본문 바로가기
  • Home

A Comparative Analysis of Skill Design Types in OpenClaw-Based Korean Office Task Automation

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(6), pp.161~167
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : April 13, 2026
  • Accepted : May 21, 2026
  • Published : June 30, 2026

Youngmi Baek 1 Jung Kyu Park 1

1대진대학교

Accredited

ABSTRACT

This paper examines how different skill design types affect task performance in a Korean office-task automation environment based on OpenClaw. We compare four SKILL.md design types: procedural, constraint, example, and checklist. Seven benchmark tasks are constructed, including reply drafting, identification of information to confirm, checklist generation, and structured summarization. The outputs are evaluated in terms of format compliance, inclusion of key information, handling of uncertainty, and avoidance of unsupported details. The results show that OpenClaw performs stably and accurately across most tasks. Differences among the four skill design types are generally small. Performance drops appear mainly in tasks involving scheduling uncertainty. Overall, task structure and uncertainty have a greater influence on performance than the instruction organization strategy of the skill template.

Citation status

* References for papers published after 2024 are currently being built.