dsh-llm-volcengine-plans

Integrating Volcengine Ark Agent Plan and Coding Plan into DeepSeek Harness has historically meant relying on a legacy standalone package or hand-wiring OpenAI compatibility parameters, while DSH rc.6 introduced a regression where third-party Models editor registration breaks the Add provider dialog. dsh-llm-volcengine-plans bundles both plans into a single community DSH package so that route registration, gateway switches, and the credential pipeline all live under one maintainable manifest, eliminating route conflicts with the legacy bundle.

Key capabilities include registering two isolated route families (volcengine-plan/ and volcengine-coding/) with per-model context window, output limit, modality, and thinking-level metadata; automatically applying the OpenAI compatibility field switches the Coding Plan gateway expects; resolving ARK_API_KEY through the DSH credential service on every request with legacy environment-variable fallbacks; exposing a ready-to-use key form under Settings → Plugins → Plugin configuration that stores the secret in $DSH_HOME/.credentials.yaml without ever returning it to the browser; and supporting validated, live-reloadable configuration for routes, endpoints, headers, timeouts, retries, and default reasoning level so changes take effect without a restart.

This package is well suited for developers and teams managing multi-model workflows with DSH who need a stable, consolidated connection to Volcengine Ark services, particularly those using both Agent Plan and Coding Plan and wanting to retire the legacy bundle while keeping all configuration in one place.