Hierarchical Self-Improvement: A Framework for Task-Specific Evolvable Agent Harnesses
- 类型：arxiv
- 标识：2608.08466
- 链接：https://arxiv.org/abs/2608.08466
- 主分类：evaluation
- 形态：application
- TLDR：Modern LLM agents are often improved by modifying prompts, tools, or workflows manually, while the executable scaffold surrounding the model---the harness---is typically treated as a fixed artifact after deployment. This work studies an alternative where the harness is task-specific and continuously evolvable: each task family maintains its own harness, which is hot-swapped across iterations through a fixed task-injection seam and rewritten using environment feedback. We introduce Hierarchical Self-Improvement (HSI), a framework in which a single frozen LLM M operates across three hierarchical
- 副分类：agent
- 待LLM分类：否
- 标题中文：分层自改进：面向任务特定可进化 Agent Harness 的框架
- TLDR中文：现代 LLM Agent 的改进通常依赖于人工修改 prompt、工具或工作流，而围绕模型的可执行支架——harness——在部署后一般被视为固定不变的产物。本文研究一种替代方案：harness 是任务特定的且持续可进化的，每个任务族维护各自的 harness，通过固定的任务注入接缝在不同迭代间热替换，并依据环境反馈进行改写。我们提出分层自改进（HSI），在该框架中，一个冻结的 LLM M 在三层
- 来源文件：
- /inbox/tom/_candidates/2026-08-22-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-08-23-agent-rag-longcontext-candidates.json