Back to work
03
Skill Passport
A research project on whether AI skills can move safely between models and agents.
Skill Passport explores a “passport” for AI skills: a description rich enough to predict whether a skill will be used correctly by a new model or agent, before anything is deployed.
The project also studies selection regret: the cost that comes from choosing the wrong skill for a task, and how better predictions can reduce it.
How it works
Describe
Capture what a skill does and what it expects.
Profile
Characterize the target model or agent.
Predict
Estimate compatibility without deploying.
Measure
Quantify selection regret from wrong choices.
Key highlights
No deployment needed
Compatibility is predicted up front.
Transferability
Asks how skills carry over between models.
Selection regret
Measures the cost of choosing the wrong skill.
