Skills on the Fly: Test-Time Adaptive Skill Synthesis for LLM Agents
What this work does
Retrieves training trajectories, synthesizes a task-specific skill for a fixed-parameter solver, and optimizes benchmark-specific synthesis prompt slots from rollout feedback.
Connection to discovery
An important harness-adaptation baseline that can improve use of past experience without updating the solver's weights.
Evidence
Inspected paired feedback/atomic prompt edits and ablations of retrieval, skill synthesis and per-dataset optimization.
Scope and limitations
Test-time adaptation in this system is not solver-weight training. Benchmark-specific prompt optimization and retrieval pools require strict split and budget controls.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.