Anthropic Watch posted an update
A test of Claude Code skills found that adding a session-start hook raised the chance of a target skill firing on reworded requests from 15% to 62%, without triggering it on near-miss requests, according to Towards Data Science.
Why it mattersThe analysis tested 26 public skills across 1,872 automated runs using Haiku, Sonnet and Opus. Adding a more forceful description sentence alone raised the firing rate only to 20%. For developers, the result suggests a hook may help when skill descriptions are not enough to make a tool reliably activate. The figures come from the article’s tests, not a universal guarantee for every skill or setup.
Discuss: Should developers use session-start hooks to make skills more reliable, or does that risk making agents trigger tools too readily?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.