Three Claude Code subagent files, three models in frontmatter: what each launch cost, and the field the validator waved through

An investigation into Claude Code subagent configurations reveals how YAML frontmatter settings like 'model' and 'effort' impact performance and costs. By testing three distinct subagent definitions—a bulk reader, a deep reviewer, and a pre-ship reviewer—the author measured API usage and token consumption. The findings confirm that while frontmatter settings effectively override session defaults, the validation process is surprisingly permissive, silently ignoring unrecognized keys rather than flagging errors. The study highlights that subagent launch costs are significantly influenced by the harness's overhead, often exceeding the size of the system prompts themselves. The author concludes that developers should not rely solely on the built-in validator; instead, they recommend verifying subagent behavior through transcript analysis to ensure that critical settings like 'effort' are correctly applied, preventing silent fallbacks to less efficient default configurations.