User-Invoked Skills
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
User-Invoked Skills are a category of agent capability that is explicitly triggered by the user rather than autonomously decided upon by the model itself, forming one half of a two-part taxonomy alongside Model-Triggered Skills.
Definition and Contrast with Model-Triggered Skills
Philipp Schmid (Google DeepMind) distinguishes user-invoked skills from model-triggered skills as two fundamental modes by which agent capabilities are activated. Where a model-triggered skill is initiated by the model's own reasoning and decision-making, a user-invoked skill is explicitly called upon by the user. 52:45
Underappreciated Power
Schmid argues that user-invoked skills are significantly underestimated in their potential impact: "I think people underestimate how powerful user-invoked skills are." 53:54 The implication is that practitioners and researchers tend to focus attention on model-triggered autonomy while overlooking the leverage available when users directly invoke specific capabilities.
Relevance to Evals
The broader context of Schmid's talk — Don't Ship Skills Without Evals — places user-invoked skills within a quality-assurance framing: regardless of whether a skill is user-invoked or model-triggered, it should not be deployed without appropriate evaluation. The distinction between the two skill types matters for eval design, since user-invoked skills have a more constrained and predictable activation pattern than model-triggered ones, potentially affecting how their behavior is tested and measured.