deepagents@1.8.0

Minor Changes

• #236 357a092 Thanks @christian-bromann! - feat(deepagents): add local shell backend

Patch Changes

• #230 a762b91 Thanks @alvedder! - chore(deepagents): re-export createSummarizationMiddleware

• #234 199c86c Thanks @christian-bromann! - fix(deepagents): update summarization behavior

• #223 bfa843d Thanks @maahir30! - fix(deepagents): prevent write_file crash when model omits content

  • Default the content parameter to an empty string so a missing argument doesn't crash the entire agent run via Zod validation failure.

@langchain/sandbox-standard-tests@0.1.0

Minor Changes

• #237 a827af7 Thanks @christian-bromann! - feat(sandbox-standard-tests): allow custom testrunner

@langchain/node-vfs@0.1.2

Patch Changes

• #237 a827af7 Thanks @christian-bromann! - fix(node-vfs): adding license file

@langchain/modal@0.1.3

Patch Changes

• #237 a827af7 Thanks @christian-bromann! - fix(modal): adding license file

@langchain/deno@0.2.1

Patch Changes

• #237 a827af7 Thanks @christian-bromann! - fix(deno): adding license file

@langchain/daytona@0.1.2

Patch Changes

• #237 a827af7 Thanks @christian-bromann! - fix(daytona): adding license file

deepagents@1.7.6

Patch Changes

• #218 ae70fa4 Thanks @christian-bromann! - fix(deepagents): pass on subagent name

• #222 163c135 Thanks @christian-bromann! - fix(deepagents): unwrap responseFormat strategy types so structuredResponse is correctly typed

deepagents@1.7.5

Patch Changes

• #211 adce96c Thanks @hntrl! - fix(skills): improve skills middleware input validation and add annotations

  Port of Python PR #1189. Hardens parseSkillMetadataFromContent with stricter
  coercion/trimming for all YAML fields, adds Unicode lowercase support in
  validateSkillName, validates and truncates compatibility length, handles
  allowed-tools as YAML list or space-delimited string, and shows
  license/compatibility annotations in the system prompt skill listing.

• #210 2eea576 Thanks @hntrl! - refactor(summarization): state rework, move to wrap pattern

  Refactors createSummarizationMiddleware to use the wrapModelCall hook instead of beforeModel. Instead of rewriting LangGraph state with RemoveMessage(REMOVE_ALL_MESSAGES) on each summarization, the middleware now tracks a SummarizationEvent in private state and reconstructs the effective message list on each call, avoiding full state rewrites. Supports chained summarizations with correct cutoff index progression.

deepagents@1.7.4

Patch Changes

• #208 4ea1858 Thanks @antonnak! - fix(skills): use systemMessage.concat() instead of systemPrompt string in SkillsMiddleware

  Aligns SkillsMiddleware.wrapModelCall with FilesystemMiddleware and SubAgentMiddleware
  by using request.systemMessage.concat() instead of request.systemPrompt string concatenation.
  This preserves SystemMessage content blocks including cache_control annotations for
  Anthropic prompt caching.

@langchain/deno@0.2.0

Minor Changes

• #216 786053f Thanks @christian-bromann! - feat(deno): better support for sandboxes with volume