📖 Introduction

Overview of self-evolving coding agents.

Coding agents increasingly learn from execution outcomes, trajectories, accumulated experience, and environmental feedback, improving the persistent components of their own software-engineering workflow. This repository accompanies our survey and curates its paper corpus, related methods, benchmarks, products, and related surveys. Self-evolving systems are organized into three layers: assets (memory, skills, tools, context), architecture (harness, workflow, multi-agent structures), and model weights.

🎯 Scope

This repository covers five groups of resources:

1. Self-Evolving Coding Agents: curated systems organized by which part of the agent gets modified: assets (memory, skills, tools, context), architecture (harness, workflow, topology), or model weights.

2. General Self-Evolution Methods in Coding Settings: General agent self-evolution methods whose improvements are evaluated on code generation, program execution, or software engineering tasks.

3. Benchmarks and Empirical Studies: Conventional benchmarks for repository-level software engineering and general coding, dedicated benchmarks for agent self-evolution, and empirical studies of coding-agent self-evolution.

4. Products: deployed coding products with persistent adaptation mechanisms, mapped to the same target vocabulary.

5. Related Surveys: surveys covering self-evolving agents, coding agents, and their intersection.

🌐 Contents

• Introduction
• Scope
• 1. Self-Evolving Coding Agents
  • 1.1 Assets Self-Evolution
    • 1.1.1 Memory Self-Evolution
    • 1.1.2 Skill Self-Evolution
    • 1.1.3 Tool Self-Evolution
    • 1.1.4 Context Self-Evolution
  • 1.2 Architecture Self-Evolution
    • 1.2.1 Harness Evolution
    • 1.2.2 Workflow Evolution
    • 1.2.3 Multi-Agent Evolution
  • 1.3 Model Self-Evolution
    • 1.3.1 Self-Play Co-Evolution
    • 1.3.2 Environment-Reward RL
    • 1.3.3 Harness Co-Evolution
• 3. Benchmarks and Empirical Studies
  • 3.1 Repository-Level Software Engineering Benchmarks
  • 3.2 General Coding Benchmarks
  • 3.3 Self-Evolution Benchmarks
  • 3.4 Empirical Studies
• 4. Self-Evolving Coding Products
• 5. Related Surveys
  • 5.1 Surveys on Self-Evolving Agents
  • 5.2 Surveys on Coding Agents

🤖 1. Self-Evolving Coding Agents

1.1 Assets Self-Evolution

1.1.1 Memory Self-Evolution

1.1.1.1 Experience-Derived Memory

1. Self-Abstraction from Grounded Experience for Plan-Guided Policy Refinement (SAGE) [2025-arXiv]
2. SWE-Exp: Experience-Driven Software Issue Resolution [2025-arXiv] · [Code]
3. LLMs as Continuous Learners: Improving the Reproduction of Defective Code in Software Issues (EvoCoder) [2024-arXiv]
4. Structurally Aligned Subtask-Level Memory for Software Engineering Agents (Subtask Memory) [2026-arXiv]
5. EvoRepair: Enhancing Vulnerability Repair Agents Through Experience-Based Self-Evolution [2026-arXiv]
6. Coupling Planning with Episodic Memory in LLM Agents for Software Issue Resolution (PMCoder) [2026-arXiv]
7. EXPEREPAIR: Dual-Memory Enhanced LLM-based Repository-Level Program Repair [2026-FSE] · [Code]
8. VeriAgent: A Tool-Integrated Multi-Agent System with Evolving Memory for PPA-Aware RTL Code Generation [2026-arXiv]
9. MemRepair: Hierarchical Memory for Agentic Repository-Level Vulnerability Repair [2026-arXiv]
10. ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory [2026-ICLR] · [Code]
11. Reflexion: Language Agents with Verbal Reinforcement Learning [2023-NeurIPS] · [Code]
12. SE-Agent: Self-Evolution Trajectory Optimization in Multi-Step Reasoning with LLM-Based Agents [2025-NeurIPS] · [Code]
13. Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving [2025-ICML Workshop] · [Code]
14. Adaptive Self-improvement LLM Agentic System for ML Library Development [2025-ICML] · [Code]
15. PILOT in the Loop: Live Self-Improvement for Long-Horizon Agents [2026-arXiv]
16. SWE-MeM: Learning Adaptive Memory Management for Long-Horizon Coding Agents [2026-arXiv]
17. From Knowledge to Noise: CTIM-Rover and the Pitfalls of Episodic Memory in Software Engineering Agents [2025-REALM] · [Code]

