[Feature Request] Unified Auto Mode as a first-class feature

What feature would you like to see?

Currently, Kimi CLI offers several flags and options that enable "automatic" behavior, but they are scattered and not unified under a clear "Auto Mode" concept:

• --yolo / --auto-approve: auto-approves all tool calls
• --print: non-interactive mode (implicitly enables --yolo)
• --max-ralph-iterations: auto-loop mode for iterative tasks
• Agent tool with run_in_background=true: delegates tasks to sub-agents

As a user, I would like a unified, first-class "Auto Mode" that:

1. Has a clear entry point — e.g., a dedicated --auto flag, /auto slash command, or default_auto_mode = true in config, rather than needing to remember multiple flags.

2. Is well-documented as a feature — currently users have to piece together --yolo, --print, and Ralph loop to achieve a fully autonomous agent experience. An official "Auto Mode" page in the docs would help discovery.

3. Provides sensible defaults — auto-approves safe operations (reads, searches), while still confirming destructive operations (writes outside project, rm, git push, etc.), or at least makes this configurable (related to #1631).

4. Works seamlessly with Plan Mode — when entering plan mode, auto-mode should still allow the user to review the plan before execution begins (related to #1867).

Why is this needed?

• New users often ask "does Kimi CLI have an auto mode?" because competitors (Claude Code, Cursor Agent, etc.) advertise this as a headline feature.
• The current --yolo naming is playful but not discoverable for users looking for "auto" or "agent" mode.
• Having a unified mode reduces cognitive load: one flag/command instead of combining --yolo --print --max-ralph-iterations 10.

Proposed interface

# Enable auto mode for this session
kimi --auto

# Or via config
[mode]
default_auto = true

# Or via slash command inside session
/auto

In Auto Mode:

• Tool calls are auto-approved by default (with granular rules as proposed in #1631)
• The session runs in non-interactive style when piped/scripted
• Iterative tasks (Ralph loop) are enabled with a sensible default limit
• The user can still intervene with Ctrl+C or pause with /pause

Related issues

• #1631 (granular auto-approval rules)
• #2072 (yolo vs non-interactive confusion)
• #1867 (yolo in plan mode)
• #1966 (AskUserQuestion in yolo mode)

Start by reviewing the existing --yolo, --auto-approve, --print, and --max-ralph-iterations behavior, then read related issues #1631, #2072, #1867, and #1966. The work is done when the project has an agreed unified entry point, documented defaults and controls, compatible Plan Mode behavior, and coverage for the proposed interactions.