement and experience reuse | Harmony Harness: Multi-Objective Self-Evolution of Agent Harnesses | Google Cloud AI Research / Carnegie Mellon University, 2026-10-10 | No author implementation repository found in the reviewed paper or official abstract. Anonymous result visualizations are not algorithm source code; release status is unconfirmed. | Planned key harmony-harness — Not reproduced due to cost (review complete); research material only, no executable entry. |
| memory-retrieval | Beyond the Final Embedding: Retrieving Memories from Token-Level Evidence | Southeast University, 2026-10-09 | No author code found in the paper or official pages on 2026-10-13. | beyond-final-embedding |
| agent-evolution | Evo2: Evolving the Evolver through Weight-Harness Co-Evolution | University of Illinois Urbana-Champaign / Google DeepMind, 2026-10-09 | No public implementation found in the paper or author project links as of 2026-10-12. Affiliations follow the first page of the PDF. | evo2 |
| Planning, search, and reflection | Harness²: Recursive Agent Harnessing for an Open World | Northwestern University, 2026-10-09 | google-research/harness2, Apache-2.0; reviewed at pinned commit 1cfe254, not automatically tracked from main |
harness2 — Low-cost component only; full experiments not reproduced. |
| autonomous-research | Janus: Enhancing Long-Horizon Autonomous Research with Co-Evolving World Models | Institute of Computing Technology, Chinese Academy of Sciences, 2026-10-09 | No executable source found as of 2026-10-12. The authors' project page contains a README and assets only. | janus |
| memory-retrieval | NEAR: Neighborhood-Aware Residuals for Memory Retrieval | Southeast University, 2026-10-09 | No author code found in the paper or official pages as of 2026-10-13. | near |
| tool-retrieval | RATED: Reliability-Aware AI Agent Tool Retrieval via Execution Dynamics | Texas Tech University, 2026-10-09 | EcoAgent / RATED | rated |
| recursive-self-improvement | RSI-Anything: Unlocking Recursive Self-Improvement with World Models | PhAI Labs / The Chinese University of Hong Kong, 2026-10-09 | Yes: RSI-Anything. | rsi-anything |
| long-term-memory | Test-Time Memory Construction | Northeastern University, 2026-10-09 | Yes: T2Mem. | t2mem |
| agent-evaluation | When Faster Decisions Do Not Make Better Agents: An Empirical Study of Jev in Agent Workflows | Nanjing University, 2026-10-09 | No verified release. The authors provided an anonymous experiment-code link, but only its frontend loaded; the API timed out, so usable source was not verified (2026-10-12). | jev-workflow |
| Program evolution | EvoAlloc: A Self-Evolving Resource Allocation Agent for Efficient Program Evolution | King Abdullah University of Science and Technology, 2026-10-08 | No public author code found in the text or project links as of 2026-10-10. | evoalloc |
| World models | Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks | University College London, 2026-10-08 | No public implementation found. The baseline1 starting code mentioned in the paper is not a Memento 3 release. | memento3 |
| Agent safety | One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails | University of Southern California, 2026-10-08 | Released; independent local implementation | option-channel-attack |
| Agent safety | OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport | Cribl AI, 2026-10-08 | The paper says code and configs are public. No verifiable public repository was found in this review, so this is not an author-implementation integration. | ontrack |
| Recursive self-improvement | Recursive Self-Improvement through Multi-Agent Self-Supervision | UC Berkeley, 2026-10-08 | Public implementation | mass |
| Long-term memory | Use and Disuse: Intent-Structured Experience Consolidation for Memory and Learning in LLM Agents | Institute of Information Engineering, Chinese Academy of Sciences, 2026-10-08 | No public author code was found in the full text or project links as of 2026-10-10. | hippocam |
| Agentic RL | When Should Agents Think? Adaptive Reasoning via Cross-Turn Estimation | Renmin University of China, 2026-10-08 | The announced repository was empty as of 2026-10-10; no implementation was available. | race |
| Agent safety | SpecGuard: Proving a Task Is Broken Before the Agent Cheats | MATS, 2026-10-06 | Yes: prmbiy/specguard, Apache-2.0, pinned at 15820b5, not tracking main. |
specguard — Low-cost component only; full experiments not reproduced. |
| Agentic RL | AdaStep: Adaptive Step Credit Weighting for Agentic Reinforcement Learning | Tsinghua University, 2026-10-02 | No author code found in the reviewed HTML as of 2026-10-05. | adastep |
| Agentic RL | Credit Where It Matters: Dependency-Aware Policy Optimization for Terminal Agents | Southeast University, 2026-10-02 | No public implementation found in the paper or public repository search as of 2026-10-06. | depgpo |
| Program evolution | FrugalEvo: Towards Cost-Aware LLM-Guided Program Evolution | National University of Singapore, 2026-10-02 | Authors' implementation, Apache-2.0; checked revision 82f7b739110944bf3e8737db303a302a568e7327 |
frugalevo |
