55aafbae9a
- Modified loadChatHistory() to check for active project before fetching all sessions - When active project exists, use project.sessions instead of fetching from API - Added detailed console logging to debug session filtering - This prevents ALL sessions from appearing in every project's sidebar Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
9.9 KiB
9.9 KiB
RalphLoop "Tackle Until Solved" Integration Summary
Date: 2025-01-22 Status: ✅ Complete
Overview
Ralph Orchestrator has been successfully integrated into the Claude Code CLI with the name "RalphLoop: Tackle Until Solved". The integration provides autonomous agent loop functionality that continues iterating until a task is complete.
What Was Built
1. RalphLoop Wrapper Script
Location: /home/uroma/obsidian-web-interface/bin/ralphloop
A Python wrapper script that:
- Integrates Ralph Orchestrator with Claude Code CLI
- Creates Ralph projects (PROMPT.md, ralph.yml)
- Runs autonomous agent loops until completion
- Supports task descriptions, file input, and configuration options
Usage:
./ralphloop "Your task description here"
./ralphloop -i task.md
./ralphloop --agent claude --max-iterations 50
Environment Variables:
ANTHROPIC_API_KEY- Required for Claude agentRALPH_AGENT- Override agent selection (claude, gemini, etc.)RALPH_MAX_ITERATIONS- Max iterations (default: 100)RALPH_MAX_RUNTIME- Max runtime in seconds (default: 14400)
2. Brainstorming Skill Integration
Location: /home/uroma/.claude/skills/brainstorming/
Files:
SKILL.md- Updated with Ralph integration instructionsralph-integration.py- Python module for complexity detection and Ralph delegation
Features:
- Automatic complexity analysis (threshold: 5 steps)
- Keyword detection for architecture/system-level tasks
- Delegates complex tasks to Ralph for autonomous iteration
- Presents Ralph's validated design as final output
Complexity Triggers:
- Architecture keywords: architecture, platform, framework, multi-tenant, distributed
- Multiple implementation phases
- Keywords: complex, complete, production, end-to-end
- User opt-in via environment variables
3. Agent Pipeline Builder Integration
Location: /home/uroma/.claude/skills/agent-pipeline-builder/
Files:
SKILL.md- Updated with Ralph integration instructionsralph-pipeline.py- Python module for pipeline complexity detection and Ralph delegation
Features:
- Automatic pipeline complexity analysis (threshold: 3 stages or 5/10 complexity)
- Generates pipeline manifests for Ralph
- Ralph autonomously builds agent definitions and orchestration scripts
- Tests each stage and validates end-to-end pipeline
Installation
Ralph Orchestrator installed at:
/home/uroma/obsidian-web-interface/.venv/bin/ralph
Installation command used:
.venv/bin/pip install ralph-orchestrator
Dependencies installed:
- ralph-orchestrator-1.2.3
- claude-agent-sdk-0.1.21
- fastapi-0.128.0
- +40 other dependencies
Architecture
┌─────────────────────────────────────────────────────────────┐
│ Claude Code CLI │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Brainstorming / Pipeline Builder │
│ (Complexity Detection Layer) │
└─────────────────────────────────────────────────────────────┘
│
┌─────────────┴─────────────┐
│ │
▼ ▼
Simple Task Complex Task
│ │
▼ ▼
Direct Mode RalphLoop Wrapper
│
▼
┌─────────────────┐
│ Ralph Orchestrator
│ (Autonomous
│ Iteration) │
└─────────────────┘
Usage Examples
Example 1: Simple Task (Direct Mode)
/brainstorming "Add a dark mode toggle"
# Complexity: 1 step → Runs directly, returns in < 2 minutes
Example 2: Complex Task (Ralph Mode)
/brainstorming "Build a multi-tenant SaaS platform with authentication, billing, and real-time notifications"
# Complexity: 16 steps → Delegates to Ralph
Output:
🔄 Delegating to RalphLoop 'Tackle Until Solved' for autonomous iteration...
Complexity: 16 steps estimated
✅ Ralph task initialized: .ralph/PROMPT.md
Command: /home/uroma/obsidian-web-interface/bin/ralphloop "Build a..."
...
