Skill Adapter: Personalized Modification and Online Routing of General-Purpose Skills
Organizations: Nanjing University · Jiangnan University
Abstract
Large language model (LLM) agents use external skills to reuse task procedures and experience with tools. General-purpose skills describe shared execution procedures, while specific tool choices, operating practices, and output requirements vary across users. In long-horizon tasks, subgoals, available inputs, and environmental states change as execution progresses. Agents must select currently applicable skills from an expanding skill pool and prepare candidates for subsequent steps. We propose Skill Adapter, which enables personalized modification and online routing of general-purpose skills. The system distills local adjustments from user requirements and execution feedback and stores them as Adapters linked to general-purpose skills. The routing agent selects general-purpose skills and applicable Adapters based on the current subtask, execution progress, and environmental feedback, incrementally maintaining a task-level online skill library that the main LLM reads on demand. The system preloads candidates for subsequent nodes and rechecks their applicability once actual inputs are determined. Reinforcement learning optimizes the two types of decisions using modification gains and routing outcomes, respectively. We evaluate Skill Adapter on ALFWorld, WebShop, and SkillsBench. Compared with SkillRL, it improves task success rates by an average of 11.43 percentage points and reduces mean execution time on all three benchmarks. Skill Adapter retains its performance advantage when the skill pool grows to 1,000 skills.
Figures & tables
| ALFWorld | WebShop | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Method | Pick | Clean | Heat | Cool | Look | Pick 2 | SR | Time (s) | Score | SR | Time (s) |
| Backbone (ReAct) | 59.05 | 60.49 | 39.58 | 42.67 | 48.72 | 56.94 | 52.86 | 88.6 | 46.00 | 19.80 | 72.4 |
| Backbone (all skills) | 30.48 | 33.33 | 20.83 | 25.33 | 28.21 | 23.61 | 27.62 | 121.5 | 24.80 | 7.53 | 103.7 |
| Backbone + BM25 | 68.57 | 61.73 | 56.25 | 56.00 | 53.85 | 65.28 | 61.67 | 81.3 | 58.13 | 25.27 | 66.8 |
| LLM as Selector | 70.48 | 67.90 | 60.42 | 56.00 | 61.54 | 68.06 | 65.00 | 99.7 | 61.00 | 28.87 | 82.5 |
| SkillRouter | 75.24 | 71.60 | 66.67 | 64.00 | 58.97 | 73.61 | 69.76 | 76.8 | 68.87 | 46.13 | 62.4 |
| Method | SR | Mean reward | Mean time (s) |
|---|---|---|---|
| Backbone (ReAct) | 7.14 | 9.65 | 468.3 |
| Backbone (all skills) | 9.05 | 14.87 | 621.4 |
| Backbone + BM25 | 13.33 | 17.24 | 525.2 |
| LLM as Selector | 14.76 | 18.63 | 587.3 |
| SkillRouter | 17.62 | 19.46 | 615.0 |
| Graph of Skills | 18.57 | 23.09 | 462.9 |
| Group | Variant | ALF. | Web. | SB. | Avg. |
|---|---|---|---|---|---|
| Skills | Trajectory pool only | 52.86 | 25.93 | 9.05 | 29.28 |
| Skills | + General-purpose skill pool | 70.24 | 41.53 | 13.81 | 41.86 |
| Routing | + Composition and usage guidance | 76.90 | 51.47 | 16.67 | 48.35 |
| Routing | + DAG preloading | 80.71 | 57.07 | 20.48 | 52.75 |
| RL | Modification policy | 87.86 | 68.80 | 23.81 | 60.16 |
| RL | Routing policy | 90.95 | 73.93 | 30.95 | 65.28 |
LLM Agent Skill Retrieval
Large language model agent skill retrieval finds and selects reusable procedures to help agents complete tasks.
See all 108 papers