Skip to content

AUTOREPLY_CONFIG.md

Latest commit

 

History

History
209 lines (165 loc) · 5.99 KB

AUTOREPLY_CONFIG.md

File metadata and controls

209 lines (165 loc) · 5.99 KB
Raw

Auto-Reply Configuration Guide

The auto-reply feature in this fork has been significantly enhanced to use opencode's built-in subagent system with OpenAI integration, making it much more intelligent and integrated!

Quick Start

Enable Auto-Reply with Subagent (Recommended)

# Enable with default subagent configuration
auto-reply on

# Enable with custom subagent settings
auto-reply on \
  --subagent-provider openai \
  --subagent-model gpt-4o-mini \
  --subagent-prompt "You are an AI assistant. Provide concise, helpful responses to continue conversations naturally." \
  --subagent-temperature 0.7 \
  --subagent-tokens 150

Traditional Phrase-Based Auto-Reply

# Enable phrase-based auto-reply
auto-reply on \
  --phrases "continue,go on,proceed,do it" \
  --triggers "next steps,what next,continue with"

# Or use hooks with external APIs
auto-reply on \
  --use-hooks \
  --primary-hook openai-assistant \
  --fallback-phrases true

Configuration Options

Subagent-Based Auto-Reply (Default & Recommended)

The subagent approach uses opencode's built-in AI client to spawn intelligent auto-reply agents, providing much better responses than static phrases.

auto-reply on \
  --use-subagent true \                    # Enable subagent replies (default: true)
  --subagent-provider openai \              # AI provider: openai, anthropic, local, etc.
  --subagent-model gpt-4o-mini \           # Model to use
  --subagent-prompt "Custom system prompt" \
  --subagent-temperature 0.7 \            # Response randomness (0-1)
  --subagent-tokens 150                    # Max response length

Default Configuration:

  • Provider: openai
  • Model: gpt-4o-mini
  • Temperature: 0.7
  • Max tokens: 150
  • System prompt: "You are an auto-reply assistant. Provide brief, helpful responses to continue conversations naturally."

Phrase-Based Auto-Reply (Simple & Fast)

auto-reply on \
  --phrases "continue,go on,proceed,do it,carry on,next step" \
  --triggers "next steps,what next,continue with,proceed with,moving forward" \
  --response-delay 1000 \                  # 1 second delay
  --cooldown 30s                          # 30 second cooldown period

Hook-Based Auto-Reply (External APIs)

auto-reply on \
  --use-hooks true \
  --primary-hook ai-assistant \
  --fallback-phrases true

Advanced Configuration Examples

1. Production AI-Powered Auto-Reply

auto-reply on \
  --use-subagent true \
  --subagent-provider openai \
  --subagent-model gpt-4o \
  --subagent-prompt "You are an expert assistant helping with software development. Provide concise, technically accurate responses that continue the conversation naturally." \
  --subagent-temperature 0.3 \
  --subagent-tokens 200 \
  --triggers "continue,next step,proceed,what next,how to continue" \
  --cooldown 1m

2. Development Team Auto-Reply

auto-reply on \
  --use-subagent true \
  --subagent-provider openai \
  --subagent-model gpt-4o-mini \
  --subagent-prompt "You are a helpful development assistant. Analyze code and conversation context to provide intelligent continuation responses." \
  --use-hooks true \
  --primary-hook code-review-assistant \
  --fallback-phrases true \
  --phrases "continue,go on,proceed,do it"

3. Minimal Configuration

# Just enable with defaults
auto-reply on

# Disable completely
auto-reply off

# Check current status
auto-reply status

Configuration Priority (High to Low)

  1. Subagent responses (highest priority, enabled by default)
  2. Hook responses (if subagent fails and hooks enabled)
  3. Phrase responses (fallback method)

How Subagent Auto-Reply Works

When enabled, the system:

  1. Detects triggers in user messages using configured trigger phrases
  2. Spawns a subagent with the specified provider and model
  3. Analyzes context including conversation history and original message
  4. Generates intelligent response using the AI model
  5. Updates statistics tracking response types and performance
  6. Respects cooldowns to prevent excessive responses

Statistics and Monitoring

Check auto-reply performance with:

auto-reply status

Statistics shown:

  • Total/Recent responses
  • Trigger count
  • Subagent vs Hook vs Phrase response counts
  • Last response time
  • Current configuration

Troubleshooting

Common Issues

  1. Subagent not working:

    • Check if AI provider credentials are configured
    • Verify model availability in your opencode config
    • Try simpler configuration first

  2. Too many responses:

    • Increase cooldown period: --cooldown 5m
    • Reduce trigger phrases
    • Disable subagent: --use-subagent false

  3. Responses too generic:

    • Customize system prompt with --subagent-prompt
    • Adjust temperature: --subagent-temperature 0.5

Reset Configuration

# Reset to defaults
auto-reply off --reset

# Disable and clear stats
auto-reply off --reset-stats

Integration Examples

With Loop Scheduling

# Enable intelligent auto-reply for scheduled tasks
auto-reply on \
  --use-subagent true \
  --subagent-provider openai \
  --subagent-model gpt-4o-mini

# Schedule tasks with auto-reply context
/loop 5m "analyze recent commits and provide status update"

With Pattern Detection

# Prevent AI loops with pattern detection
pattern-detection --enable --threshold 0.8 --repetitions 3

# Enable auto-reply with intelligent responses
auto-reply on \
  --use-subagent true \
  --subagent-provider openai \
  --subagent-model gpt-4o-mini

Performance Considerations

  • Subagent responses are slower but more intelligent (~1-3 seconds)
  • Phrase responses are instant but generic
  • Hook responses vary by external API speed
  • Cooldown periods prevent excessive API usage
  • Temperature settings affect response quality vs randomness

The subagent approach is recommended for most use cases as it provides the best balance of intelligence and integration with opencode's existing architecture!