✅ Ralph completed in 47 iterations
Example 3: Pipeline with Ralph
/agent-pipeline-builder "Create a research→analysis→writing pipeline for AI news"
# Complexity: 6/10 → Delegates to Ralph
Ralph autonomously:
- Creates agent definitions for each stage
- Implements orchestration script
- Tests each stage independently
- Validates end-to-end pipeline
- Returns complete, tested pipeline
Configuration
Environment Variables
# Force Ralph mode for all tasks
export RALPH_AUTO=true
# Force Ralph for brainstorming only
export BRAINSTORMING_USE_RALPH=true
# Force Ralph for pipelines only
export PIPELINE_USE_RALPH=true
# Set max iterations
export RALPH_MAX_ITERATIONS=50
# Set max runtime (4 hours)
export RALPH_MAX_RUNTIME=14400
# Choose agent
export RALPH_AGENT=claude
Manual Ralph Invocation
# Direct wrapper usage
/home/uroma/obsidian-web-interface/bin/ralphloop "Your task"
# With file input
/home/uroma/obsidian-web-interface/bin/ralphloop -i task.md
# With custom settings
/home/uroma/obsidian-web-interface/bin/ralphloop \
--agent claude \
--max-iterations 50 \
"Your task"
Files Created
Ralph Working Directory
.ralph/
├── PROMPT.md # Task prompt with success criteria
├── ralph.yml # Configuration file
├── state.json # Progress tracking
└── iterations/
├── 001.md # First iteration output
├── 002.md # Second iteration output
└── final.md # Final validated output
Integration Files
/home/uroma/obsidian-web-interface/
├── bin/
│ └── ralphloop # Wrapper script
└── .venv/
└── bin/
└── ralph # Ralph binary
/home/uroma/.claude/skills/
├── brainstorming/
│ ├── SKILL.md # Updated with Ralph docs
│ └── ralph-integration.py # Complexity detection & delegation
└── agent-pipeline-builder/
├── SKILL.md # Updated with Ralph docs
└── ralph-pipeline.py # Pipeline complexity & delegation
Testing
Test Complexity Detection
# Simple task (no Ralph)
python3 /home/uroma/.claude/skills/brainstorming/ralph-integration.py \
"Add a dark mode toggle" --test-complexity
# Output: Complexity: 1 steps, Should use Ralph: False
# Complex task (use Ralph)
python3 /home/uroma/.claude/skills/brainstorming/ralph-integration.py \
"Build a multi-tenant SaaS platform" --test-complexity
# Output: Complexity: 16 steps, Should use Ralph: True
# Pipeline complexity
python3 /home/uroma/.claude/skills/agent-pipeline-builder/ralph-pipeline.py \
--test-complexity
# Output: Pipeline complexity: 6/10, Should use Ralph: True
Success Criteria
✅ Ralph Orchestrator installed in project venv ✅ RalphLoop wrapper script created and tested ✅ Brainstorming skill integrated with Ralph ✅ Agent-pipeline-builder skill integrated with Ralph ✅ Complexity detection working correctly ✅ Environment variable configuration supported ✅ Documentation complete
Next Steps (Optional Enhancements)
- Web UI Integration - Add Ralph progress monitoring to the web interface
- Checkpoint/Resume - Implement state persistence for interrupted Ralph runs
- Multi-Agent Coordination - Enable Ralph to coordinate multiple agents in parallel
- Metrics Dashboard - Track Ralph performance and iteration patterns
- Custom Prompts - Allow users to provide custom Ralph templates
Troubleshooting
Ralph not found
# Check installation
ls -la /home/uroma/obsidian-web-interface/.venv/bin/ralph
# Reinstall if needed
/home/uroma/obsidian-web-interface/.venv/bin/pip install ralph-orchestrator
Complexity detection too aggressive
# Adjust threshold in ralph-integration.py
COMPLEXITY_THRESHOLD = 10 # Increase from 5
Force direct mode
# Disable Ralph for current session
export RALPH_AUTO=false
References
- Ralph Orchestrator: https://github.com/mikeyobrien/ralph-orchestrator
- Claude Agent SDK: https://github.com/anthropics/claude-agent-sdk
- Brainstorming Skill:
/home/uroma/.claude/skills/brainstorming/SKILL.md - Pipeline Builder Skill:
/home/uroma/.claude/skills/agent-pipeline-builder/SKILL.md
Integration completed: 2025-01-22 Integration by: Claude Code CLI with brainstorming & agent-pipeline-builder skills