1.1.1.2 External Knowledge Memory

1. Improving Code Localization with Repository Memory [2026-ICLR]
2. Your Code Agent Can Grow Alongside You with Structured Memory (MemCoder) [2026-arXiv]
3. Learning to Commit: Generating Organic Pull Requests via Online Repository Memory [2026-arXiv]
4. Self-Improving AI Coding Agents Through Accumulated Behavioral Rules: A Closed-Loop Framework [2026-IEEE ICE]

1.1.2 Skill Self-Evolution

1.1.2.1 Skill Acquisition

1. CODESKILL: Learning Self-Evolving Skills for Coding Agents [2026-arXiv]
2. Automatically Learning Skills for Coding Agents (GSkill) [2026-ACM CAIS]
3. Socratic-SWE: Self-Evolving Coding Agents via Trace-Derived Agent Skills [2026-arXiv]
4. Learning Globally Reusable Skills for Coding Agents (GSE) [2026-arXiv]
5. SkillForge: Self-Distilling Agents for Project-Specific Issue Resolution [2026-arXiv] · [Code]
6. Verilog-Evolve: Feedback-Driven and Skill-Evolving Verilog Generation [2026-arXiv] · [Code]
7. Trace2Skill: Verifier-Guided Skill Evolution for Long-Context EDA Agents [2026-arXiv]
8. From Procedural Skills to Strategy Genes: Towards Experience-Driven Test-Time Evolution [2026-arXiv] · [Code]

1.1.2.2 Skill Optimization

1. EffiSkill: Agent Skill Based Automated Code Efficiency Optimization [2026-arXiv]
2. Do Personalized Skills Help Coding Agents? An Empirical Study of Developer Interaction Histories [2026-arXiv]
3. SkillMOO: Multi-objective Optimization of Agent Skills for Software Engineering [2026-ASE] · [Code]
4. daVinci-kernel: Co-Evolving Skill Selection, Summarization, and Utilization via RL for GPU Kernel Optimization [2026-arXiv] · [Code]

1.1.2.3 Skill Governance

1. Ratchet: How Reliable Must an LLM Judge Be to Retire a Skill? [2026-arXiv] · [Code]
2. SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution [2026-arXiv] · [Code]
3. Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents [2026-arXiv] · [Code]

1.1.2.4 Skill Safety

1. Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries [2026-arXiv] · [Code]
2. When Self-Evolution Backfires: Pre-Commit Gating against Skill Contamination in LLM Agents [2026-arXiv]
3. EVOMAL: Self-Poisoning in Self-Evolving Coding Agents [2026-arXiv]

1.1.3 Tool Self-Evolution

1. Live-SWE-agent: Can Software Engineering Agents Self-Evolve on the Fly? [2025-arXiv] · [Code]
2. SIGA: Self-Evolving Coding-Agent Adapters for Scientific Simulation [2026-arXiv]

1.1.4 Context Self-Evolution

1.1.4.1 Context Retrieval

1. EVOR: Evolving Retrieval for Code Generation [2024-EMNLP Findings] · [Code]
2. CodeMEM: AST-Guided Adaptive Memory for Repository-Level Iterative Code Generation [2026-ACL Findings] · [Code]

1.1.4.2 Context Compression

1. A Self-Evolving Framework for Efficient Terminal Agents via Observational Context Compression (TACO) [2026-arXiv] · [Code]
2. SWE-Pruner: Self-Adaptive Context Pruning for Coding Agents [2026-arXiv] · [Code]

1.1.4.3 Prompt Evolution

1. Automated Prompt Engineering for Cost-Effective Code Generation Using Evolutionary Algorithm (EPiC) [2026-TOSEM] · [Code]
2. Prompt Optimization for LLM Code Generation via Reinforcement Learning [2026-arXiv]
3. SePO: Self-Evolving Prompt Agent for System Prompt Optimization [2026-arXiv]
4. From Failing to Passing: Evolving Natural Language Prompt Optimization Rules for LLM Code Generation [2026-arXiv]

1.1.4.4 Environment Evolution

1. Libra: Training the Environment for Agentic Information Retrieval [2026-arXiv] · [Code]
2. Learning to Build the Environment: Self-Evolving Reasoning RL via Verifiable Environment Synthesis (EvoEnv) [2026-arXiv]

