SkillFlow: Flow-Driven Recursive Skill Evolution for Agentic Orchestration

TL;DR

SkillFlow introduces a flow-based framework for autonomous skill evolution in agentic systems, improving task orchestration by addressing strategy collapse, credit assignment, and skill development challenges.

Contribution

It presents a novel flow-matching loss and recursive skill evolution mechanism enabling principled, autonomous skill growth in LLM-based agentic systems.

Findings

Outperforms baselines on 14 datasets across various tasks.

Uses flow diagnostics for transparent credit assignment.

Enables autonomous skill creation and pruning.

Abstract

In recent years, a variety of powerful LLM-based agentic systems have been applied to automate complex tasks through task orchestration. However, existing orchestration methods still face key challenges, including strategy collapse under reward maximization, high gradient variance with opaque credit assignment, and unguided skill evolution whose decisions are typically made by directly prompting an LLM to judge rather than derived from principled training signals. To address these challenges, we propose SkillFlow, a flow-based framework that takes a trainable Supervisor as the agent and a structured environment with dynamic skill library and frozen executor, automating task orchestration through multi-turn interaction. SkillFlow employs Tempered Trajectory Balance (TTB), a regression-based flow-matching loss that samples trajectories proportional to reward, preserving diverse orchestration strategies rather than collapsing to a single mode. The same flow objective yields a jointly learned backward policy that provides transparent per-step credit assignment at zero additional inference cost. Building on these flow diagnostics, a recursive skill evolution mechanism determines when to evolve, what skills to create or prune, and where decision gaps lie -- closing the loop from training signal to autonomous capability growth. Experimental results on 14 datasets show that SkillFlow significantly outperforms baselines across question answering, mathematical reasoning, code generation, and real-world interactive decision making tasks. Our code is available at https://anonymous.4open.science/r/SkillFlow-E850.