PASE
An agent skill evolution framework that decomposes a skill into an executable program module and a natural language description, and achieves targeted updating and iterative evolution of the skill through program patches.
Timeline (last 90 days)
The PASE framework represents agent skills as a separation between executable programs and natural language descriptions: the executable program carries the procedural logic, while the natural language retains descriptive knowledge.
Experiments on multiple models and multiple benchmarks show that PASE consistently outperforms all baselines, with gains exceeding 2.8%.
PASE performs better in analyses of evolutionary stability and generalization ability.
PASE's data and code will be released via GitHub.
The PASE paper emphasizes that programmatic knowledge plays a key role in skill evolution.
All Facts (5)
The PASE framework represents agent skills as a separation between executable programs and natural language descriptions: the executable program carries the procedural logic, while the natural language retains descriptive knowledge.
50%UnverifiedExperiments on multiple models and multiple benchmarks show that PASE consistently outperforms all baselines, with gains exceeding 2.8%.
50%UnverifiedPASE performs better in analyses of evolutionary stability and generalization ability.
50%UnverifiedThe PASE paper emphasizes that programmatic knowledge plays a key role in skill evolution.
50%UnverifiedPASE's data and code will be released via GitHub.
50%