1.2 Architecture Self-Evolution

1.2.1 Harness Evolution

1.2.1.1 Archive-Based Evolution

1. Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents (DGM) [2026-ICLR] · [Code]
2. Mendel Gödel Machine: Comparative Evolution Enables State-of-the-Art Self-Improving Coding Agents (Mendel GM) [2026-OpenReview] · [Code]
3. Huxley Gödel Machine: Human-Level Coding Agent Development by an Approximation of the Optimal Self-Improving Machine (Huxley GM) [2026-ICLR] · [Code]
4. Self-Improvement via Fast Tree-Search (SIFT) [2026-ICLR]
5. The Red Queen Gödel Machine: Co-Evolving Agents and Their Evaluators (RQGM) [2026-arXiv]
6. HyperAgents [2026-arXiv] · [Code]

1.2.1.2 Iteration-Based Evolution

1. A Self-Improving Coding Agent (SICA) [2025-ICLR] · [Code]
2. Self-Taught Optimizer (STOP): Recursively Self-Improving Code Generation [2024-COLM] · [Code]
3. Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution [2026-arXiv] · [Code]

1.2.1.3 Trace-Driven Evolution

1. Self-Harness: Harnesses That Improve Themselves [2026-arXiv] · [Code]
2. AutoSaddler: Automatic Harness Optimization with Durable Updates from Agent Execution Traces [2026-arXiv] · [Code]
3. Evo-Harness: Context-to-Harness Skill Compilation for Self-Evolving Agents [2026-arXiv] · [Code]
4. Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses (AHE) [2026-arXiv] · [Code]
5. From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flaws (HarnessFix) [2026-arXiv] · [Code]
6. ReCreate: Reasoning and Creating Domain Agents Driven by Experience [2026-ACL] · [Code]
7. Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement [2026-arXiv] · [Code]
8. One Recipe, Many Harnesses: What Self-Evolution Encodes Across Languages and Models [2026-arXiv]

1.2.1.4 Population-Based Search

1. DarwinX: Evolving Agent Harnesses Through Natural Selection [2026-arXiv]
2. HarnessBank: Semantic Gene-Bank Search with Gated Verification for Agent-Harness Self-Evolution [2026-arXiv]
3. EvolveNet: Collaborative Harness Evolution for Agent Self-Improvement [2026-arXiv] · [Code]
4. HarnessCompass: Guiding Automatic Harness Evolution toward Generalizable and Effective Agent Harnesses [2026-arXiv]
5. Automated Design of Agentic Systems (ADAS) [2025-ICLR] · [Code]

1.2.1.5 Pipeline-Driven Evolution

1. Meta-Harness: End-to-End Optimization of Model Harnesses [2026-arXiv] · [Code]
2. AgentDevel: Reframing Self-Evolving LLM Agents as Release Engineering [2026-arXiv]
3. HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry [2026-arXiv] · [Code]
4. Confucius Code Agent: Scalable Agent Scaffolding for Real-World Codebases (CCA) [2025-arXiv] · [Code]
5. Autogenesis: A Self-Evolving Agent Protocol [2026-arXiv] · [Code]
6. Self-Evolving Agents with Anytime-Valid Certificates (SEA) [2026-arXiv]

1.2.1.6 Runtime Evolution

1. TTHE: Test-Time Harness Evolution [2026-arXiv] · [Code]
2. Adapting the Interface, Not the Model: Runtime Harness Adaptation for Deterministic LLM Agents (Life-Harness) [2026-arXiv] · [Code]
3. Argus: A General-Purpose Agentic Reasoning Runtime for Long-Horizon Tasks [2026-arXiv] · [Code]

1.2.2 Workflow Evolution

1. AFlow: Automating Agentic Workflow Generation [2025-ICLR] · [Code]
2. SEW: Self-Evolving Agentic Workflows for Automated Code Generation [2025-arXiv] · [Code]
3. EvoFlow: Evolving Diverse Agentic Workflows On The Fly [2025-arXiv]
4. EvoAgentX: An Automated Framework for Evolving Agentic Workflows [2025-EMNLP Demos] · [Code]
5. FlowEvo: Self-Evolving Agents through the Co-Evolution of Workflows and Executable Skills [2026-arXiv] · [Code]
6. JUDGEFLOW: Agentic Workflow Optimization via Block Judge [2026-arXiv]