| Self-improvement and experience reuse | SEER: Self-Evolving Event Reasoning and Retrieval for Time Series Forecasting | Google Cloud AI Research / University of Virginia, 2026-10-02 | BennyTMT/SEER exists. Its inspected root contained README, assets, paper and citation only; executable core code was not confirmed. | Planned key seer-forecasting — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Failure recovery | Sentry: Learning to Recover from LLM Agent Failures at Test Time | Stanford University, 2026-10-02 | The author repository at 83064d18a99f47a0a8032049eaa2e1e4085351ad had documentation, paper and images only; README commands are not released executable source. |
sentry |
| Harness optimization | ActiveSaddler: Automated Curriculum Learning for Agent Harness Optimization | POSTECH, 2026-10-01 | Yes: microsoft/AutoSaddler, ActiveSaddler branch | active-saddler |
| Coding Agent | AutoCompact: Learning When to Compact Context in Long-Horizon Coding Agents | First author's affiliation not listed on the first page, 2026-10-01 | No public author implementation found as of 2026-10-03. | autocompact |
| Long-horizon agents | Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief States | First author's affiliation not listed on the first page, 2026-10-01 | No public author implementation found as of 2026-10-03. | belief-state-pos |
| Causal retrieval intervention | Causal Memory Policy: Making Memory Utility Identifiable by Intervening on Retrieval | Illinois Institute of Technology, 2026-10-01 | The paper links an anonymous release. Access failed during this review; upstream code was not run. | causal-memory-policy |
| Simulated lookahead and action selection | Code Owns the Simulation, Jev Owns the Evaluation | The Chinese University of Hong Kong, 2026-10-01 | No public author repository was found during the 2026-10-04 full-text review. | jev-lookahead |
| Failure attribution | DeFA: Dependency-Guided Failure Attribution for LLM Agents | Qwen DianJin Team, Alibaba Cloud Computing, 2026-10-01 | No public author implementation found as of 2026-10-02. | defa |
| Workflow evolution | It Takes Workflows to Evolve Better Workflows | William & Mary, 2026-10-01 | No public author implementation found as of 2026-10-02. | flowright |
| Agent memory | Mem++: Non-Destructive Memory for Long-Term Organizational LLM Agents | The University of Texas at Austin, 2026-10-01 | AIDAChip-Inc/mem-plus-plus. | mem-plus-plus |
| Local agent harnesses | Mingbird: A Local-First Agent Harness Enabling Small Open Models to Complete Real Tasks | University of Science and Technology Beijing, 2026-10-01 | Yes: Mingbird/Mingbird-agent, Apache-2.0 | mingbird |
| Agent safety | PACE: Provenance-Aware Capability Enforcement for Tool-Using LLM Agents | First author's affiliation not listed on the first page, 2026-10-01 | No public author implementation found as of 2026-10-03. | pace-capability |
| RSI safety | Safety Must Survive Self-Improvement: Why Failures Persist and How Agents Recover | First author's affiliation not listed on the first page, 2026-10-01 | No public author implementation found as of 2026-10-02. | safe-self-improvement |
| Agent verification | VeriHarness: Scaling Agentic Verification for Long-Horizon Tasks | Google Cloud AI Research, 2026-10-01 | No public author implementation found as of 2026-10-02. | veriharness |
| Jev | JevSpawn: Adaptive Agentic Inference through Compositional Action Spaces | First author's affiliation not listed on the first page, 2026-09-30 | No public author implementation found as of 2026-10-03. | jev-spawn |
| Agent memory | MemFit: Efficient Long-Term Agentic Memory | First author's affiliation not listed on the first page, 2026-09-30 | No public author implementation found as of 2026-10-03. | memfit |
| RSI | Self-Evolving Coding Rules for AI Coding Agents | First author's affiliation not listed on the first page, 2026-09-30 | No public author implementation found as of 2026-10-03. | rule-evolve |
| Context management | Context Language Models | University of Washington / Meta Superintelligence Labs, 2026-09-29 | Yes: facebookresearch/context-language-models | context-lm |
| Harness design | Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI | Apodex / University of Illinois Urbana-Champaign, 2026-09-29 | Yes: qiancheng-apodex/MetaSkill-AI4AI | meta-skills |
| Harness optimization | Mixture of Self-Improving Branches for Agent Harness Optimization | Affiliation not listed on the first page, 2026-09-29 | No public author repository found as of 2026-10-01. | branch-mixture |
| Long-term memory | ReMem: Rethinking Perception and Memory in Long-Context Recommendation Agents | The Hong Kong Polytechnic University, 2026-09-29 | Yes: Quhaoh233/ReMem | remem |
| Agent reasoning | Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning | Meta Superintelligence Labs, 2026-09-29 | No public author repository was found as of 2026-10-01. | meta-reasoning |
| Agent memory | UpliftMem: Learning Set-Level Uplift for Agent Memory Retrieval | Beihang University, 2026-09-29 | No author repository found as of 2026-09-30. | upliftmem |
| Agent evaluation | UserProxyBench: Evaluating LLM User Simulators for Agent Benchmarks and Training | Affiliation not listed in the paper, 2026-09-29 | No author repository was found as of 2026-09-30. | user-proxy-bench |
| RSI | Video-RSI: Recursive Self-Improvement of Video Understanding Agents via Harness Evolution | Tsinghua University, 2026-09-29 | bingjunluo/Video-RSI. | video-rsi |
| Agent evaluation | Certified Selective Automation of LLM Agent Evaluation | Independent Researcher, 2026-09-28 | No public author repository was found as of 2026-09-30. | certified-selective-eval |
