Explore adaptive effort/thinking level for Claude Code sessions

[tfvchow]

Description

Explore whether Claude Code supports or could benefit from adaptive effort levels rather than a static CLAUDE_CODE_EFFORT_LEVEL: "high" setting.

Currently the --effort flag only accepts low/medium/high. Investigate:

• Whether there's a dynamic/adaptive mode (or upcoming feature)
• Whether effort level could be adjusted based on task complexity (e.g., low for simple reads, high for /spec workflows)
• Token cost implications of always running at "high"

Context

Discovered during v7.2.1 migration triage. Upstream added CLAUDE_CODE_EFFORT_LEVEL: "high" to settings.json. Fork adopted it but wants to understand if a smarter approach exists.

Priority

Exploration only — current "high" setting works fine.