upstream/VoltAgent-awesome-ai-agent-papers/catalogue/README.md

@@ -2,9 +2,9 @@
 title: "Awesome AI Agent Papers"
 task: ""
 lineage_type: import
-upstream_source: https://github.com/VoltAgent/awesome-ai-agent-papers/blob/aa50c0c3/README.md
-upstream_sha: aa50c0c3
-imported_at: 2026-07-03
+upstream_source: https://github.com/VoltAgent/awesome-ai-agent-papers/blob/c8502b6a/README.md
+upstream_sha: c8502b6a
+imported_at: 2026-08-08
 prompt_class: catalogue
 upstream_changes: accepted
 author: upstream
@@ -69,12 +69,13 @@ Hundreds of papers are published on arXiv every week, and a growing number of th
 <br>
 
 <details open id="multi-agent">
-<summary><h3 style="display:inline">Multi-Agent (53)</h3></summary>
+<summary><h3 style="display:inline">Multi-Agent (54)</h3></summary>
 
 <br>
 
 | Paper | arXiv ID |
 |---|:---:|
+| **[Agon: An Autonomous Large-Scale Omnidisciplinary Research System Built on Prompt Economy](https://arxiv.org/pdf/2606.24177)** - Treats prompt engineering as an engineering discipline rather than an art, minimizing human time while expecting maximum output. Carries the smallest prompt footprint among cross-disciplinary autoresearch systems (18 roles, 230.6 KiB total), has run across 10+ research fields without modification, and the longest observed run went 30 days unattended. | <a href="https://arxiv.org/abs/2606.24177"><img src="https://img.shields.io/badge/arXiv-2606.24177-b31b1b.svg" alt="arXiv" /></a> |
 | **[PerceptUI: LLM Agents as Human-Aligned Synthetic Users for UI/UX Evaluation](https://arxiv.org/pdf/2606.05697)** - A persona-conditioned framework that predicts how a specific user would answer UI/UX evaluation questions and explains why in natural language. Trained via contrastive reflection fine-tuning and reflective prompt evolution, reaching human-level realism and generalizing to unseen questions and personas. | <a href="https://arxiv.org/abs/2606.05697"><img src="https://img.shields.io/badge/arXiv-2606.05697-b31b1b.svg" alt="arXiv" /></a> |
 | **[AutoNumerics: An Autonomous, PDE-Agnostic Multi-Agent Pipeline for Scientific Computing](https://arxiv.org/pdf/2602.17607)** - A multi-agent pipeline that reads a PDE problem description in plain text and writes, debugs, and validates a classical numerical solver end-to-end. Generates spectral and finite-difference code (no neural networks), scoring ~6 orders of magnitude below FNO and CodePDE baselines. | <a href="https://arxiv.org/abs/2602.17607"><img src="https://img.shields.io/badge/arXiv-2602.17607-b31b1b.svg" alt="arXiv" /></a> |
 | **[Beyond Offline A/B Testing: Context-Aware Agent Simulation for Recommender System Evaluation](https://arxiv.org/abs/2604.09549)** - Evaluates recommender systems via agent-RS interactions. | <a href="https://arxiv.org/abs/2604.09549"><img src="https://img.shields.io/badge/arXiv-2602.06039-b31b1b.svg" alt="arXiv" /></a> |
@@ -206,7 +207,7 @@ Hundreds of papers are published on arXiv every week, and a growing number of th
 <br>
 
 <details id="eval--observability">
-<summary><h3 style="display:inline">Eval & Observability (80)</h3></summary>
+<summary><h3 style="display:inline">Eval & Observability (81)</h3></summary>
 
