Skip to content

deepagents@1.13.5

Choose a tag to compare

@github-actions github-actions released this 16 Sep 22:17
· 31 commits to main since this release
280f573

Patch Changes

  • #861 7f8d150 Thanks @thushanth-bengre-langchain! - feat(filesystem): replace read_file's line-number gutter with a status header

    Before:

         1	hello
         2	world
    

    After:

    @@ lines 1-2 of 2 @@
    hello
    world
    

    read_file now states its line range once, in a @@ lines A-B[ of T] | ... @@ header above unmodified source, instead of a per-line prefix. Truncation and a clamped negative offset are now disclosed as terse header fields instead of prose.

    The quickjs provider's stripLineNumbers is updated to strip the new header (with a fallback to the legacy format) so sandboxed code still gets clean file content.

  • #847 9be2581 Thanks @richardscarrott! - fix(deepagents): pin skills metadata on per-thread basis

    createSkillsMiddleware stored loaded skills in a loadedSkills closure, created once per middleware instance and reused for every invocation that instance serves. A single agent commonly serves many threads, so the first thread to load won for all of them: later threads were given the first thread's skills metadata and never saw their own. Skills are now loaded per invocation from graph state.