Settings that move the bill
Five lines in settings.json that decide more than any prompt.
3 min
The five
model: the default variant. Prefer the 200k-context form; opt into 1M per session.
effortLevel: keep the global value at medium and raise it per session (/effort) for hard problems. Effort controls how much thinking and how many tool calls a turn spends.
env.CLAUDE_CODE_SUBAGENT_MODEL: the model every subagent uses. Point it at the current Sonnet, or remove it to inherit the session model.
env.ANTHROPIC_DEFAULT_*_MODEL: tier aliases. A stale value here picks an older model on every tier switch; delete them unless you need a specific pin.
enabledPlugins: each plugin's skill descriptions and hooks join every request. Disable the ones you do not use.
Apply them safely
readyclaw headroom lists the changes it would make and why; readyclaw headroom --apply asks per change and writes a timestamped backup first. Restart Claude Code afterwards.
npx readyclaw scan