1.2.3 Multi-Agent Evolution

1. Self-Evolving Multi-Agent Collaboration Networks for Software Development (EvoMAC) [2025-ICLR] · [Code]
2. AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation [2026-ICML]
3. SEMAG: Self-Evolutionary Multi-Agent Code Generation [2026-arXiv]
4. Group-Evolving Agents: Open-Ended Self-Improvement via Experience Sharing (GEA) [2026-arXiv] · [Code]
5. SAGE: Multi-Agent Self-Evolution for LLM Reasoning [2026-arXiv]
6. Evolve as a Team: Collaborative Self-Evolution for LLM-Based Multi-Agent Systems [2026-arXiv] · [Code]

1.3 Model Self-Evolution

1.3.1 Self-Play Co-Evolution

1. Toward Training Superintelligent Software Agents through Self-Play SWE-RL [2026-ICML]
2. ReVeal: Self-Evolving Code Agents via Iterative Generation-Verification [2026-ICLR] · [Code]
3. CURE: Co-Evolving LLM Coder and Unit Tester via Reinforcement Learning [2025-NeurIPS] · [Code]
4. ZeroCoder: Can LLMs Improve Code Generation Without Ground-Truth Supervision? [2026-arXiv]
5. Learning to Solve and Verify: A Self-Play Framework for Code and Test Generation (Sol-Ver) [2025-NeurIPS]
6. ACE: Self-Evolving LLM Coding Framework via Adversarial Unit Test Generation and Preference Optimization [2026-arXiv]
7. Anchored Self-Play for Code Repair [2026-ICML]
8. Absolute Zero: Reinforced Self-Play Reasoning with Zero Data [2025-NeurIPS] · [Code]
9. OPD-Evolver: Cultivating Holistic Agent Evolver via On-Policy Distillation [2026-arXiv] · [Code]

1.3.2 Environment-Reward RL

1. Agent-RLVR: Training Software Engineering Agents via Guidance and Environment Rewards [2026-ICLR]
2. SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution [2025-NeurIPS] · [Code]
3. Self-Improving Language Models for Evolutionary Program Synthesis: A Case Study on ARC-AGI (SOAR) [2025-ICML] · [Code]

1.3.3 Harness Co-Evolution

1. Harness-R1: Learning to Edit Executable Runtime Harnesses from Agent Failure Trajectories [2026-arXiv] · [Code]
2. HELIX: Model-Harness Co-evolution for Recursive Self-Improvement [2026-arXiv] · [Code]

📊 3. Benchmarks and Empirical Studies

This section brings together repository-level software engineering benchmarks and general coding benchmarks to assess coding capabilities, self-evolution benchmarks to evaluate agents' ability to improve themselves, and empirical studies to examine the performance gains, computational costs, and failure modes of self-evolving coding agents.

3.1 Repository-Level Software Engineering Benchmarks

1. SWE-bench: Can Language Models Resolve Real-World GitHub Issues? [2024-ICLR]
2. SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks? [2026-ICML]
3. SWE-EVO: Benchmarking Coding Agents in Long-Horizon Software Evolution Scenarios [2025-arXiv]
4. SWE-bench Multimodal: Do AI Systems Generalize to Visual Software Domains? [2025-ICLR]
5. Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving [2025-NeurIPS Datasets and Benchmarks]
6. SWE-PolyBench: A Multi-Language Benchmark for Repository Level Evaluation of Coding Agents [2025-arXiv]
7. SWE-bench Goes Live! (SWE-bench-Live) [2025-NeurIPS Datasets and Benchmarks]
8. SWE-bench Multilingual [2025-Benchmark]
9. BeyondSWE: Can Current Code Agent Survive Beyond Single-Repo Bug Fixing? [2026-arXiv]
10. LoCoBench: A Benchmark for Long-Context Large Language Models in Complex Software Engineering [2025-arXiv]
11. CoderEval: A Benchmark of Pragmatic Code Generation with Generative Pre-trained Models [2024-ICSE]
12. DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories [2024-ACL Findings]
13. SWE-QA: Can Language Models Answer Repository-level Code Questions? [2026-ACL Findings]
14. LongCodeBench: Evaluating Coding LLMs at 1M Context Windows [2025-COLM]
15. CRUST-Bench: A Comprehensive Benchmark for C-to-safe-Rust Transpilation [2025-COLM]
16. PATCHEVAL: A New Benchmark for Evaluating LLMs on Patching Real-World Vulnerabilities [2025-arXiv]
17. SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks [2025-NeurIPS]
18. Vul4J: A Dataset of Reproducible Java Vulnerabilities Geared Towards the Study of Program Repair Techniques [2022-MSR]
19. CRAVE: Code Review Agent Verdict Evaluation [2025-Dataset]
20. Self-Evolving Multi-Agent Collaboration Networks for Software Development (rSDE-Bench) [2025-ICLR]
21. GameCraft-Bench: Can Agents Build Playable Games End-to-End in a Real Game Engine? [2026-arXiv]
22. ProgramBench: Can Language Models Rebuild Programs From Scratch? [2026-arXiv]
23. FrontierSWE [2026-Benchmark]
24. CompileBench: Can AI Compile 22-year-old Code? [2025-Benchmark]

