github.com/OpenHands/software-agent-sdk/pull/6030 Upgrade LiteLLM to 1.105.1 (supports System One Decision models) In service of github.com/OpenHands/software-agent-sdk/issues/6015

Allegedly prompts from CC. This… is great, they have multiple little agents running alongside the main agent, and enriching the user experience

Cap litellm below 1.106 until the SDK can use mcp 2.x github.com/OpenHands/software-agent-sdk/pull/6089 See [Bug]: Unlocked installs resolve litellm 1.106.0, which cannot import against mcp 1.x #6088 for details.

Fixing a workflow in the CI that was consistently failing: ci: fix shell quoting in Debian snapshot defaults step

I will do a SDK release today. If you have something you would like to merge, please let me know 🙂 (in the 🧵 )

Release 1.57 is out 🙂

github.com/OpenHands/software-agent-sdk/issues/4260 - is this issue still relevant? can we close it?

cc @curved-angle-80449 Seems it was a close call? 👀 😳

I will need a review of this PR if someone is around 🙂 github.com/OpenHands/software-agent-sdk/pull/6140

I investigated the reported Qwen3.8-27B/vLLM issue with Agent Canvas. Using the supplied image and flags, direct tool calls worked. With OpenHands SDK 1.58.1, the default

reasoning_effort="high"

failed because the checkpoint accepts low/medium/xhigh. Setting it to medium made the agent loop work with streaming on and off. vLLM’s Rust frontend obscures the cause with `raise_exception is unknown`; its Python frontend reports the unsupported reasoning-effort value directly. Do we want to address this through custom-endpoint defaults, model capability validation, or configuration guidance? I have reproduction scripts and traces. I wasn't sure if there is a way to keep this from happening again or if this is just too narrow of an edge case.

Tiny PR: github.com/OpenHands/software-agent-sdk/pull/6257 The SDK allows you to specify a max budget for a conversation - but this was never exposed by the agent server - until now! This would allow development of a UI where we say - "Start a conversation for this, but don't spend more than $5" The reason I want it is it will give me more confidence when developing automations that I'm not going to blow through a pile of cash accidentally 🙏

Hey, I will cut a release of the sdk with some important fixes. If you want to have a PR in this release, please put the PR link in the 🧵

Hi @modern-fireman-99755, here is my PR #6260 which adds the

has_utf8_locale()

guard for the SDK path, and enforces UTF8 locale for all variants of LC. (follow-up to #6255 by JunweiH0). It includes regression tests for non-UTF-8 locales. Would love to get it into the release if possible! Link: github.com/OpenHands/software-agent-sdk/pull/6260. The relevant issue is here: github.com/OpenHands/OpenHands/issues/18982 with a local repo copy here: (Not sure if I mistakenly duplicated it: github.com/OpenHands/software-agent-sdk/issues/6261).

RSI achieved externally!

PRs needed for sdk release 🙂 • chore(sdk): remove ACP curated model lists past their 1.59.0 deadline • fix(typescript): add integration_refs to agent profile types

High priority issues to fix — 2026-10-28 1. OpenHands/OpenHands#18102 — Virtualize the conversation message list for very long conversations [high] Why: PR #17820 open, mergeable, 13 checks green, awaiting review from @enyst/@VascoSch92; unassigned. 2. OpenHands/software-agent-sdk#6284 — v1.59.0 release blocked by expired ACP curated model lists [high] Why: blocks the v1.59.0 release; PR #6286 open (26 green checks), awaiting review; unassigned. 3. OpenHands/software-agent-sdk#6285 — TS agent profile types miss integration_refs vs the 1.59.0 contract [high] Why: fails the release PR's TS client CI; PR #6287 open, awaiting review from @simonrosenberg; unassigned. 4. OpenHands/OpenHands#16727 — Conversation corrupted after VM restart: "Missing event file" [high] Why: data loss, conversations unloadable; PR #16806 open, mergeable, awaiting review; unassigned. 5. OpenHands/OpenHands#16492 — Backend API keys in origin-readable localStorage [high] Why: security — any same-origin script can read every backend key; no PR in flight, unassigned. Generated by OpenHands AI on behalf of the user.

High priority issues to fix — 2026-10-28 1. OpenHands/software-agent-sdk#6224 — Long-generation timeouts wrongly retried as transient [high] Why: PR #6230 open, mergeable, checks green — waiting on maintainer review. unassigned. 2. OpenHands/software-agent-sdk#4963 — Windows PowerShell terminal poisoned after a terminating error [high] Why: PR #4962 open, mergeable, checks green — waiting on review. unassigned. 3. OpenHands/software-agent-sdk#5307 — Docker worktree built across incompatible host/container paths [high] Why: PR #5318 open, mergeable, checks green; review requested from @simonrosenberg. unassigned. 4. OpenHands/software-agent-sdk#5181 — Output-cap defaults and reasoning overrides mishandled [high] Why: PR #6040 open by @brash-coat-26221, mergeable, checks green. unassigned. 5. OpenHands/software-agent-sdk#5152 — Browser tools silently vanish on unlocked installs [high] Why: PR #5219 open but has conflicts and changes requested by @brash-coat-26221; review requested from @simonrosenberg. unassigned. Generated by OpenHands AI on behalf of the user.