| Context management | Continuous Context Management | University of Miami, 2026-09-28 | No public author repository was found as of 2026-09-30. | ccm |
| Agent RL | Dr.Credit: Rubric-Grounded Process Credit Assignment for Deep Research Agents | University of Chinese Academy of Sciences, 2026-09-28 | No public author repository found as of 2026-09-30. | dr-credit |
| Agentic RL | GraphHCA: Closed-Form Hindsight Credit Assignment for Long-Horizon LLM Agents | Beihang University, 2026-09-28 | No author code found or released as of 2026-09-30. | graphhca |
| Harness learning | Harness Learning Enables Generalizable Test-Time Adaptation | Johns Hopkins University, 2026-09-28 | No public author code found as of 2026-09-30. | harness-learning |
| Long-term memory | Mnemon: Raw Records, Fast Judgments, Slow Thoughts | Affiliation not listed in the paper, 2026-09-28 | No public author repository found as of 2026-09-30. | mnemon |
| Planning | SAGE: Symbolic Action-Gating and Editing for LLM Task Planners | Affiliation not listed on the first page, 2026-09-28 | mtbui2010/sage is public. | sage-planner |
| Research automation | Advancing Model Research in AgentX: Long-Horizon Autonomy for Industrial Recommender Systems | Kuaishou, 2026-09-24 | No public AgentX-Model implementation, production code, experimental graph, or logs found as of 2026-09-28. | agentx-model-replay |
| Programming agents | An Empirical Study of Harness Design for Coding Agents | University of Massachusetts Amherst, 2026-09-17 | No public author implementation found as of 2026-09-19. | harness-design-study |
| Multi-agent systems | CERA-MoA: Co-Evolving Router and Agents for Mixture-of-Agents | IIIS, Tsinghua University, 2026-09-16 | No public author implementation repository was found as of 2026-09-19. | cera-moa |
| Trajectory refinement | Dependency-Aware Trajectory Refinement for Efficient Multi-Turn Agent Fine-Tuning | ShanghaiTech University, 2026-09-16 | Alibaba-NLP/VLLM-KB is public. | dependency-refinement |
| Long-term memory | Interactive Memory Learning for Long-Term Conversations | Harbin Institute of Technology, Shenzhen / Pengcheng Laboratory, 2026-09-15 | No public author repository found as of 2026-09-16. | interactive-memory |
| GUI Agent | Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents | Zhejiang University, 2026-09-15 | Yes: ZJU-REAL/EvoSkill-GUI | evoskill-gui |
| Programming agents | RepoAtlas: Guiding Coding Agents via Evolving Multimodal Repository Views | Beihang University, 2026-09-15 | No public author repository found as of 2026-09-16. | repoatlas |
| Scientific agents | ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents | Gen-Verse research collaboration, 2026-09-15 | Yes: Gen-Verse/ScienceBuddy | sciencebuddy |
| Agent evaluation | Verifiable Social Reasoning for LLM Assistants | Google Research / Hebrew University of Jerusalem / University of Cambridge, 2026-09-15 | No public author repository was found as of 2026-09-16. | fuse-evaluator |
| Agent RL | HarnessBandit: Joint Learnability-Transferability Scheduling for Multi-Harness Agentic Reinforcement Learning | Harbin Institute of Technology / Alibaba Cloud, 2026-09-12 | No public author repository found as of 2026-09-16. | harness-bandit |
| Coding agents | Beyond Vector Similarity: Hierarchical Context-Aware Graph RAG vs Standard RAG in Enterprise Code Migration | Google Cloud, 2026-09-11 | No pipeline found in the paper or Google entry; spring-petclinic is input, not author implementation. | hcrg — Low-cost component only; full experiments not reproduced. |
| Skill evolution | COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization | The Chinese University of Hong Kong, Shenzhen, 2026-09-10 | https://github.com/Jerry-LuP/COBRA-Skills | cobra-skills |
| Harness self-improvement | Ecdysis: Efficient and Effective Training of Runtime Harnesses for LLM Agents | Chengdu Institute of Computer Applications, Chinese Academy of Sciences, 2026-09-10 | cuiyu-ai/Ecdysis. | ecdysis |
| Persistent optimization planning | MAPLE: Memory-Augmented Planning with Language and Evolution | Harbin Institute of Technology, Shenzhen, 2026-09-10 | MAPLE. | maple |
| Agent memory | Grounding Agent Memory: Environment-Probing Curation for Enterprise Agents | Microsoft, 2026-09-09 | No public author repository was found on 2026-09-14. | grounded-memory |
| Search agents | SearchAtlas: Analyzing Agentic Search Strategies via Evidential Query Graphs | Duke University, 2026-09-09 | https://github.com/DukeNLP/SearchAtlas | searchatlas |
| Agent RL | T1: Terminal Agent Reinforcement Learning for Long-Horizon Tasks | National University of Singapore / Tencent, 2026-09-09 | No public author repository found as of 2026-09-14. | t1-terminal-rl |
| Skill retrieval | When Synthetic Data Hurts: On Catastrophic Forgetting in Skill Retrieval for LLM Agents | Manulife, 2026-09-09 | Author EMNLP 2026 repository. | skill-retention |
| Environment feedback scaffolding | Environments as Scaffold: Enriching Feedback to Bootstrap Self-Evolving Agents in Long-Horizon Tasks | Fudan University, 2026-09-08 | EnvAsScaffold | feedback-scaffold |
| Event-tree memory compression | MemForest: Efficient Agent Memory Management via EventTree Partitioning and Progressive Merging | Shanghai Jiao Tong University, 2026-09-08 | MemForest. | memforest |
| Self-improving program graphs | Procedural Graphs: Self-Evolving Execution Structures for LLM Agents | Google, 2026-09-08 | No public author repository was found on 2026-09-12. | procedural-graphs |
