All projects
Open source tool for coding agents
skill-trigger-bench
Find out why your Agent Skill never triggers.
Nothing tells you when an Agent Skill was skipped. stb lint checks every installed SKILL.md offline: descriptions that say what a skill does but never when to use it, a combined description set over the roughly 15k character budget so some skills are silently never shown, two descriptions similar enough to fight over the same prompts, duplicate names and broken references. A small live eval then runs prompts through the agent and counts how often it actually reaches for your skill.
- Role
- Solo. I wrote it and maintain it.
- Period
- 2026
- Status
- Installable from GitHub, MIT licensed
- Built with
- Python · TF-IDF · Agent Skills
- Offline
- lint of the whole installed skill set
- ~15k
- character description budget it checks against