Unclear It is not clear what condition the new check tests, so it is not known which effort level changes get skipped.
What
effortLevel sets how much effort the model puts into its answers. Claude Code used to apply it every time it arrived. Now:
- Skipped when satisfied: a settings or control update that carries
effortLevelis applied only if a new check on the model, the level and the engine state says it is not already satisfied. Otherwise it is ignored entirely, including the report of the change. apply_flag_settings: the same skip applies when a controlling program, for example over Remote Control, changes session settings this way. Before, anyeffortLevelwas always validated and applied.ultracode: an effort change that arrives without an explicitultracodevalue now calls a new reset step.
Why
This avoids applying the same effort level twice, for example from Remote Control. It also means some effort changes may be dropped silently, so a setting that seems to have no effect may have been judged already in place.
Names in the bundleeffortLevel
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
Confirmed since
Anthropic's documentation has since written up effortLevel, on All settings.
* `effortLevel`: one of `"low"`, `"medium"`, `"high"`, or `"xhigh"`settings-reference see the edit
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear what condition the new check tests, so it is not known which effort level changes get skipped.
Anthropic's documentation agrees
Anthropic's documentation has since written up effortLevel, on All settings.