[FEATURE] Configurable pushback / proactive review mode

Resolved 💬 1 comment Opened May 29, 2026 by aleksandar-mojanchevski Closed Jul 2, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Currently Claude tends toward agreeableness — it executes what's asked without volunteering concerns unless directly prompted. This is fine for fast execution tasks but counterproductive
when a second opinion or code review mindset is what's actually needed.

Suggestion: Add a user-configurable setting (in settings.json or via /config) that controls how proactively Claude surfaces issues, bugs, and alternative approaches — independent of what
the user asked for. Something like a "challenge level" dial:

  • Execute — do what's asked, minimal unsolicited opinions
  • Review — flag issues spotted along the way, suggest alternatives
  • Challenge — actively push back, question assumptions, propose different approaches

The current default sits too close to "Execute" even when "Review" would produce better outcomes. A user who is debugging a production issue with a senior engineer mindset gets more value
from being told "this catch block silently swallows failures" than from a model that completes the task and stays quiet.

Why this matters: Optimizing for human approval ratings produces a model that's good at sounding helpful rather than being helpful. Giving users explicit control breaks that loop — the
user is opting in to friction, so agreement-seeking stops being the right target.

Proposed Solution

I have no idea

Alternative Solutions

_No response_

Priority

Low - Nice to have

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