Memory control gets finer-grained. Quiet but important: tool-runner now forwards container IDs properly.
Anthropic TypeScript SDK v0.117.0: Enhanced memory control and streaming fixes
Original: anthropics/anthropic-sdk-typescript: sdk: v0.117.0
Importance: SDK マイナーバージョンアップで機能追加と複数バグ修正を含むが、既存実装への破壊的変更はなく、利用者への即座の影響は中程度。
Summary
Anthropic released TypeScript SDK v0.117.0 with enhanced memory control and multiple bug fixes. Key additions include output_behavior parameter for dream creation to control whether new memory stores are created or updated in-place. Fixes address streamed message accumulation, per-request timeout handling, tool-runner response container forwarding, and path resolution alignment across SDKs. Build improvements include dotfile inclusion during git installs, and infrastructure was modernized by switching from yarn to pnpm.
Key Points
- output_behavior parameter added to dream creation for memory management control
- Streaming message_delta field accumulation logic fully corrected
- Per-request timeout settings now properly respected
- Tool-runner supports response container ID forwarding
- Development infrastructure migrated from yarn to pnpm
View developer notes (APIs, breaking changes, migration)
v0.117.0 adds output_behavior parameter to dream API for memory store management control. Fixes include complete message_delta field application during streaming (issue #289), proper per-request timeout honor in non-streaming long-request checks, response container id forwarding in tool-runner (#271), and path resolution/skill-archive/heartbeat alignment with other SDKs (#264). Build now includes dotfiles during git installs, yarn replaced with pnpm, and breaking-change detection runs on every push via CI.
Source: https://github.com/anthropics/anthropic-sdk-typescript/releases/tag/sdk-v0.117.0
Outlet: SDK (anthropic-sdk-typescript)
This article is an AI-generated summary (OpenAI GPT-4o-mini) of publicly available information from Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Sakana, and other vendors. The original source URL is always provided in accordance with fair-use citation requirements. Summaries are AI-generated and may contain mistranslations or misinterpretations. Always verify details with the original source.