| Agentic recommendation memory | AtomRec: Evolving Atomic Memory for Agentic Recommendation | Xi'an Jiaotong-Liverpool University, 2026-09-04 | No public author repository was found (checked 2026-09-07). | atomrec |
| Hierarchical skill coevolution | CoSkill: Joint Reinforcement Learning of Reasoning and Meta-Skill Agents for Hierarchical Skill Evolution | Institute of Automation, Chinese Academy of Sciences, 2026-09-04 | Released: jinyuan-cookie/CoSkill. | coskill |
| Structure-preserving verifier and reward | SiLR: Structure-Preserving Admission and Process Reward for LLM Tool Agents | Institute of Science Tokyo, 2026-09-03 | No public author repository was found (checked 2026-09-07). | silr |
| Multi-harness RL audit | What Does Multi-Harness RL Learn? Credit Assignment and Portability in Coding Agents | New York University, 2026-09-03 | The paper says result artifacts were released, but no verifiable public repository link was found (checked 2026-09-07). | multi-harness-rl |
| skill-retrieval | Imagine Before Acting: Prospective Skill Identification for LLM Agents | Sichuan University, 2026-08-28 | No author implementation found in the paper or official pages as of 2026-10-13. | skilldreamer |
| Paper-to-code state subplanning | DeepRepro: State-Aware Subplanning for Paper-to-Code Reproduction in Evolving Repositories | Institute of Computing Technology, Chinese Academy of Sciences, 2026-08-27 | Released DeepRepro code. | deeprepro |
| Red-team skill evolution | RedEvoAgent: Automatic Red-Teaming Agent with Experience-Driven Skill Evolution | City University of Hong Kong, 2026-08-27 | No public author code found as of 2026-08-31. | redevoagent |
| Agent skill pretraining | SPT: Skills as Pre-Training Data for Agentic Language Models | Beijing University of Posts and Telecommunications, 2026-08-27 | Artifact availability stated; no public repository found as of 2026-08-29. | spt |
| Software trajectory selection | SWE-Prime: Fewer Trajectories, Better Performance | Sun Yat-sen University, 2026-08-27 | No public author code found as of 2026-08-29. | swe-prime |
| Behavior-sensitive harness verification | Verify Smarter, Evolve Further: Efficient Harness Evolution through Behavior-Aware Verification | Fudan University, 2026-08-27 | jhxu5214/HarnessLens. | harnesslens |
| Agent data quality | What Makes Good Agentic Data? An ACE Lens on Data Generation for LLM Agents | Huawei Technologies Co., Ltd, 2026-08-27 | Survey paper; no public author code found on 2026-08-31. | ace-data |
| Trainable collaborative memory | When Memory Takes Gradients: Collaborative Vector Memory for Agentic Recommender Systems | Shenzhen Technology University, 2026-08-27 | No public author code was found when checked on 2026-08-29. | covemem |
| Adaptive tool use and reflection for video research | AdaVDR: Adaptive Tool Use and Reflection for Video Deep Research | Accio Team, Alibaba Group, 2026-08-26 | Released on GitHub | adavdr |
| Counterfactual causal skill graphs | CaSKG: Counterfactual-Causal Skill Graphs for Scalable Agent Skill Retrieval | Jilin University / Ant Group, 2026-08-26 | Released on GitHub | caskg |
| On-demand harness generation and evolution | JIT-Agent: Scaling Harness Intelligence via Just-in-Time Harness Evolution | LV-NUS Lab, 2026-08-26 | Released on GitHub | jit-agent |
| Cost-aware routing by online progress | ProgRouter: Online Progress-Guided Orchestration for Multi-Agent LLM Workflows under Quality-Cost Tradeoffs | Aston University, 2026-08-26 | No author-released code found (checked 2026-08-28). | progrouter |
| Agent runtime | SKILL.state: Scalable Long-Horizon Agent Skills | Google, 2026-08-26 | No official author-released code found. A similarly named community implementation is not author code. | skill-state — Low-cost component only; full experiments not reproduced. |
| Workflow prefix-state scheduling | TOPAS: Workflow-Aware Prefix-State Scheduling for Multi-Agent LLM Serving | University of Science and Technology of China, 2026-08-26 | No public author code found on 2026-08-28. | topas |
| ML development trajectory planning | TraceML: An Empirical Analysis of Human-Agent Planning in Machine Learning Development | Carnegie Mellon University, 2026-08-26 | Released: TraceML dataset and extraction pipeline | traceml |
| Training with environment feedback prompts | AHEAD: Agentic Hints for Effective Agent Development | AWS AI Labs / Purdue University, 2026-08-25 | No author training code identified (checked 2026-08-26); released model weights. | ahead |
| Verifiable skill construction | SkillForge: Automated Skill Discovery and Refinement for Tool-Using Agents | AMAP, Alibaba Group, 2026-08-25 | No public author code found as of 2026-08-26. | skillforge |
| Self-improving tools for multi-dimensional verification | SMITH: Self-Improving Tool-Using Agents through Multi-Aspect Verification | Appier AI Research / National Taiwan University, 2026-08-25 | Yes: appier-research/smith. | smith |
| Asynchronous single-stream agent RL | SPO++: Stabilizing Asynchronous Agentic Reinforcement Learning via Measure-Theoretic Token Correction | Renmin University of China, 2026-08-25 | No public author code was found on 2026-08-26. | spo-plus-plus |
| Gaussian-guided agent RL | Agent-G²: Gaussian Guidance for Agentic Reinforcement Learning | Baidu, 2026-08-24 | Released: ZJU-REAL/Agent-G2 | agent-g2 |
