The Empire, Long Divided, Must Unite: Architectural Convergence in Three LLM Agent Harnesses
- ID: eda36985
- 原文链接: https://arxiv.org/abs/2608.23953
- PDF: https://arxiv.org/pdf/2608.23953v1
- 作者: Dai Jiahong
- 日期: 2026-08-25
- 更新: 2026-08-25
- 分类: agents
- 来源类型: paper
- 标签: agents, agent-harness, architecture, source-analysis, arxiv
- 质量评分: 5/5
中文导读
单作者源码级多案例研究：把三个刻意对立哲学的开源 coding agent harness——LangChain deepagents（全家桶）、Earendil pi（激进极简）、DeepSeek dsh（一切皆插件）——钉在固定 commit 上逐 commit 对照。两个成熟 harness 朝相反方向演化（deepagents 在删自研脚手架，pi 在攒基础设施），却收敛到同一中间形态，要素有五：商品化的请求循环、只追加且可重放的会话记录、以数据形式保存的模型怪癖、上下文渐进披露、显式扩展缝；第三个 harness 事后作为 held-out 检查，五条全中，其中一处直接复用别家实现，作者因此把收敛拆成平行发现、扩散、照抄三种来源而非宣称独立发明。唯一既未收敛也未出现的承重维度是外部可验证性——外部第三方无需信任运行时就能查验的防篡改记录，作者将其视为 provenance 敏感领域 harness 的下一个分化轴。
为什么值得关注
三个对立哲学的 agent harness 源码级对照收敛到五要素中间形态；外部可验证性是集体缺位的下一个分化轴
English Abstract
An agent harness is what turns a language model into an autonomous agent: the surrounding code that builds the model's context, mediates its tools, runs the loop, and persists state across a long-horizon run. This layer, not the model it wraps, is increasingly the binding constraint on agent behaviour. We present a source-level, multi-case study of three open coding-agent harnesses built from deliberately opposing philosophies: LangChain's deepagents (batteries-included), Earendil's pi (radical minimalism), and DeepSeek's dsh (everything-is-a-plugin). Reading each at a pinned commit and following its commit history, we find that the two mature harnesses have travelled in opposite directions (deepagents subtracting authored scaffolding, pi accreting durable infrastructure), yet converged toward one architectural middle form of five recurring elements: a commoditised loop, an append-only replayable session record, model quirks kept as data, progressive disclosure of context, and explicit extension seams. A third harness, read afterward as a held-out check, exhibits all five, and in one seam reuses another's implementation outright. We therefore do not claim independent invention, and decompose the convergence into parallel discovery, diffusion, and literal reuse. Finally, one load-bearing dimension shows no convergence, and indeed no presence: external verifiability, a tamper-evident record an outside party can check without trusting the runtime. We read this absence not as an oversight but as a predictive gap, the next axis on which harnesses for provenance-sensitive domains will differ.
Obsidian Notes
- Metadata and abstract fetched via
opencli arxiv paper 2608.23953 -f json(2026-08-27); response parsed list-or-dict tolerant. - 单一作者跨三仓库的源码级对照，方法与结论对自家 harness 评审有直接迁移价值。
- 中文导读与价值判断锚定在论文摘要上，未补充摘要之外的实验细节。