SDK v0.0.94

• Plugins now run only in worker threads (the default, CLINE_PLUGIN_MODE=worker) or in the subprocess sandbox (sandbox). The in_process mode is removed, along with its silent fallback. A host that ships no plugin-sandbox-bootstrap.js fails every plugin load with PluginBootstrapMissingError, which names CLINE_PLUGIN_BOOTSTRAP_PATH, instead of quietly running plugins inside the hub. The hub daemon's uncaught error handlers are fatal again. The SSH remote hub ships no bootstrap yet, so it reports plugin tools as missing
• Provider errors that arrive after a model stream has started are retried up to 3 times, and the final failure is reported as an error in logs and Langfuse traces. Exhausted retries no longer also trigger the agent's whole-turn retry, so retry budgets don't multiply
• Model request failure logs include the provider's request id
• setHubDaemonLauncher lets an embedding host start the hub daemon through a bundled Cline CLI, and ensureDetachedHubServer gains a beforeSpawn hook that runs only when a new daemon actually starts. The standalone SSH remote helper entrypoint is removed: SSH remotes run the hub through the uploaded Cline CLI
• NodeHubClient accepts resolveConnectionHeaders, so a remote proxy's WebSocket upgrade can be authenticated with HTTP headers refreshed on every reconnect. It is threaded through HubRuntimeHost and RemoteRuntimeHost
• AskQuestionExecutor may return text and image blocks, so attachments sent with an answer reach the model
• New buildTelemetrySelfcheckReport export in @cline/shared
• The cline provider's bundled default (the first Cline recommended model) is now Gemini 4 Argon instead of Claude Sonnet 5.5, because the recommended list now leads with it
• Refreshed the model catalog (6,862 → 6,952 models). Adds the Boundless and Standard Compute providers, and SambaNova now comes from the catalog with MiniMax M3 as its default. The Cline recommended list adds Gemini 4 Argon and drops Grok 4.7, and the Cline free list replaces Step 5 Preview with Nemotron 3.5 Super VL. OpenCode Go no longer lists MiniMax M2.7. Default models change for Cortecs, CrossModel, Eden AI, Google Gemini, GitHub Copilot, Google Vertex AI, LLM Gateway, Ofox, OpenCode Go, and Prime Inference (Gemini 4 Argon), Fireworks AI, Nebius Token Factory, and Requesty (Muse Spark Open), CoreWeave, DigitalOcean, NanoGPT, and Nvidia (Nemotron 3.5 Super VL), Aixy and Neon (Claude Haiku 5.5), Agnes AI, Ambient, IteraCompute, Poe, Scaleway, and SenseNova

Full Changelog: sdk/sdk/v0.0.93...sdk/sdk/v0.0.94