3.2 General Coding Benchmarks

1. Evaluating Large Language Models Trained on Code (HumanEval) [2021-arXiv]
2. Program Synthesis with Large Language Models (MBPP) [2021-arXiv]
3. Measuring Coding Challenge Competence with APPS [2021-NeurIPS Datasets and Benchmarks]
4. Competition-Level Code Generation with AlphaCode (CodeContests) [2022-Science]
5. LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code [2025-ICLR]
6. BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions [2025-ICLR]
7. Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation (EvalPlus) [2023-NeurIPS]
8. MultiPL-E: A Scalable and Extensible Approach to Benchmarking Neural Code Generation [2023-IEEE TSE]
9. DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation [2023-ICML]
10. CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution [2024-ICML]
11. xCodeEval: A Large Scale Multilingual Multitask Benchmark for Code Understanding, Generation, Translation and Retrieval [2024-ACL]
12. EffiBench-X: A Multi-Language Benchmark for Measuring Efficiency of LLM-Generated Code [2025-NeurIPS Datasets and Benchmarks]
13. Aider Polyglot Benchmark [2024-Benchmark]
14. Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces [2026-ICLR]
15. Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem (Terminal-Bench Pro) [2025-arXiv]
16. TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks [2026-arXiv]
17. DebugBench: Evaluating Debugging Capability of Large Language Models [2024-ACL Findings]
18. Anchored Self-Play for Code Repair (BugSourceBench) [2026-ICML]
19. CodeIF-Bench: Evaluating Instruction-Following Capabilities of Large Language Models in Interactive Code Generation [2025-arXiv]
20. LongCoder: A Long-Range Pre-trained Language Model for Code Completion (LCC) [2023-ICML]
21. EvoR: Evolving Retrieval for Code Generation (EvoR-bench) [2024-EMNLP Findings]
22. Can LLM Already Serve as A Database Interface? A BIg Bench for Large-Scale Database Grounded Text-to-SQLs (BIRD) [2023-NeurIPS Datasets and Benchmarks]
23. Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows [2025-ICLR]
24. DA-Code: Agent Data Science Code Generation Benchmark for Large Language Models [2024-EMNLP]
25. InterCode: Standardizing and Benchmarking Interactive Coding with Execution Feedback [2023-NeurIPS Datasets and Benchmarks]
26. VerilogEval: Evaluating Large Language Models for Verilog Code Generation [2023-ICCAD]
27. RTLLM: An Open-Source Benchmark for Design RTL Generation with Large Language Model [2024-ASP-DAC]
28. Comprehensive Verilog Design Problems: A Next-Generation Benchmark Dataset for Evaluating Large Language Models and Agents on RTL Design and Verification (CVDP) [2025-arXiv]
29. KernelBench: Can LLMs Write Efficient GPU Kernels? [2025-ICML]
30. SOL-ExecBench: Speed-of-Light Benchmarking for Real-World GPU Kernels Against Hardware Limits [2026-arXiv]
31. LiveBench: A Challenging, Contamination-Limited LLM Benchmark (Coding Subset) [2025-ICLR]
32. SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks (Software Engineering Subset) [2026-arXiv]

3.3 Self-Evolution Benchmarks

1. HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness? [2026-arXiv]
2. EvoAgentBench: Benchmarking Agent Self-Evolution via Ability Transfer (Coding and Software Engineering Subsets) [2026-arXiv]
3. SkillEvolBench: Benchmarking the Evolution from Episodic Experience to Procedural Skills (Code Modification Subset) [2026-arXiv]
4. RSIBench-Data: Benchmarking Data-Centric Research for Recursive Self-Improvement [2026-arXiv]
5. AI4AI-Bench: Benchmarking LLM Agents in Algorithmic Design for Recursive Self-Improvement [2026-arXiv]
6. SWE-Bench-CL: Continual Learning for Coding Agents [2025-arXiv]