| Automated harness optimization | AutoSaddler: Automatic Harness Optimization with Durable Updates from Agent Execution Traces | Microsoft / POSTECH, 2026-08-24 | No released source found; the paper links a project page. | autosaddler |
| Action-level skill optimization | AUSO: Action-Level Unified Skill Optimization from Internalization to Utilization | University of Science and Technology of China, 2026-08-21 | Yes: Action-Skill. | auso |
| Agentic RL | SAPO: Single-Rollout Autoregressive Policy Optimization for Agentic Reinforcement Learning | Xiamen University, 2026-08-20 | No public author code was found as of 2026-08-24. | sapo |
| Agentic RL | RTPO: Reverse-Turn Policy Optimization for Stabilizing Agentic RL Training | Adelaide University, 2026-08-19 | No public author code was found as of 2026-08-24. | rtpo |
| Agentic RL | SPADE: Self-Play in Adaptive Synthetic Executable Environments | University of Washington, 2026-08-19 | spade | spade |
| Agentic RL | PlanPO: Group Planning-Aware Policy Optimization for Multi-Turn Agentic LLMs | Xiamen University, 2026-08-18 | No public author code was found as of 2026-08-24. | planpo |
| Tool use | Grounding AI Agents in Contracts: An Empirical Evaluation of Spec-Driven Test Generation | Google, 2026-08-17 | No author repository identified on Google's official entry or in v1. The 90 historical production defects are internal; this is not an assertion that no code exists anywhere. | spec-driven-tests — Low-cost component only; full experiments not reproduced. |
| Agentic RL | TRCA: Transition-wise Rubric Credit Assignment for Long-horizon LLM Agents | Authors did not disclose affiliation, 2026-08-17 | No public author code found as of 2026-08-24. | trca |
| Memory | HyMem: Hierarchical Context Management for Long-Horizon Agents via Information Isolation | Institute of Automation, Chinese Academy of Sciences, 2026-08-16 | No public author code was found as of 2026-08-24. | hymem |
| Reflection | LoongReflect: Boosting Long-Horizon Reflection in Search Agents via Global Perspective Distillation | Authors did not disclose affiliation, 2026-08-12 | No public author code found as of 2026-08-24. | loongreflect |
| Agentic RL / efficient long context | Efficient Reinforcement Learning for Long-Horizon Tool-Use Agentic Tasks | Capital One AI Foundations, 2026-08-11 | No author-released code found (checked 2026-08-13). | sinkflex-rl |
| Self-improvement | OpenLoopEvolve: A Verifiable Self-Evolution Framework for Loop Policies in Long-Horizon Complex Tasks | Tsinghua University, 2026-08-10 | Yes: OpenLoopEvolve. | openloopevolve |
| Coding agents | Coupling Planning with Episodic Memory in LLM Agents for Software Issue Resolution | Vanderbilt University, 2026-08-07 | No public author code was found as of 2026-08-24. | pmcoder |
| Multi-turn agent RL | ResidencyRL: Reinforcement Learning in Simulated Clinical Environments | Google DeepMind, 2026-08-07 | Explicitly not released; Code Availability cites internal training infrastructure and safety restrictions. Third-party benchmarks are not the implementation. | Planned key residency-rl — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Recursive turn credit | AgentOPSD | Tsinghua University, 2026-08-06 | ZethWang/AgentOPSD is public; README says complete code is pending. | agent-opsd |
| Code retrieval agents | CodeGrep: An RL-Trained Retrieval Agent for LLM Coding Agents | NetEase Guangzhou AI Lab, 2026-08-06 | No public author code found as of 2026-08-09. | codegrep |
| Search agent RL | Contextual Information Policy Optimization for Search Agents | Beihang University, 2026-08-06 | No public author code was found on 2026-08-09. | cipo |
| Environment rehearsal and agent RL | EnvACE | Shanghai Jiao Tong, 2026-08-06 | Within-yao/EnvACE is public. | envace |
| Harness optimization evaluation | HarnessOpt-Bench: Evaluating LLMs at Harness Optimization | Scale AI, 2026-08-06 | No public author code was found on 2026-08-09. | harnessopt-bench |
| Global skill evolution | Learning Globally Reusable Skills for Coding Agents | Tianjin University, 2026-08-06 | No public author code found as of 2026-08-09. | gse |
| Safety in skill evolution | When Self-Evolution Backfires: Pre-Commit Gating against Skill Contamination in LLM Agents | Institution not stated in the paper, 2026-08-06 | No public author code was found on 2026-08-09. | vag |
| Harness policy RL | EvoHarness-RL: Learning Self-Evolving Runtime Harness for Long-Horizon LLM Agents | University of Illinois Urbana–Champaign / Meta AI, 2026-08-05 | No public author code found as of 2026-08-09. | evoharness-rl |
| End-to-end agent memory | MemoryCPT: An End-to-End Agent Memory Framework for Cost-Performance Trade-off | Hong Kong University of Science and Technology / Tencent LIGHTSPEED STUDIOS, 2026-08-05 | No public author code found as of 2026-08-09. | memorycpt |
| Observation-calibrated distillation | OCSD | Nanjing University, 2026-08-05 | yiy1x/OCSD. | ocsd |
| Training with environment-derived intermediate tasks | State2State: Environment-Derived Mid-Training for LLM Agents | Tsinghua University AIR / Alibaba Group, 2026-08-05 | https://github.com/THUNLP-MT/State2State | state2state |
| Tool planning | ToolLIFT: Lifting Tool-Specific Trajectories into Function-Level Graphs for Generalizable Tool Planning | Authors did not disclose affiliation, 2026-08-04 | No public author code was found as of 2026-08-24. | toollift |
