Parametric Skills

Since intelligence fundamentally relies on efficient skill acquisition (Chollet, 2019), the ability to acquire and exploit reusable skills is critical for LLM agents. Such skills, whether manually authored or extracted from task trajectories, are textual recipes that encode mature problem-solving experience. Despite their widespread use, their utility is limited by the model's ability to comprehend and follow skill instructions, especially in complex, long-context scenarios where key instructions are difficult to locate and apply. To address this limitation, we propose ParametricSkills, a framework that converts free-form textual skills into parameters at test time, enabling context-free skill exploitation. Specifically, we first construct a large-scale, high-quality skill library, and synthesize single-turn and multi-turn skill exploitation trajectories built around these skills with OpenCode. Using these data, we then train a hypernetwork that parameterizes both the skill content and the test-time exploitation methodology by receiving textual skills and converting them into LoRA adapters. On six complex software engineering (SWE) subtasks, ParametricSkills outperforms in-context learning by an average of 6.44 LLM-judge points while improving BERT Score by 1.17 points and F1 by 5.53 percentage points. Beyond performance, we find that parametric skills, being inherently cumulative, offer a preliminary yet promising avenue toward test-time continual learning.