3.4 Empirical Studies

1. Rethinking the Evaluation of Harness Evolution for Agents [2026-arXiv]
2. Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents [2026-arXiv]
3. Phantom Guardrails: When Self-Improving Agent Harnesses Fix Failures That Never Happened [2026-arXiv]
4. Behind EvoMap: Characterizing a Self-Evolving Agent-to-Agent Collaboration Network [2026-arXiv]
5. Don't Blame the Large Language Model: How Agent Harness Evolution Shapes Coding Agent Quality [2026-arXiv]
6. The Scaffold Effect in Coding Agents: Harness Choice as a Hidden Variable in Coding-Agent Evaluation [2026-arXiv]
7. Prompt-Induced Waste in Coding Agents: Reasoning, Effort, Harness Design, and End-to-End Cost [2026-arXiv]
8. Harness Engineering: Anatomy, Architecture, and Evolution of Coding Agents -- A Source-Code Study of Eleven Systems [2026-arXiv]
9. Self-Authored Verification Is Unreliable in Heuristic Self-Improving Agents [2026-arXiv]
10. Memory Reward Inflation in Self-Improving LLM Agents [2026-arXiv]
11. Auditing Harness Tampering in Self-Improving Agents [2026-arXiv]
12. From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent Skills [2026-arXiv]
13. Memory Transfer Learning: How Memories are Transferred Across Domains in Coding Agents [2026-arXiv]
14. How Memory Management Impacts LLM Agents: An Empirical Study of Experience-Following Behavior [2026-ACL]

🧩 4. Self-Evolving Coding Products

These products and open-source tools support persistent memory, reusable skills, context updates, and harness customization through automatic or user-guided refinement. The table summarizes adaptation targets and mechanisms. Inclusion reflects documented capabilities, not necessarily a fully autonomous, experimentally validated self-evolution loop.

Target legend: 🔵 Assets — Memory, Skill, Tool, Context (including environment adaptation) · 🟣 Architecture — Harness, Workflow, Multi-Agent.

From memory to architecture: Most products adapt assets such as memory, skills, and context. Prime Agent and DeepSeek Harness also expose mechanisms for modifying the surrounding agent harness.

📚 5. Related Surveys

5.1 Surveys on Self-Evolving Agents

1. A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence [2026-TMLR]
2. A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems [2025-arXiv]
3. A Systematic Survey of Self-Evolving Agents: From Model-Centric to Environment-Driven Co-Evolution [2026-TechRxiv]
4. Self-Improvements in Modern Agentic Systems: A Survey [2026-arXiv]
5. Self-Improving Agents in the Era of Experience: A Survey of Self- to Meta-Evolution [2026-OpenReview]
6. A Survey on Self-Evolution of Large Language Models [2024-arXiv]
7. Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering [2026-arXiv]
8. Diving into Reliable Self-Evolving Agents: A Survey [2026-OpenReview]
9. The Path to Recursive Self-Improving Agents: Foundation, Framework, and Future Directions [2026-Preprints.org]
10. Self-Evolving Agents as Dynamic Graph Transformation: A Survey and New Perspective [2026-arXiv]

5.2 Surveys on Coding Agents

1. Large Language Model-Based Agents for Software Engineering: A Survey [2025-TOSEM]
2. Agents in Software Engineering: Survey, Landscape, and Vision [2025-Automated Software Engineering]
3. LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision, and the Road Ahead [2025-TOSEM]
4. Large Language Models for Software Engineering: A Systematic Literature Review [2024-TOSEM]
5. A Survey on Large Language Models for Code Generation [2026-TOSEM]
6. Advances and Frontiers of LLM-Based Issue Resolution in Software Engineering: A Comprehensive Survey [2026-arXiv]

Citation

@misc{zhou2026selfevolvingcodingagents,
      title={Self-Evolving Coding Agents}, 
      author={Hao Zhou and Haichuan Hu and Tianyu Luo and Ye Shang and Chunrong Fang and Zhenyu Chen and Liang Xiao and Quanjun Zhang},
      year={2026},
      eprint={2608.03392},
      archivePrefix={arXiv},
      primaryClass={cs.SE},
      url={https://arxiv.org/abs/2608.03392}, 
}

🤝 Contributions are welcome! If you find any missing or incorrect information, please feel free to open an issue or submit a pull request.