| Verifiable unified memory | VerMem | Sun Yat-sen University, 2026-08-04 | Released code: Sun-SYSU-24/VerMem. | vermem |
| Retrieval–memory co-evolution | CoEvo-Mem | Institution not stated in the paper, 2026-08-03 | No released official code found on 2026-08-08. | coevo-mem |
| Hindsight on search trajectories | HindSearch: Trajectory-Level Hindsight Critique for Search-Augmented Reinforcement Learning | Santa Clara University, 2026-08-03 | https://anonymous.4open.science/r/hindsearch-anon-EBDC | hindsearch |
| Tool planning | HyperAgent: Planning and Acting over Tool-Schema Hypergraphs for Tool-Use LLM Agents | University of New South Wales, 2026-07-31 | No public author code found as of 2026-08-24. | hyperagent |
| Evaluation and verifiable agents | FinanceHarness: Autonomous Financial Deep Research Framework | Google Cloud AI Research, 2026-07-30 | FinanceHarness, pinned to 1163c5e7a8f35ff6217c3a480f7b04562355e0cc |
finance-harness-protocol — Low-cost component only; full experiments not reproduced. |
| Agent group credit | Group-Reflective Self-Distillation | Baidu Inc., 2026-07-30 | BinbZheng1/GRSD. | grsd |
| Multi-agent systems | MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems | Cornell University, 2026-07-30 | No public author code found as of 2026-08-24. | manta |
| Computer-use agent reward evaluation | OSReward / OS-Shepherd | The University of Hong Kong, 2026-07-30 | Project page: code, benchmark, data and weights. | os-shepherd |
| Agent RL | TAPO | Peking University, 2026-07-30 | No official implementation was found when checked on 2026-08-01. | tapo |
| Cost-aware stopping for tool use | CAM-DF | Peking University, 2026-07-29 | The paper mentions cached artifacts and evaluation scripts but no public repository was found. | cam-df |
| Cross-task skill evolution | SkillRise | Zhejiang University, 2026-07-29 | Public implementation | skillrise |
| Agentic RL / turn-level credit | CAST | University of Science and Technology of China, 2026-07-28 | Wloner0809/CAST. | cast |
| Hierarchical skill memory | HiSkill | Beijing University of Posts and Telecommunications, 2026-07-28 | Author code. | hiskill |
| Continual agent memory | UniMem | CASIA, 2026-07-28 | No official public repository was found as of 2026-07-29. | unimem |
| Self-improvement and experience reuse | AgentDebugX: An Open-Source Toolkit for Failure Observability, Attribution, and Recovery in LLM Agents | University of Illinois Urbana-Champaign, 2026-07-21 | AgentDebugX/AgentDebugX, pinned to 2437c73c3183e6b6495259dfbb12464a7f07f6b8 |
Planned key agentdebugx — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Agentic RL / hindsight skill | SEED | Tsinghua University, 2026-07-16 | jinyangwu/SEED. | seed |
| Active memory | Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents | Unspecified (HTML does not map the first author to an affiliation), 2026-07-15 | Released: ericjiang18/MemCon; MIT | memcon |
| Agentic OPD / rollout budgeting | TurnOPD | Academic author team, 2026-07-07 | No official author repository found as of 2026-07-29. | turn-opd |
| Research automation | AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems | Kuaishou, 2026-06-26 | No public author implementation was found as of 2026-08-24. | agentx |
| Multi-agent collaboration | Formalizing Mathematics at Scale | FAIR at Meta / CERMICS, ENPC, Institut Polytechnique de Paris, 2026-05-28 | autoform-bot, MIT; formalization library atlas-lean. No plugin was installed. | Planned key autoformbot — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Scientific research agents | ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence | Google Cloud AI Research, 2026-05-25 | Author artifact repository, Apache-2.0, contains generated papers and solver code only. No source for ScientistOne orchestration or CoE auditing was found. Project page | scientist-one — CPU evaluation interface; does not run a model. |
| Training service optimization | PROMPTS: Performance Optimization via Multi-Agent Planning for LLM Training and Serving | University of Maryland / Google, 2026-05-18 | No public author repository found as of 2026-09-14. | prompts |
| Agent retrieval | Superintelligent Retrieval Agent: The Next Frontier of Agentic Retrieval | Meta Superintelligence Labs / Rice University (work done at Meta), 2026-05-07 | facebookresearch/sira. | sira |
| Step-aligned Agent RL | StepPO | University of Science and Technology of China, 2026-04-20 | No official code found. | steppo |
| Tool calling and environment execution | LogAct: Enabling Agentic Reliability via Shared Logs | Meta, 2026-04-09 | Official MIT-licensed repository, checked at commit db395449fb2dbb534c8eec848900252ac8324ab2; the agent plugin is not installed |
logact — Low-cost component only; full experiments not reproduced. |
| Policy–tool graph co-evolution | SEARL | Shanghai AI Laboratory, 2026-04-09 | No official author repository was found. | searl |
| Agent self-improvement | CROP: Token-Efficient Reasoning in Large Language Models via Regularized Prompt Optimization | Google, 2026-04-08 | As of 2026-10-30, neither the paper nor the Google publication page linked to a verifiable author repository for CROP. The upstream TextGrad project is not a CROP implementation. | crop — Low-cost component only; full experiments not reproduced. |
