You'll noticeTier: how much it should matter to you
3Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
SkillsArea: what it touches
ImprovementsKind: in v2.1.287,
ImprovementsSection of the release
What
Skills are saved instructions Claude can run on request. Claude Code can now attach a note when your message contains /<skill>, the name of a skill. The note tells the model that the message mentions the skill, and to use the Skill tool only if you are asking for the skill to be run. This sits alongside the notes for mentioning agents and other participants.
Claude Code also gains a stricter way of sending requests through its relay, used in one place depending on a setting.
Why
Typing a skill's name in passing should not by itself make Claude run that skill.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtIt is not clear whether either change is switched on, or which setting chooses the stricter relay.
Anthropic's release notes agreeFixed --output-format stream-json and the SDK not streaming the turns of a context: fork skill run by typing /<skill> as the prompt, as they…
