Skill++: Towards Lifelong Agents via Routing-Aware Skill Refinement
Hanrong Zhang, Philip S. Yu and colleagues at the University of Illinois Chicago, McGill, Rutgers, MBZUAI and Trip.com Group present Skill++, which keeps a growing skill library routable by rewriting only the name and description of each new skill.
Ask this paper
Problem. Under progressive disclosure, a router picks skills by name and description only. A new skill with similar metadata can displace an earlier one, breaking routing that used to work even though neither the earlier skill nor the router changed. The authors frame this as class-incremental continual learning.
Method. For each incoming skill, Skill++ rewrites the description to state what the body supports, how it differs from its closest existing skill, and what related work it excludes. It needs no router training and no labeled feedback.
Results. As a library grows to 731 skills over ten sessions, Skill++ selects all required skills for 76.3% of queries, against 36.4% with the original metadata. It nearly eliminates substitution errors and halves crowding errors.
Generality. It outperforms all compared approaches across three benchmarks and five LLM routers, and the gains carry over to the Codex CLI harness.
Abstract
Large language model (LLM) agents extend their capabilities with skills, each with a name, a description, and a body. Under progressive disclosure, an LLM router selects skills by name and description alone and loads only the selected bodies. Skill libraries grow continually as users and agents add new skills. Because the router relies on this metadata, a similar new skill can displace an earlier one, breaking previously correct routing even though the earlier skill and the router are unchanged. This routing interference is aggravated for public skills, whose descriptions may not match their bodies. We formulate the problem as lifelong skill routing, a class-incremental continual learning problem. A solution must keep earlier skills selectable without suppressing new ones. Existing approaches fall short, since continually retraining the router risks catastrophic forgetting and rewriting methods either ignore similar skills or require per-skill feedback. We therefore propose Skill++, which rewrites only the name and description of each incoming skill, without router training or labeled data. The new description states what the body supports, how the skill differs from its closest predecessor, and what related work it excludes. As a library grows to 731 skills over ten sessions, Skill++ selects all required skills for 76.3% of queries, versus 36.4% under the original metadata. Across three benchmarks and five LLM routers, it outperforms all compared approaches, and its gains carry over to the Codex CLI agent harness.