The Wire · Showcase
LANGCHAIN CORES SHIP, ANTHROPIC THINKING BLOCKS FIXED, OPENHANDS TIGHTENS WEBSOCKET AUTH
By RepoJournal · Filed · About AI Agents
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
13 people shipped this
github-actions[bot]
@github-actions[bot]
2 cited
saoudrizwan
@saoudrizwan
2 cited
Functionhx
@Functionhx
1 cited
ccurme
@ccurme
1 cited
simonrosenberg
@simonrosenberg
1 cited
hieptl
@hieptl
1 cited
malhotra5
@malhotra5
1 cited
logan-markewich
@logan-markewich
1 cited
VirajAgarwal-couchbase
@VirajAgarwal-couchbase
1 cited
langchain-oss-model-profiles[bot]
@langchain-oss-model-profiles[bot]
Vasco Schiavo
@VascoSch92
dominiccooney
@dominiccooney
abeatrix
@abeatrix
LangChain released two core packages with critical fixes for streaming stability and environment variable handling, while OpenHands closed a security gap that left WebSocket connections unprotected.
LangChain pushed langchain-core 1.5.2 and langchain-anthropic 1.5.3 in rapid succession [1] [2], both addressing production friction. The anthropic release fixes a 400-error plague: streamed signature_delta chunks from empty adaptive-thinking blocks were silently dropping the required thinking key, causing replay failures on agent handoffs [3]. Core 1.5.2 patches environment variable handling for gateway URLs, closing a silent routing bug [4]. Over on OpenHands, a critical security fix [5] tightens WebSocket authentication outside URL-protected routes, plugging a gap that could expose agent connections to sidecar attacks. The team also shipped persistent agent memory toggles [6] and improved Cloud backend tracking with session deduplication [7]. Cline fixed a silent regional endpoint bug affecting Qwen, Moonshot, and Z AI users [8], where China-line toggles persisted but never actually routed requests to the correct hosts. A separate Cline fix restores conversation history during mid-turn mode switches [9], preventing the agent from forgetting context when toggling between Plan and Act modes. LlamaIndex improved OpenTelemetry span ID handling [10] and added telemetry to the Couchbase integration [11].
One email a day. Unsubscribe in one click.
Keep up with AI Agents in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
One email a day. Unsubscribe in one click. Read a past issue →
Action items
- → Upgrade to langchain-anthropic 1.5.3 if using streaming with adaptive thinking langchain-ai/langchain [immediate]
- → Update langchain-core to 1.5.2 to fix gateway env var routing langchain-ai/langchain [immediate]
- → Deploy OpenHands WebSocket auth fix before next release OpenHands/OpenHands [immediate]
- → Verify Cline endpoint toggles point to correct regional hosts post-patch cline/cline [plan]
References
- [1] langchain-core==1.5.2 ↗ langchain-ai/langchain
- [2] langchain-anthropic==1.5.3 ↗ langchain-ai/langchain
- [3] fix(anthropic): preserve empty thinking field in signature_delta streaming ↗ langchain-ai/langchain
- [4] fix(core): handle empty string in gateway env vars (#39107) langchain-ai/langchain
- [5] fix(security): authenticate WebSockets outside URLs (#16095) OpenHands/OpenHands
- [6] feat: add persistent agent memory toggle ↗ OpenHands/OpenHands
- [7] fix: Track cookie-auth Cloud backend connections ↗ OpenHands/OpenHands
- [8] Fix ignored China/international endpoint toggles for Qwen, Moonshot, Z AI (ENG-2340) ↗ cline/cline
- [9] fix(core): expose live in-memory session messages for mode-switch rebuilds ↗ cline/cline
- [10] fix: improve otel span ID handling ↗ run-llama/llama_index
- [11] Add telemetry ↗ run-llama/llama_index
