Model-Triggered Skills

concept · updated Jul 15, 2026

person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation

A model-triggered skill is a capability or tool that an AI agent autonomously decides to invoke based on context and the skill's description, without explicit user instruction — contrasted with User-Invoked Skills, which are called explicitly by the user.

Mechanism and Distinction

Philipp Schmid (Google DeepMind) introduces model-triggered skills as one of two fundamental invocation patterns in skill-based agent architectures. In his framing, "based on the context and the description, the model decides to use or read a skill" — meaning the model itself performs the routing decision, inferring from conversational context and the skill's declared purpose whether to activate it. This stands in direct contrast to user-invoked skills, where the human explicitly requests a particular capability. 1

Points of Disagreement

No counterarguments or alternative framings of model-triggered skills appear in the current material.

Implications for Evaluation

The autonomous nature of model-triggered skills makes them a key motivation for rigorous evals before deployment. Because the model — not the user — controls invocation, unexpected or unintended skill activations are a realistic failure mode, underscoring Schmid's broader argument that skills should not be shipped without adequate evaluation coverage. 1

Footnotes

  1. Philipp Schmid — Don't Ship Skills Without Evals, 52:45 2