DeepSeek Harness 插件

dsh-ark-plan

Correctly activate DeepSeek v4 flash on the Volcano Ark plan API for DeepSeek Harness: default config injection (ark provider + reasoningEfforts + reasoningEffort=max) plus an ark_plan_doctor(英文原文)

跳到安装方式

来源信息

GitHub 仓库
snow-The/dsh-ark-plan
最近更新
2026年8月18日
分类
工具与能力
GitHub stars
0
载体类型
plugin
目录证据
上游声明已找到 dsh.bundle
证据路径
package.json#dsh.bundle
核对版本
0.1.0-rc.8
上游核对日期
2026-08-20

该证据由上游目录提供。本站没有安装、运行或安全审核这个插件。

安装

默认先复制一段 Prompt,让 Agent 读 GitHub 仓库和源码;需要自己装时再切到命令。

复制这段 Prompt,发给 DSH、Codex 或其他 Agent,让它先读 GitHub 仓库和源码。

请先不要安装或执行任何命令。阅读这个插件的 GitHub 仓库、README 和关键源码,然后用清楚、直接的方式回答以下问题,帮助我判断它是否适合我的需求:

1. 这个插件是什么,解决什么问题;
2. 适合哪些用户和典型使用场景;
3. 安装后如何使用,并给出一个最小使用示例;
4. 有哪些已知限制,以及隐私、安全、兼容性或维护风险;
5. 给出“推荐 / 有条件推荐 / 不推荐”的明确建议和理由。

请区分仓库明确说明、根据源码推断和未知信息。证据不足时请明确说明,不要猜测或照抄 README。

GitHub:https://github.com/snow-The/dsh-ark-plan
插件名:dsh-ark-plan
作者:snow-The

检查来源文件

安装前先看这个插件目录里的 README 和其他文件。

文件资源管理器3 个文件
README.md来源说明 · 只读预览

dsh-ark-plan

Activate DeepSeek v4 flash on the Volcano Ark plan API (https://ark.cn-beijing.volces.com/api/plan/v3) in DeepSeek Harness — the budget lane: 10× cheaper off-peak token bundles for the same model.

Strategy: built-in route + baseURL override (no custom provider)

The plugin reuses pi-ai's built-in deepseek provider — which already knows the models (reasoning incl. effort=max, 1M context, 384K output) — and only overrides baseURL to point at Ark's plan API. No hand-rolled "ark" provider, nothing to re-declare, nothing to silently degrade.

Two lanes coexist:

LaneProviderbaseURLKey envKey prefix
Official (default)deepseek-officialapi.deepseek.comDEEPSEEK_API_KEYsk-...
Ark budgetdeepseek (DeepSeek V4 Flash(ARK))ark plan APIARK_API_KEYark-...

Default model stays on the official route; the ark lane is there to burn the cheap plan tokens when you want them — switch the model picker to the deepseek provider's models to use it, then switch back when the bundle runs out (don't buy more: the plan's value ends there).

Install

dsh plugin --profile web add github:snow-The/dsh-ark-plan
# restart dsh web

Configure keys

# ~/.dsh/.credentials.yaml (or the Models page)
DEEPSEEK_API_KEY: sk-...      # official DeepSeek key (official lane)
ARK_API_KEY: ark-...          # Volcano Ark plan key (ark lane)

The two envs are deliberately separate — a sk- key against the ark baseURL answers 401 AuthenticationError and vice versa.

Verify

Ask your agent to run ark_plan_doctor. It reads the current config, flags the known mistakes, and live-tests ark /chat/completions with the deepseek thinking format (thinking.type=enabled + reasoning_effort=max) — expect serverModel = deepseek-v4-flash-ga-<date> (e.g. ...-ga-260731, the 0731 build) and thinking: true.

Traps this plugin encodes

| Trap | Symptom if missed | |---|---| | Stale standalone ark provider (v0.1 layout) | A duplicate "ark \| 自定义" entry in the Models page with a red dot. Remove llm-pi-ai.providers.ark. | | reasoningEffort: max not set | Thinking runs at the endpoint default (high), not max. | | Wrong key family for the baseURL | sk-... against ark → 401; ark-... against api.deepseek.com → 401. Match key to endpoint. | | ARK_API_KEY missing | The ark lane exists but answers 401 until its key is filled. | | baseURL not the ark plan API | Plugin defaults do nothing if deepseek.baseURL points at api.deepseek.com — that's just the official route. |

How the defaults work

dsh-settings registers each entry's patch config as the base of its settings section, so plugin defaults only apply where the user has not configured the same keys. No overwriting of user settings.

License

MIT