Microsoft’s SkillOpt Shows Optimized Agent Skill Artifacts Transfer Across Model Scales and Between Codex and Claude Code Harnesses
Microsoft's SkillOpt shows agent skills trained on one model transfer to another—lifting Claude performance 60 points on spreadsheets with a Codex-trained artifact.
Why it matters
Agent skill portability across model architectures and harnesses is a genuine technical finding that changes how teams think about skill reuse and agent engineering—but transfer quality is highly task-dependent, requiring close evaluation per workload.
The key facts
5 to knowSkillOpt exports best_skill.md artifacts that work across untrained environments
Codex-trained spreadsheet skill lifted Claude from 22.1 to 81.8 (vs 80.4 in-harness)
Transfer retention varies sharply by task: 102% on spreadsheets, 10% on math
Key finding in Section 4.3, beyond the headline 52/52 result
Published August 2026 via MarkTechPost
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Most coverage of Microsoft's SkillOpt centers on its 52/52 result. The more consequential finding is in Section 4.3: the exported best_skill.md keeps working in environments it was never trained on. A Codex-trained SpreadsheetBench skill lifted Claude Code from 22.1 to 81.8, slightly above the 80.4…