AI Insight
SkillFlow is a new framework for coordinating AI agents to complete complex tasks using large language models. It employs a flow-based training method called Tempered Trajectory Balance that maintains diverse problem-solving strategies instead of converging to a single approach, while also providing transparent credit assignment for each decision step. The system includes a recursive skill evolution mechanism that automatically determines when to create or remove skills based on training signals, demonstrating superior performance across 14 datasets including question answering, mathematical reasoning, code generation, and interactive decision-making tasks.
Why it matters
This work addresses fundamental challenges in AI agent orchestration that could improve automated systems' ability to handle complex, multi-step tasks across various domains. The framework's ability to maintain strategic diversity and autonomously evolve its capabilities could lead to more robust and adaptable AI assistants for real-world applications.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
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://github.com/beita6969/SkillFlow.
Source: SkillFlow: Flow-Driven Recursive Skill Evolution for Agentic Orchestration