LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents
For LLM agent systems, this provides a more efficient and modular way to inject reusable skills without exposing them as plaintext.
LatentSkill converts textual skills into LoRA adapters via a hypernetwork, storing skill knowledge in weight space to reduce context overhead. On ALFWorld, it improves success by 21.4/13.4 points with 64.1% fewer prefill tokens; on Search-QA, it improves exact match by 3.0 points with 72.2% lower overhead.
Agent systems increasingly use textual skills to encode reusable task procedures, but injecting these skills into the prompt at every step incurs substantial context overhead and exposes skill content as plaintext. We present LatentSkill, a framework that converts textual skills into plug-and-play LoRA adapters through a pretrained hypernetwork. LatentSkill stores skill knowledge in weight space rather than context space, removing per-step skill tokens while preserving modular loading, scaling, and composition. On ALFWorld and Search-QA, LatentSkill outperforms the corresponding in-context skill baseline while using substantially fewer prefill tokens: it improves ALFWorld success by 21.4 and 13.4 points on the seen and unseen splits with 64.1% fewer prefill tokens, and improves Search-QA exact match by 3.0 points with 72.2% lower skill-token overhead. Further analysis shows that generated skill LoRAs form a structured semantic geometry, can be precisely controlled via the LoRA scaling coefficient, and can be composed through parameter-space arithmetic when skill components are aligned. These findings suggest that weight-space skills provide an efficient, modular, and less exposed substrate for extending LLM agents.