| Agent evaluation | TFRBench: A Reasoning Benchmark for Evaluating Forecasting Systems | Google Cloud AI Research / University of Kentucky, 2026-04-07 | A benchmark and leaderboard are available; the reviewed pages did not identify a separate algorithm source repository | tfrbench — CPU evaluation interface; does not run a model. |
| Multi-agent collaboration | PaperOrchestra: A Multi-Agent Framework for Automated AI Research Paper Writing | Google Cloud AI Research, 2026-04-06 | No verified author implementation. Author code SHA/license are not applicable. Community Ar9av/PaperOrchestra is not author code. |
Planned key paperorchestra — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Agent data synthesis | Reasoning-Driven Synthetic Data Generation and Evaluation | EPFL, 2026-03-31 | No author implementation identified in the reviewed text; no claim of global absence. | simula — CPU evaluation interface; does not run a model. |
| AI research agents | AIRA2: Overcoming Bottlenecks in AI Research Agents | FAIR at Meta / University College London, 2026-03-27 | As of 2026-10-04, no public author implementation was found in the paper, Meta's official page, or author homepages. | aira2 |
| RSI | HyperAgents | University of British Columbia, 2026-03-19 | Author code. | hyperagents |
| Skill design | Memento-Skills | Memento Team, 2026-03-19 | Public implementation | memento-skills |
| Agent self-improvement | POLCA: Stochastic Generative Optimization with LLM | University of Wisconsin-Madison, 2026-03-16 | Yes: pinned rlx-lab/POLCA. Setup points to Trace experimental; pinned epsilon-net/summarizer implementation verified beyond the README. | polca — Low-cost component only; full experiments not reproduced. |
| Agentic RL | Dual-Modality Multi-Stage Adversarial Safety Training: Robustifying Multimodal Web Agents Against Cross-Modal Attacks | University of California, Berkeley, 2026-03-04 | DMAST_official, pinned SHA e11444116fe26d94d46b8f90ef6bebd8cb728040 |
dmast — Low-cost component only; full experiments not reproduced. |
| Active memory | U-Mem | National University of Singapore, 2026-02-25 | Anonymous review repository | u-mem |
| Agent evaluation | ConvApparel: A Benchmark Dataset and Validation Framework for User Simulators in Conversational Recommenders | Google, 2026-02-18 | No algorithm source found in the paper or Google paper page. Official data is not source code. | convapparel — CPU evaluation interface; does not run a model. |
| Personalized agents | Learning Personalized Agents from Human Feedback | Meta Superintelligence Labs / Princeton University (work done at Meta), 2026-02-18 | facebookresearch/PAHF. | pahf |
| Agent safety | CausalArmor: Efficient Indirect Prompt Injection Guardrails via Causal Attribution | Google Cloud AI Research, 2026-02-08 | No author implementation found in the reviewed paper or Google entry. AgentDojo is an environment, not CausalArmor code. | causal-armor — Low-cost component only; full experiments not reproduced. |
| Automated research | AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents | FAIR at Meta / University of Oxford, 2026-02-06 | facebookresearch/airs-bench, task and evaluation code, CC BY-NC 4.0. | Planned key airs-bench — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Self-improvement and experience reuse | MARS: Modular Agent with Reflective Search for Automated AI Research | Google Cloud AI Research, 2026-02-02 | jfc43/MARS, MIT. Root includes exp-logs, figures, README and LICENSE. Generated code and trajectories are released; the full controller was not verified. | Planned key mars-research — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Memory skills | MemSkill | Nanyang Technological University, 2026-02-02 | Public implementation | memskill |
| Reinforcement learning for planning | PEARL | Institute of Information Engineering, Chinese Academy of Sciences, 2026-01-28 | No author code found or released. | pearl |
| Coding agents | Dynamic Cogeneration of Bug Reproduction Test in Agentic Program Repair | Google, 2026-01-27 | No Passerine implementation found in reviewed paper or official listing. Input/reference repositories are not author code. | dynamic-brt — Low-cost component only; full experiments not reproduced. |
| Evaluation and verifiable agents | LifeAgentBench: Benchmarking LLMs for Long-Horizon, Cross-Dimensional Lifestyle Health Reasoning | University of California San Diego, 2026-01-20 | gdfwj/LifeAgentBench, official project page, pinned to cfc227736965f4cd42261ee780f41050ae44cc67 |
Planned key lifeagentbench — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Agent evaluation and self-improvement | CoReflect: A Reflective Co-Evolution Framework for Improving Conversational Evaluation | University of Illinois Urbana-Champaign, 2026-01-18 | No author implementation is verifiable from pinned v2 or its abstract. Exact-title search found only third-party commentary, not source evidence. Code license and pinned SHA are not applicable; the paper and figure are CC BY 4.0. | Planned key coreflect — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Skill-bank reinforcement learning | SAGE | University of Wisconsin–Madison, 2025-12-18 | No official author repository was found. | sage |
| Zero-data multi-agent systems | Agent0 | University of North Carolina at Chapel Hill, 2025-11-20 | No official author code found or released. | agent0 |
| Agentic RL infrastructure | Agent-R1 | University of Science and Technology of China, 2025-11-18 | Author code. | agent-r1 |
