Feature request: configurable tone/personality setting

Request

Add a configurable setting (not just prompt instructions) to control Claude Code's conversational tone — specifically the ability to disable unsolicited editorializing, interpretive framing, and confidence signaling. Too much chirpy chatbot.

Problem

CLAUDE.md instructions can nudge tone but don't reliably constrain it. There's no hard toggle equivalent to disabling a feature. Users who want facts-only output without commentary, "what this means for you" framing, or chirpy chatbot affect currently have no reliable mechanism to get it.

Suggested approach

A tone or personality setting in settings.json (e.g., "tone": "terse" or "editorial": false) that is enforced at a level deeper than prompt instructions — similar to how other behavioral flags work in the harness.

Context

Raised by a power user who found that even with detailed CLAUDE.md directives, Claude Code defaults to confident editorializing when reporting research findings, including reframing the user's own analysis back to them as if it were novel insight.