Sdk version

1.59.0

is out

High priority issues to fix — 2026-10-29 1. OpenHands/software-agent-sdk#6224 — Deterministic long-generation timeouts retried as transient (27-min hangs) [high] Why: PR #6230 open, mergeable, checks green, awaiting maintainer review; unassigned 2. OpenHands/software-agent-sdk#4963 — Windows PowerShell terminal poisoned after a terminating error [high] Why: PR #4962 open, mergeable, checks green, awaiting maintainer review; unassigned 3. OpenHands/software-agent-sdk#4558 — Encrypted LLM profiles decrypted only on the conversation path [high] Why: security-related; fallback/sub-agents load keys without the cipher; PR #4413 open but CI failing; unassigned 4. OpenHands/software-agent-sdk#6088 — Unlocked installs resolve litellm 1.106.0, which cannot import against mcp 1.x [high] Why: breaks every tool-calling request; PR #18687 open but CI failing; unassigned 5. OpenHands/software-agent-sdk#6285 — TypeScript agent profile types miss integration_refs [high] Why: fix already merged (PR #6287); confirm the contract in a release; unassigned Generated by OpenHands AI on behalf of the user.

Hey 👋 I did some brief research into the bugs we've been seeing in the SDK and why we keep introducing them (see attachment). Having bugs is normal given our complex codebase and the challenges of working with APIs. I wanted to understand what we can do to prevent them, or at least reduce their frequency. There are a couple of actionable takeaways we can address quickly. To get started, I’ve opened a few small PRs. (others are more complex and big changes). I’d like to revisit this in a couple of weeks to see if we've managed to reduce the incoming bug rate. 🙂 To review: • ci: run downstream test suites when upstream packages change • test(cross): check LocalConversation and RemoteConversation parity • fix(sdk): align RemoteConversation errors with BaseConversation • test(sdk): drive pause tests with TestLLM instead of patching litellm • docs(agents): require a failure-path test for each changed behavior

Somewhat major architectural proposal of making the think tool optional and off by default https://github.com/OpenHands/software-agent-sdk/pull/6159 The motivation behind this is I noticed that deepseek flash was going to think to a very frequently and often overthinking when it used it. I ran a small swebench eval and it indeed reduced cost by 10% will not hurting accuracy.

I think we could auto-merge the automated PRs that clean up the

.pr/

directory from main. 🤔 It’s a type of PRs that only deletes files, nothing else must happen in its diff WDYT? An Automation, e.g. daily or on PR opened, could do it. Example: https://github.com/OpenHands/OpenHands/pull/19050

github.com/OpenHands/software-agent-sdk/pull/6030 Upgrade LiteLLM to 1.105.1 -- completed integration testing and addressed prior review comments -- stepping stone to decision model support in OpenHands SDK.

github.com/OpenHands/software-agent-sdk/pull/6068 feat(llm): add DecisionModel with OTEL tracing

🤖 LLM model list update — 2026-10-31 New models detected, PR opened: • https://github.com/OpenHands/software-agent-sdk/pull/6350 — Added gpt-6.1-terra (OpenAI), kimi-k3.1-pro (Moonshot), nemotron-3.5-super-vl-120b-a12b (Nvidia), qwen3.8-omni-flash (Qwen) Generated by OpenHands AI on behalf of the user.

High priority issues to fix — 2026-11-02 1. software-agent-sdk#6246 — ACPAgent.ask_agent (/btw) fork keeps all tools, never closed [high] Why: PR #6361 open, mergeable, one check failing. unassigned. 2. software-agent-sdk#6370 — agent-server fail-open on corrupt settings.json (confirmation mode silently off) [high] Why: security, fail-open; no PR yet. unassigned. 3. software-agent-sdk#6181 — TypeScript downloads return stale cached bytes [medium] Why: PR #6182 clean and mergeable, awaiting review from @calm-artist-7206. unassigned. 4. software-agent-sdk#5722 — Kill stdio MCP subprocesses on sync_close [medium] Why: PR #4862 open, awaiting review from @calm-artist-7206; assigned to @brash-coat-26221. 5. software-agent-sdk#5152 — browser-use disappears on unlocked installs [high] Why: PR #5219 has changes requested by @brash-coat-26221. unassigned. Generated by OpenHands AI on behalf of the user.

I would be grateful for feedback on #6383: Detect and recover from prefix-cache collapse on hosted/self-served LLM backends (per-call telemetry + backoff policy)

Starting an SDK release.