deepagents@1.14.0

Minor Changes

• #862 6e6b036 Thanks @richardscarrott! - feat(deepagents): reload skills when skillsMetadata is set to null

  Skills are loaded once per thread and kept in state, so a long-lived thread never saw skills added, edited or deleted after its first run. Setting skillsMetadata to null now makes the next model call re-read every source and replace the stored list. A thread whose sources hold no skills also counts as loaded, so it stops listing the backend on every run — including threads checkpointed by earlier versions, which store [] and need an explicit null to reload.

  Loading moved from beforeAgent to beforeModel, so a reload is served by the next model call rather than the next run. Two consequences:

  • A middleware of your own can invalidate mid-run, from afterModel, and the following model call sees the fresh list.
  • With a StateBackend, a load now sees files written during the run, so a SKILL.md the agent just wrote is picked up.

  A fork inherits a pending invalidation: a fork spawned in the same iteration that set skillsMetadata to null reloads from its own sources rather than inheriting the parent's last loaded list.

  skillsMetadata is also part of the agent's state type now when skills is passed to createDeepAgent, which previously meant mounting createSkillsMiddleware by hand to reach it. Inference mirrors the runtime condition, so an omitted option or an empty literal contributes no state, while a non-literal string[] does.

  A middleware of your own can reach the field too: skillsMetadataValue is exported for declaring skillsMetadata on a custom middleware's state schema, since a middleware can only read and write the fields its own schema declares. Treat the value as opaque — it exists to be passed to StateSchema, and its concrete type may change.