Vibe-Skills
A skill-routing and workflow-orchestration layer that helps an AI agent pick and sequence the right skills from a large installed pool, instead of invoking more of them.
Vibe-Skills is an orchestration layer that helps an AI agent choose and sequence Skills from a large installed pool, rather than simply calling more of them. It first clarifies the task's objective and delivery requirements, breaks a complex task into verifiable subtasks, and then selects only the capabilities that are actually relevant, executing them in dependency order.
It was benchmarked on SkillsBench, an evaluation of whether agents can use Skills to complete professional tasks, using a large-scale multi-skill setting where every task is evaluated with all 195 specialized Skills installed simultaneously rather than only the one skill meant for that task. Compared to a baseline agent without it, using DeepSeekV4Flash-VE and OpenHands, it increased the average task score and reduced both token consumption and tool calls.
It is written in Python and released under the Apache 2.0 license.