Intel / article

Grounded Skill Synthesis from Code at Scale for Agentic Intelligence

Source: arxiv.org
Author: Yongqi Tong, Pan Wang, Hang Wang, Jianshe Li, Xin Zhang, Jiang-Ming Yang, Wei Wu
Date: Published Sep 9, 2026

Context

Code2Skill extracts reusable procedures from source code so an can consult them when tackling a new task. It turns selected code into records describing operations, workflows and their limits, then checks whether those descriptions preserve the implementation's behavior.

The paper describes a round trip: a model reconstructs code from the record without seeing the original function body, and a model-based judge compares the reconstruction with the source. Accepted records retain links to their evidence. Adding retrieved skills improved results in 57 of 72 matched evaluations. This supports code repositories as a source of useful procedural guidance, while the model-based checking remains a consistency test rather than a proof of correctness.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.

Key quotes

“Source code offers a complementary path: it requires no prior agent experience yet provides executable evidence for grounding abstractions.”

“Skills synthesized from tested AI-generated code achieve a 93.50% pass rate, compared with 93.00% for human-written code, providing initial evidence that the pipeline can expand with the growing volume of AI-generated software.”

Read the source arxiv.org