@zhongjingwei/dsh-model-retry
DSH rc.8 Web 设置中的活跃提供商模型请求重试预算行
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:502399493zjw-lgtm/dsh-model-retry#86f13c789b6c09dc38fefd5e60ad5e6851d30161Model Retry 以 @zhongjingwei/dsh-model-retry 发布，当前版本为 0.1.0-rc.9。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
English | 中文
DeepSeek Harness Web plugin for configuring active providers' model-request retry budget from Settings → General.
Compatibility: pinned to stock DSH
0.1.0-rc.8. rc.8 moved retry ownership into provider profiles, so this version writes each active normal provider'sretryPolicy.maxRetriesthrough the published settings API.
The demo uses stock DSH 0.1.0-rc.8 and a plugin tarball packed from this source, all under an isolated DSH_HOME. It changes the retry budget from 2 to 5, then uses DSH's deterministic local LLM mock to return retryable server errors on attempts 1–5 and succeed on attempt 6. Session-title generation is disabled so those six mock requests belong only to the demonstrated model request. DSH executes and reports the retries; the plugin owns the shared retry-budget setting.
For stock rc.8:
dsh plugin --profile web add @zhongjingwei/dsh-model-retry@0.1.0-rc.9
Restart Web after installation. Remove the plugin with:
dsh plugin --profile web remove @zhongjingwei/dsh-model-retry
0.llm.providers and uses revision-fenced settings.mutate calls.retryPolicy.maxRetries for normal/default policies while preserving existing backoff and retryable-code fields.always policies unlimited and leaves retry execution/backoff behavior owned by DSH.pnpm install
pnpm test
pnpm run build
pnpm run verify:package
pnpm pack
The package is an external Cordis plugin and does not patch DSH core.
MIT