| Procedural memory | LEGOMem | Microsoft Research, 2025-10-06 | No public repository found on the paper page as of 2026-07-27. | legomem |
| Tool and environment interaction | Watch and Learn: Learning to Use Computers from Online Videos | The Ohio State University, 2025-10-06 | No author code or weights verified. The project page's Code placeholder has no link and does not establish a release. Code SHA and license are not applicable. | Planned key watch-and-learn — Not reproduced due to cost (review complete); research material only, no executable entry. |
| Multi-turn user-facing agent RL | MUA-RL | Meituan, 2025-08-26 | zzwkk/MUA-RL | mua-rl |
| Tool learning | ToolGrad: Efficient Tool-Use Dataset Generation with Textual Gradients | Google, 2025-08-06 | Authors' code. | toolgrad |
| Agent RL | Agent Lightning | Microsoft Research, 2025-08-05 | microsoft/agent-lightning. | agent-lightning |
| Tool memory | MemTool | PricewaterhouseCoopers Commercial Technology and Innovation Office, 2025-07-29 | No public repository found on the paper page as of 2026-07-27. | memtool |
| Web agent reinforcement learning | WebAgent-R1 | University of Virginia, 2025-05-22 | weizhepei/WebAgent-R1 | webagent-r1 |
| Agent group credit | GiGPO | Nanyang Technological University, 2025-05-16 | No official author code found. | gigpo |
| Multi-turn agent RL | RAGEN | Northwestern, 2025-04-24 | RAGEN-AI/RAGEN | ragen |
| Reinforcement learning for tool use | ToolRL | University of Illinois Urbana-Champaign, 2025-04-16 | No official author repository found or released. | toolrl |
| Tool calling during reasoning | ReTool | ByteDance Seed, 2025-04-15 | No official author repository found or released. | retool |
| Deep research RL | DeepResearcher | HKU, 2025-04-04 | Author code. | deepresearcher |
| Search agent RL | Search-R1 | University of Illinois Urbana-Champaign, 2025-03-12 | PeterGriffinJin/Search-R1. | search-r1 |
| Tool use and environments | SOPBench: Evaluating Language Agents at Following Standard Operating Procedures and Constraints | University of California, Santa Barbara, 2025-03-11 | Yes: Leezekun/SOPBench at SHA 8790ed7406acbd22270f90bd1c4038de3fc5db8f, MIT. Data have a separate DATA_LICENSE and were not copied locally. | sopbench-protocol — Low-cost component only; full experiments not reproduced. |
| Long-horizon agent RL | LOOP | Apple, 2025-02-03 | apple/ml-loop | loop |
| General-purpose software agents | OpenHands | All-Hands-AI, 2024-07-23 | All-Hands-AI/OpenHands. | openhands |
| Agent–computer interfaces for software engineering | SWE-agent | Princeton University, 2024-05-06 | Public implementation | swe-agent |
| General agent evaluation | GAIA | Meta AI, 2023-11-21 | Yes: official benchmark. | gaia |
| Virtual context | MemGPT | University of California, Berkeley, 2023-10-12 | The authors' public code, now Letta. | memgpt |
| Agent search | LATS | University of Illinois Urbana-Champaign, 2023-10-06 | lapisrocks/LanguageAgentTreeSearch. | lats |
| Multi-agent systems | AutoGen | Microsoft Research, 2023-08-16 | microsoft/autogen is public. | autogen |
| Multi-agent software engineering | MetaGPT | DeepWisdom, 2023-08-01 | Public implementation | metagpt |
| Decoupled planning | ReWOO | Microsoft Research, 2023-05-29 | billxbf/ReWOO. | rewoo |
| Tool instructions and evaluation | ToolBench | Tsinghua University, 2023-05-25 | No official code repository from the authors was found. | toolbench |
| Lifelong learning | Voyager | NVIDIA, 2023-05-25 | Released code. | voyager |
| Tool feedback | CRITIC | Microsoft, 2023-05-19 | Author code. | critic |
| Reasoning search | Tree of Thoughts | Princeton University, 2023-05-17 | princeton-nlp/tree-of-thought-llm. | tree-of-thoughts |
| Memory and reflection | Generative Agents | Stanford University, 2023-04-07 | Public author code. | generative-agents |
| Multi-agent collaboration | CAMEL | King Abdullah University of Science and Technology, 2023-03-31 | Author code. | camel |
| Expert model orchestration | HuggingGPT | Zhejiang University, 2023-03-30 | microsoft/JARVIS. | hugginggpt |
| Self-iteration | Self-Refine | Carnegie Mellon University, 2023-03-30 | madaan/self-refine. | self-refine |
| Self-reflection | Reflexion | Northeastern University, 2023-03-20 | Released | reflexion |
| Automated tool reasoning | ART | University of Washington, 2023-03-16 | bhargaviparanjape/language-programmes. | art |
| Tool learning | Toolformer | Meta AI Research, 2023-02-09 | No official training code released; public GitHub implementations are third party. | toolformer |
| Program reasoning | PAL | Carnegie Mellon University, 2022-11-18 | reasoning-machines/pal. | pal |
| Reasoning and acting | ReAct | Princeton University, 2022-10-06 | Released | react |
| Neurosymbolic routing | MRKL | AI21 Labs, 2022-05-01 | The complete Jurassic-X/MRKL implementation was not released. | mrkl |
| Embodied planning | SayCan | Robotics at Google, 2022-04-04 | Simulated tabletop implementation. | saycan |
| Web-browsing question answering | WebGPT | OpenAI, 2021-12-17 | Full training code was not released. The WebGPT comparisons dataset is public. | webgpt |
| Formal mathematics agents | AutoformBench: A Lean Formalization Benchmark for Advanced Mathematics | FAIR at Meta, Not recorded | Released: facebookresearch/autoformbench, pinned to af1aa38fd44e019e566bce0b5a168609a886d8e6. |
autoformbench — CPU evaluation interface; does not run a model. |
Browse by: