Decreased invocation rate observed with Claude Sonnet 4.5

[kvenkatrajan]

On analysis of the test results it was observed that Claude Sonnet 4.5 resulted in decreased skill invocation rate compared to Claude Opus 4.5/4.6

On further investigation it was found that the frontmatter needs to be updated to improve invocation reliably on Sonnet. Here's the analysis
https://gist.github.com/kvenkatrajan/52e6e77f5560ca30640490b4cc65d109

[kvenkatrajan]

#1041 - addresses it in Sensei

We need to run Sensei for the tools and rerun the tests to validate

[github-actions]

Thanks for opening this issue and sharing the detailed analysis, @kvenkatrajan! 🙏

This has been labeled as a bug since it describes a regression in skill invocation reliability, skills because it affects the skill frontmatter/invocation system, and intent-detection since the root cause relates to how Sonnet models detect and trigger skills from descriptions. The investigation gist is a great starting point — once the required frontmatter changes are identified, this should be actionable as a targeted code update.

Generated by Issue Triage for issue #1044