 <br>
 
@@ -214,6 +215,7 @@ Hundreds of papers are published on arXiv every week, and a growing number of th
 |---|:---:|
 | **[PerspectiveGap: A Benchmark for Multi-Agent Orchestration Prompting](https://arxiv.org/pdf/2606.08878)** - A 110-scenario benchmark testing whether LLMs can compose orchestration prompts that distribute context to sub-agents without information leakage. Across 10 topologies and 27 commercial models, GPT-5.5 leads with 62% pass rate while the average is 14.9%. | <a href="https://arxiv.org/abs/2606.08878"><img src="https://img.shields.io/badge/arXiv-2606.08878-b31b1b.svg" alt="arXiv" /></a> |
 | **[ClawBench: Evaluating Browser Agents on Live Production Websites with Submission-Interception](https://arxiv.org/abs/2604.08523)** - Benchmarks browser agents on 283 everyday tasks (V1 153 + V2 130) across 163 live production sites, with a Chrome-extension plus CDP layer that blocks only the final write request so agents can run end-to-end on real sites without real-world side effects. Two-stage scoring (interception + LLM judge); leaderboard at https://claw-bench.com. | <a href="https://arxiv.org/abs/2604.08523"><img src="https://img.shields.io/badge/arXiv-2604.08523-b31b1b.svg" alt="arXiv" /></a> |
+| **[StructEval: Benchmarking LLMs' Capabilities to Generate Structural Outputs](https://arxiv.org/pdf/2505.20139)** - Benchmarks structured-output generation and cross-format conversion across 18 text and renderable formats, with syntax, structural, and visual evaluation checks. | <a href="https://arxiv.org/abs/2505.20139"><img src="https://img.shields.io/badge/arXiv-2505.20139-b31b1b.svg" alt="arXiv" /></a> |
 | **[From Features to Actions: Explainability in Traditional and Agentic AI Systems](https://arxiv.org/pdf/2602.06841v1)** - Compares attribution-based explanations with trace-based diagnostics across static and agentic settings to study how explainability methods translate to multi-step agent trajectories. | <a href="https://arxiv.org/abs/2602.06841v1"><img src="https://img.shields.io/badge/arXiv-2602.06841-b31b1b.svg" alt="arXiv" /></a> |
 | **[Agentic Uncertainty Reveals Agentic Overconfidence](https://arxiv.org/pdf/2602.06948v1)** - Investigates whether agents can accurately predict their own success rates in agentic tasks. | <a href="https://arxiv.org/abs/2602.06948v1"><img src="https://img.shields.io/badge/arXiv-2602.06948-b31b1b.svg" alt="arXiv" /></a> |
 | **[AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents](https://arxiv.org/pdf/2602.06855v1)** - Introduces 20 research tasks from real ML papers covering idea generation, experiments, and refinement for benchmarking science agents. | <a href="https://arxiv.org/abs/2602.06855v1"><img src="https://img.shields.io/badge/arXiv-2602.06855-b31b1b.svg" alt="arXiv" /></a> |
@@ -258,6 +260,7 @@ Hundreds of papers are published on arXiv every week, and a growing number of th
 | **[The Why Behind the Action: Unveiling Internal Drivers via Agentic Attribution](https://arxiv.org/pdf/2601.15075v2)** - Proposes a hierarchical framework for general agentic attribution that identifies internal factors driving agent actions through temporal likelihood dynamics and perturbation-based analysis. | <a href="http://arxiv.org/abs/2601.15075v2"><img src="https://img.shields.io/badge/arXiv-2601.15075-b31b1b.svg" alt="arXiv" /></a> |
 | **[Tokenomics: Quantifying Where Tokens Are Used in Agentic Software Engineering](https://arxiv.org/pdf/2601.14470v1)** - Analyzes token consumption patterns across software development lifecycle stages in a multi-agent system to identify where tokens are consumed and which stages drive cost. | <a href="http://arxiv.org/abs/2601.14470v1"><img src="https://img.shields.io/badge/arXiv-2601.14470-b31b1b.svg" alt="arXiv" /></a> |
 | **[APEX-Agents](https://arxiv.org/pdf/2601.14242v2)** - Introduces a benchmark of 480 long-horizon, cross-application productivity tasks created by investment banking analysts, consultants, and lawyers for evaluating AI agent capabilities in realistic work environments. | <a href="http://arxiv.org/abs/2601.14242v2"><img src="https://img.shields.io/badge/arXiv-2601.14242-b31b1b.svg" alt="arXiv" /></a> |
+| **[DRIFT: Detecting Representational Inconsistencies for Factual Truthfulness](https://arxiv.org/pdf/2601.14210)** - Lightweight probes (3M-37M params) trained on middle-layer hidden states catch factually wrong generations before they reach the user, at less than 0.1% overhead. Up to 13 AUROC points above final-layer baselines, and the probes transfer across datasets without retraining. | <a href="https://arxiv.org/abs/2601.14210"><img src="https://img.shields.io/badge/arXiv-2601.14210-b31b1b.svg" alt="arXiv" /></a> |
 | **[CooperBench: Why Coding Agents Cannot be Your Teammates Yet](https://arxiv.org/pdf/2601.13295v2)** - Introduces a benchmark of 600+ collaborative coding tasks to evaluate whether coding agents can coordinate as effective teammates under various coordination structures. | <a href="http://arxiv.org/abs/2601.13295v2"><img src="https://img.shields.io/badge/arXiv-2601.13295-b31b1b.svg" alt="arXiv" /></a> |
 | **[Insider Knowledge: How Much Can RAG Systems Gain from Evaluation Secrets?](https://arxiv.org/pdf/2601.13227v1)** - Investigates how RAG systems can game nugget-based LLM judge evaluations through metric overfitting, demonstrating near-perfect scores when evaluation elements are leaked or predictable. | <a href="http://arxiv.org/abs/2601.13227v1"><img src="https://img.shields.io/badge/arXiv-2601.13227-b31b1b.svg" alt="arXiv" /></a> |
 | **[Replayable Financial Agents: A Determinism-Faithfulness Assurance Harness for Tool-Using LLM Agents](https://arxiv.org/pdf/2601.15322v1)** - Introduces the Determinism-Faithfulness Assurance Harness for measuring trajectory determinism and evidence-conditioned faithfulness in tool-using LLM agents across 74 configurations and 12 models. | <a href="http://arxiv.org/abs/2601.15322v1"><img src="https://img.shields.io/badge/arXiv-2601.15322-b31b1b.svg" alt="arXiv" /></a> |