mirror of
https://github.com/deepseek-ai/deepseek-harness.git
synced 2026-09-14 04:01:35 +00:00
Tianyi Cui
a7a5be1703
Run the dsh-archive-agent-notes audit over every active Agent Note on current master, judging each record by whether its rationale still guides work rather than by size or age. - Archive 453 implemented bilingual triplets (417,882 English words): completed UI chrome, narrow adapters, closed bug fixes, implementation walkthroughs whose package READMEs, docs pages, generators, or successor notes now carry the useful behavior, and 51 records fully superseded by a later active note. Keep 201 implemented notes whose ownership rules, negative guarantees, durable or wire semantics, security rules, reintroduction conditions, or still-tempting rejected alternatives remain useful. - Reject 7 proposals whose premise is gone or whose work shipped in amended form under other records; delete 2 rejected notes that no longer prevent a plausible mistake. - Retarget every remaining inbound link to the archived path, and repair active prose that named an archived record as the owner of a live fact: parenthetical citations drop, ownership sentences redirect to the README, docs page, or active note that states the fact, and history citations say so. Chinese files link the English archived path because the pairing gate treats the frozen tree as outside the bilingual corpus. - Seal 1,359 new frozen artifacts; existing seals are unchanged and outbound links from archived notes are neither inspected nor repaired. - Regenerate docs/config-catalog.md after the hook-bridge comment edits shifted two source line numbers.
3.0 KiB
3.0 KiB
description, kind
| description | kind |
|---|---|
| LLM 能力包组：一个提供方无关的模型调用服务、DeepSeek 与 pi-ai 提供方适配器、请求重试执行器，以及具备回放感知的 token 计量。 | package-group |
llm/ — LLM 能力家族
English | 中文
概述
llm 组提供 harness 的模型调用能力：一个提供方无关的服务，任何组合都可以通过它向模型提供方发起流式请求，外加适配器、提供方专用请求元数据、重试执行与计量。核心 llm 包定义所有插件与会话日志使用的消息、内容块与流式分片词汇；提供方适配器把某个提供方的协议格式翻译为该词汇；DeepSeek 请求扩展插件在模型输入之外贡献具有生命周期归属的元数据；llm-retry 在持久 agent 步骤边界上重跑失败的请求；token-meter 从持久日志测量请求与上下文压力。本页是组的映射；每个包 README 负责各自的包级约定。
目录
包
| 包 | 职责 | ctx key |
|---|---|---|
llm/ |
通过已注册的提供方适配器流式发起一次模型调用，并共享 harness 的消息、块与分片词汇 | ctx.llm |
llm-deepseek/ |
以 DeepSeek chat-completions 直连、thinking 与图片输入服务 deepseek-official 路由 |
注册到 ctx.llm |
llm-pi-ai/ |
通过 pi-ai 目录与协议格式服务配置的提供方路由，包括手工声明的网关 | 注册到 ctx.llm |
deepseek-llm-api-extensions/ |
在官方 DeepSeek 请求上注册具有生命周期归属的顶层字段 | ctx.deepseekLlmApiExtensions |
plugin-package-inventory-deepseek/ |
为官方 DeepSeek 请求贡献活跃 Loader 包清单 | 贡献 dsh_plugin_packages |
llm-retry/ |
在持久 agent 步骤边界上按各提供方策略重试失败的模型请求 | 监听 agent/request-error |
token-meter/ |
用固定启发式规则从持久会话日志测量请求与上下文压力 | ctx.tokenMeter |
相关文档
- LLM 流式子系统——消息与块类型、组装后的模型请求、
StreamChunk协议与适配器约定。 - Token 计量子系统——
ctx.tokenMeter背后的测量语义。 - 孪生 LLM 适配器——为什么 DeepSeek 路由交付两个结构不同的适配器。
- 按路由的模型上下文——loop 如何路由模型请求并压缩上下文。
开发备注
无。