This paper presents Ark (Agent Research Kit), a minimal open-source coding agent designed for research and education that preserves the essential architectural mechanisms of modern coding agents while emphasizing simplicity and clarity, and introduces ArkBench, a lightweight benchmark comprising ten representative software maintenance and evolution tasks.
Abstract
Coding agents have rapidly emerged as the primary interface for AI-assisted software development. However, despite their growing adoption, relatively little is known about their internal architecture, and no systematic architectural description comparable to those available for compilers or operating systems currently exists. This paper addresses this gap by documenting the main architectural components of coding agents, explaining their responsibilities, interactions, and execution flow. To support this effort, we also present Ark (Agent Research Kit), a minimal open-source coding agent designed for research and education that preserves the essential architectural mechanisms of modern coding agents while emphasizing simplicity and clarity. We also introduce ArkBench, a lightweight benchmark comprising ten representative software maintenance and evolution tasks. Using gpt-5.4-mini, Ark successfully solved 8 of the 10 tasks while requiring modest token consumption. Finally, we compare the architecture of Ark with those of state-of-the-art coding agents using a recently proposed architectural taxonomy. We hope that both Ark and ArkBench provide a practical foundation for teaching, research, and experimentation on coding agents.
This work compares six tool architectures that hold the underlying information and actions similar while varying how they are organized and exposed to the model, across three actors and a total of 11,700 trajectories to show that, even when tools provide similar capabilities, tool architecture changes agent behavior.
Xiangzhe Xu, H. Saghir, Qian-Hui Wu et al.· 1 citation
ICAE-Bench, a benchmark for evaluating coding agents under interactive project-building settings, starts from a fuzzy product requirement, simulating the dynamic paradigm with an automated User Agent, and introduces three key designs.
Zhongyuan Peng, Dan Huang, Chuyu Zhang et al.· arXiv.org· 3 citations· ⚡1
This survey aims to clarify the conceptual boundaries of self-evolving coding agents and provide a foundation for designing more adaptive, reliable, and software-aware agentic systems.
H. Zhou, Haichuan Hu, Tianyu Luo et al.· 0 citations
Overall, repository-preserved Agent Plans under these tool-specific directories appear to be a narrow but informative artifact for studying task intent and execution guidance in human-agent workflows.
M. Abubakar, Seyedmoein Mohsenimofidi, Jai Lal Lulla et al.· 1 citation
A unified, taxonomy-driven, and deployment-oriented survey of agentic AI systems, synthesizing recent advances through a modular reference architecture and a four-dimensional taxonomy that characterizes agents along the axes of autonomy, tool use, collaboration, and safety–governance is presented.
Sparsh Bajoria, Shreyanshu Ranjan, Adhitya M et al.· Cognitive Computation· 0 citations
Large Language Models (LLMs) have enabled the emergence of autonomous AI agents capable of
reasoning, planning, tool use, and iterative decision-making. Despite rapid development, the field
remains architecturally fragmented, with limited conceptual clarity regarding memory
integration, planning mechanisms, and operational reliability.
This study presents a systematic review and critical synthesis of LLM-based autonomous agents,
focusing on architectural paradigms, memory models, planning strategies, and real-world
deployment constraints. Using a structured review approach, this study examines existing LLM
based agent systems across key design components to uncover common patterns, differences in
implementation, and recurring structural weaknesses.
The review reveals persistent and structurally significant challenges across all four dimensions:
long-horizon reasoning stability degrades as task length increases; memory consistency is
undermined by retrieval noise, embedding drift, and summarisation errors; tool alignment failures
propagate errors across modular pipelines; and evaluation standardisation remains insufficient
to support reliable cross-paper comparison. A consistent cross-paradigm finding emerges:
autonomy and reliability trade off systematically as agent complexity increases, with current
systems achieving capability gains through heuristic design rather than principled theoretical
foundations. Based on this synthesis, the review proposes a consolidated analytical framework
that maps common structural elements and trade-offs across reviewed systems, and outlines a
research agenda directed toward formalised agent architectures, memory consistency guarantees,
verified planning algorithms, standardised reliability metrics, and benchmark frameworks
adequate for long-horizon, real-world evaluation conditions.
Unknown authors· International Journal of Com...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
