AgentSkillsCN

ralph-commands

拉尔夫·维格姆自主循环的用户控制指令

SKILL.md
--- frontmatter
name: ralph-commands
description: User control commands for Ralph Wiggum autonomous loops
license: MIT
compatibility: opencode
metadata:
  version: "1.0"
  audience: ralph-users
  stack: opencode-ralph-commands

Ralph User Control Commands

This skill provides user intervention commands for managing Ralph Wiggum autonomous loops during execution.

Available Commands

/ralph-status

Display current Ralph loop progress, task status, and budget information.

Usage:

code
/ralph-status

Output:

code
🚀 Ralph Loop Status
==================
Session: session_001
State: running
Current Task: oauth_implementation (iteration 2)

📊 Task Progress:
• oauth_implementation: in_progress (1/3 success criteria)
• stream_display: pending (dependency: oauth_implementation)
• track_metadata: pending (dependency: stream_display)

💰 Budget Status:
Total: $15.00 allocated, $0.00 spent
OAuth Implementation: $4.50 allocated, $0.00 spent

⏱️  Runtime: 00:15:32

/ralph-pause

Temporarily halt the autonomous loop for user intervention.

Usage:

code
/ralph-pause

Effect:

  • Stops current sub-agent execution
  • Prevents new sub-agents from spawning
  • Preserves all current state
  • Allows user to examine and modify code

/ralph-resume

Continue the autonomous loop from where it was paused.

Usage:

code
/ralph-resume

Effect:

  • Resumes autonomous loop execution
  • Continues from paused state
  • Maintains all context and progress

/ralph-hint "specific guidance"

Add targeted feedback for the next iteration of a specific task.

Usage:

code
/ralph-hint "Focus on implementing PKCE code challenge generation using crypto/sha256"

Effect:

  • Adds specific guidance to iteration feedback
  • Provided to next sub-agent execution
  • Escalates hint level for that task

/ralph-cost-limit [amount]

Update the total cost limit for the Ralph loop.

Usage:

code
/ralph-cost-limit 20.00

Effect:

  • Updates total budget allocation
  • Validates against current spend
  • Prevents overspending

/ralph-task-status [task_name]

Show detailed status for a specific task.

Usage:

code
/ralph-task-status oauth_implementation

Output:

code
📋 Task: oauth_implementation
Status: in_progress
Iterations: 2/8 allocated
Last Verification: Failed (25% score)

Failed Checks:
- oauth_url_missing
- pkce_missing

Next Actions:
- Add Soundcloud OAuth URL generation
- Implement PKCE code challenge

User Hints:
- "Use crypto/rand for secure random generation"

/ralph-complete [task_name]

Manually mark a task as complete (override verification).

Usage:

code
/ralph-complete oauth_implementation

Effect:

  • Bypasses verification for that task
  • Marks task as completed
  • Allows dependent tasks to proceed
  • Records manual completion in audit log

/ralph-reset [task_name]

Reset a task back to pending state.

Usage:

code
/ralph-reset oauth_implementation

Effect:

  • Clears all iteration history
  • Resets progress to 0
  • Useful for starting over on problematic tasks

/ralph-rollback [checkpoint]

Roll back to a specific git checkpoint.

Usage:

code
/ralph-rollback
# or
/ralph-rollback abc1234

Effect:

  • Restores git state to checkpoint
  • Recovers from broken implementations
  • Preserves Ralph progress tracking

Integration with Ralph Orchestrator

These commands are designed to be used with the Ralph Orchestrator agent. When a user types a command, the orchestrator:

  1. Parses the command and extracts parameters
  2. Updates shared state using the context manager
  3. Modifies loop behavior based on the command
  4. Provides feedback to the user about the action taken

Examples

Example 1: Checking Status During Development

code
User: /ralph-status

Ralph: 🚀 Ralph Loop Status
==================
Session: session_001  
State: running
Current Task: oauth_implementation (iteration 3)

📊 Task Progress:
• oauth_implementation: in_progress (2/3 success criteria)
• stream_display: pending  
• track_metadata: pending

💰 Budget Status:
Total: $15.00 allocated, $2.25 spent

Example 2: Providing Targeted Hint

code
User: /ralph-hint "Make sure to use the existing PocketBase migration patterns in pb_migrations/"

Ralph: 💡 Hint added for oauth_implementation
📝 Will include this guidance in next sub-agent iteration
🔄 Hint level escalated to 1

Example 3: Emergency Pause

code
User: /ralph-pause

Ralph: ⏸️ Ralph loop paused
🛑 Current sub-agent will complete current operation
📋 No new sub-agents will be spawned
💬 Use /ralph-resume to continue

Safety Features

Confirmation for Destructive Actions

Commands like /ralph-complete and /ralph-rollback require confirmation:

code
Ralph: ⚠️ This will mark oauth_implementation as complete without verification.
Are you sure? [y/N]

Audit Logging

All user interventions are logged:

json
{
  "timestamp": "2025-01-25T16:30:00Z",
  "command": "ralph-hint",
  "parameters": "\"Focus on PKCE implementation\"",
  "user": "session_001"
}

Budget Protection

Cost limit changes are validated against current spend:

code
Ralph: ❌ Cannot reduce cost limit below current spend ($2.25)
Minimum allowed: $2.25

These commands give users fine-grained control over the autonomous development process while maintaining the benefits of the Ralph Wiggum methodology.