[Feature]: Add `always_active` config option for ralph-loop

Prerequisites

• I will write this issue in English (see our Language Policy)
• I have searched existing issues and discussions to avoid duplicates
• This feature request is specific to oh-my-opencode (not OpenCode core)
• I have read the documentation

Problem Description

Currently, ralph-loop must be explicitly invoked with /ralph-loop "task" for each session. For users who want the "work until done" behavior as their default workflow, this requires:

1. Remembering to start every task with /ralph-loop
2. Re-invoking /ralph-loop after session restarts
3. Extra friction for teams who've adopted ralph-loop as their standard practice

Many power users (myself included) find that ralph-loop's iterative completion model is superior to the default "respond once and stop" behavior. We'd like to make it the default without modifying prompts.

Proposed Solution

Add an always_active configuration option to ralph_loop:

{
  "ralph_loop": {
    "enabled": true,
    "default_max_iterations": 100,
    "always_active": true  // NEW: Auto-activate ralph-loop for all prompts
  }
}

Behavior when always_active: true:

• Every user prompt automatically triggers ralph-loop behavior
• Agent continues working until <promise>DONE</promise> is detected
• Same iteration limits and completion detection as /ralph-loop
• Can still be cancelled with /cancel-ralph

Implementation Suggestion

Leverage the existing keyword-detector hook pattern:

// In keyword-detector or new UserPromptSubmit hook
if (config.ralph_loop.always_active) {
  // Inject ralph-loop mode into session state
  // Similar to how "ultrawork" mode is auto-activated
}

Alternatively, the Stop hook could check config.ralph_loop.always_active in addition to checking for the state file.

Alternatives Considered

Alternative	Why Not Ideal
Always prefix prompts with /ralph-loop	Tedious, error-prone, easy to forget
Modify system prompt to include ralph behavior	Doesn't leverage the Stop hook's continuation mechanism
Custom UserPromptSubmit hook	Would work but requires users to write code; should be built-in
Use /ulw-loop always	Same problem - still requires explicit invocation

Use Cases

1. Teams with "always iterate" policy - Organizations that want agents to work autonomously until completion
2. Complex projects - Where tasks naturally require multiple iterations
3. Background workflows - Where ralph-loop's continuation model is expected behavior

Related Issues

• #952 - ralph-loop not loading (existing bugs)
• #900 - ralph-loop argument parsing
• #1233 - Ralph Loop completion detection

Doctor Output (Optional)

N/A - this is a feature request

Additional Context

The keyword-detector hook already sets modes based on keywords (ultrawork, search, analyze). Adding ralph-loop as an "always on" mode would follow the same architectural pattern but triggered by config rather than keywords.

Feature Type

Configuration Enhancement

Contribution

• I'm willing to help with testing
• I can help with documentation