SKILL-DISCO: Distilling and Compiling Agent Traces into Reusable Procedural Skills

2026-06-25 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, medium

Summary

SKILL-DISCO is a novel distillation-and-compilation framework designed to extract reusable procedural skills from successful agent execution traces. It tackles the inefficiency of agents repeatedly solving similar task instances, which leads to increased reasoning costs and extended execution times. The framework conceptualizes successful traces within FSM-defined scenarios as paths in an unknown transition graph, defining procedural skills as parameterized control-flow subgraphs that can be reused. SKILL-DISCO distills these reusable PFSM subgraphs and compiles them into callable, executable, and verifiable procedural skills. Evaluations on ALFWorld and WebArena benchmarks demonstrate that SKILL-DISCO significantly enhances success rates and reduces the number of agent turns across different model scales, highlighting the value of representing shared experience as reusable execution structures.

Key takeaway

For AI Scientists and Machine Learning Engineers developing autonomous agents, if you are struggling with high reasoning costs and long execution traces in repetitive tasks, you should explore frameworks like SKILL-DISCO. This approach allows your agents to distill and compile reusable procedural skills from successful past experiences, significantly improving success rates and reducing agent turns. Implementing structured skill representations can make your agent systems more efficient and robust.

Key insights

Distilling reusable procedural skills from agent traces significantly reduces redundant reasoning and execution costs.

Principles

• Procedural skills are reusable parameterized control-flow subgraphs.
• FSM-defined scenarios are key for extracting procedural skills.
• Shared experience as execution structures boosts agent efficiency.

Method

SKILL-DISCO distills reusable PFSM subgraphs from successful traces and compiles them into callable, executable, and verifiable procedural skills.

In practice

• Apply to FSM-defined task scenarios for skill extraction.
• Improve agent success rates on ALFWorld and WebArena.
• Reduce agent turns in complex, repetitive tasks.

Topics

• Agent Systems
• Procedural Skills
• Skill Distillation
• Control-Flow Graphs
• ALFWorld
• WebArena

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• AIP: A Graph Representation for Learning and Governing Agent Skills — Structuring agent skills as a directed execution graph significantly enhances reliability and simplifies development compared to free-form prose.
• Learn to equip AI agents with reusable skills — Agent skills provide on-demand, specialized knowledge to AI agents through standardized, reusable task-specific modules.
• SkillWiki: A Living Knowledge Infrastructure for Agent Skills — SkillWiki provides a unified infrastructure for agent skill management, from ingestion to evolution, linking skills to evidence.
• Agent Skills: The Complete Workflow (Build →Test→Benchmark→Iterate) — Agent skills are modular, reusable instruction packages for repeatable workflows, requiring rigorous testing.
• Agent Skills Work but the Research Shows Most Teams Are Building Them Wrong — Curated, focused agent skills improve AI performance, but require structured organization and security diligence to scale effectively.
• Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward — Agent skills modularize LLM capabilities, extending functionality dynamically while introducing significant security challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.