$ the-wire · showcase
OpenHands 1.17.0 adds cloud LLM connections, custom automation manifests
By RepoJournal · Filed · About Agentic Coding · Composed from the cited sources · methodology
OpenHands shipped 1.17.0 with LLM provider connections on cloud, task-outcome-aware automation runs, and Canvas support for the @openhands/extensions manifest contract, while cline fixed a set of desktop state bugs that silently dropped or reordered user prompts.
OpenHands cut v1.17.0, whose features include enabling LLM provider connections on cloud, making the automation run UI task-outcome aware, and showing manifest-declared value statements on dashboard cards and rows [1]. The Canvas work lands alongside the new `@openhands/extensions` automation setup contract, exercised against prompt, plugin, upload, event-trigger, and review manifest variants [2]. Separately, a confirmation prompt that could be rendered inside a collapsed EventGroup, hiding the verification question and Cancel/Continue buttons, is now kept visible outside grouped events [3].
On the cline desktop side, four state-machine fixes landed in the same area: a queued prompt's bubble was being dropped when the previous send resolved after it started, because `LocalRuntimeHost.runTurn` schedules the queue drain as a microtask before returning the result, so `chat_queued_prompt_start` reached the webview ahead of the prior turn's completion [4]. A new task now stays on "starting" while the hub reports the fresh session idle, instead of applying in-flight `session.created` and `session.updated` events that carry status "idle" [5]. When a send fails before the turn begins, for example switching to Codex and the OAuth refresh throws, the sidecar synthesizes a messages-less error result and the user turn was never appended, so post-send hydration wiped the optimistic bubble and the prompt vanished; `sendPrompt` now resolves false and hands the prompt back to the composer [6].
The Codex picker regression is fixed in the same batch: `toProviderConfig` was filling `knownModels` for `openai-codex` from the shared `openai-native` catalog, so `list_provider_models` overwrote the filtered Codex list with the full OpenAI API catalog (gpt-4o, gpt-4.1, chatgpt-image-latest) and the runtime handler lost the Codex context caps. `filterOpenAICodexModels` is now applied when building the catalog [7].
LangChain released `langchain-openai==1.6.2`, adding GPT-6 Astra reasoning efforts to the OpenAI integration and bumping httpx2 from 2.10.0 to 2.12.0 [8]. The monorepo's automated model-profile refresh added, removed, and changed entries across openrouter, including `inception/mercury-2.5` at 260,000 context and 65,536 output with reasoning and tools [9]. OpenHands also converted the automation SDK version cache from an effect to React Query while preserving `useTracking()`'s provider-free contract, after the first attempt accidentally required every consumer to sit below a `QueryClientProvider` [10].
Action items
- → Upgrade to OpenHands v1.17.0 to get the grouped-event confirmation prompt fix and cloud LLM provider connections OpenHands/OpenHands [plan]
- → Update cline desktop builds to pick up the queued-prompt, new-task status, and failed-send prompt preservation fixes cline/cline [plan]
- → Bump langchain-openai to 1.6.2 if you need GPT-6 Astra reasoning effort handling langchain-ai/langchain [monitor]
References
- [1] v1.17.0 ↗ OpenHands/OpenHands
- [2] feat: support generic custom automation manifests in Canvas ↗ OpenHands/OpenHands
- [3] fix: keep confirmation prompt visible outside grouped events ↗ OpenHands/OpenHands
- [4] fix(desktop): keep a queued prompt's bubble when the previous send resolves after it starts (#13979) ↗ cline/cline
- [5] fix(desktop): keep a new task on "starting" while the hub reports the fresh session idle (#13981) ↗ cline/cline
- [6] desktop: preserve the prompt when the provider connection fails (#14008) ↗ cline/cline
- [7] fix: Codex subscription model list and picker (#14011) ↗ cline/cline
- [8] langchain-openai==1.6.2 ↗ langchain-ai/langchain
- [9] chore(model-profiles): refresh model profile data (#40317) ↗ langchain-ai/langchain
- [10] refactor: use React Query for automation SDK version ↗ OpenHands/OpenHands