01 / The short version
What happened
Deep Agents has updated its skills support, an open standard for giving agents domain knowledge via markdown files. Publishers say tools can now be bound to skills, loading only when the skill is read, with mid-conversation tool addition supported on newer Anthropic and OpenAI models to preserve prompt caching. Pinned skills let apps inject instructions upfront for lower latency. Skills reload via metadata reset, and publishers note provider caches typically expire within minutes to an hour of inactivity. All changes are in the latest deepagents package.
See the exact references02 / Key takeaways
What you need to know
- 01
Tools bind to skills and load only when the skill is read
- 02
Pinned skills inject instructions before the next model call
- 03
Skills reload via skills_metadata reset or client commands
Keep in perspective
What to watch for
The source is publisher documentation; performance and caching claims are not independently validated.
Go to the source
Exact references
These are the original pages used for this brief. Publisher claims are not independent evaluations.
01Primary source · LangChainRead the original announcementhttps://www.langchain.com/blog/revamping-skills-in-deep-agentsAI-generated from the linked source. It can miss context; verify consequential details in the original. How the radar works