Migrated Patterns from Other Projects

This document tracks patterns, agents, and systems that have been migrated from other projects into the Abstract Agent Team.

Overview

The Abstract Agent Team serves as a meta-framework for multi-agent systems. We extract proven patterns from production projects and generalize them for reuse.

Source projects:

AI Tutor - Production education platform
Super Alignment to Utopia - Research simulation project
Multiverse School - Multi-tenant education platform

Migrated Components

1. DevOps Agent (from AI Tutor)

Source: /Users/annhoward/src/ai_tutor/agents/devops.md Destination: agents/devops.md

What it does:

VM management and configuration (GCP)
Deployment pipelines and automation
Systemd service management
NATS queue monitoring
Incident response and recovery
Log analysis and alerting

When to use:

Projects with cloud infrastructure (GCP, AWS, Azure)
Autonomous workers on VMs
Production deployments requiring monitoring
NATS-based error reporting systems

Key responsibilities:

Check staging/production health
Restart failed services
Monitor NATS queues
Send Matrix/email alerts
Automated recovery workflows

2. Autonomous Worker Pattern (from AI Tutor)

Source: /Users/annhoward/src/ai_tutor/sre/scripts/run_devops_agent.sh Destination: autonomous_worker_templates/

Components:

autonomous-worker.sh.template - Generalized bash script template
autonomous-worker.service.template - Systemd service file
autonomous-worker.timer.template - Systemd timer configuration
README.md - Comprehensive setup guide

What it does:

Runs Claude Code autonomously on a schedule (hourly, daily, etc.)
Pulls latest code from git
Executes agent tasks (monitoring, testing, fixing bugs)
Creates PRs with completed work
Logs all activity with metrics

When to use:

Continuous monitoring and recovery
Scheduled maintenance tasks
Autonomous bug fixing
Queue draining (NATS, GitHub issues)
Daily/weekly reporting

Real-world usage:

AI Tutor: DevOps agent runs hourly, checks staging health, fixes frontend/backend issues, sends Matrix alerts
Super Alignment: Orchestrator runs hourly, processes roadmap, coordinates multi-agent workflows

3. MCP Hot-Reload Proxy System (from Multiverse School)

Source: /Users/annhoward/src/themultiverse.school/multiverse_mcp/ Destination: mcp_proxy_system/

Components:

servers/proxy_server.py (v1)
servers/proxy_server_v2.py (improved)
servers/proxy_server_v3.py (recommended)
utils/mcp_proxy.py - Python client library
utils/dynamic_server_loader.py - Subprocess manager
utils/mcp_installer.py - Git-based installer
README.md - Full documentation

What it does:

94% context savings - Load 3 proxy tools instead of 50+ individual tools
Hot-reload - Add/remove MCP servers without restarting Claude Code
Programmatic orchestration - Call tools in loops, conditionals, workflows
Dynamic installation - Install from git repos instantly

When to use:

Projects with multiple MCP servers (3+)
Development workflows requiring hot-reload
Programmatic tool orchestration (batch operations)
Large tool catalogs that exceed context limits

Real-world usage:

Multiverse School: Registrar MCP (student management), Matrix MCP (messaging), Dashboard MCP (analytics) all hot-loaded dynamically

4. Research Validation Agents (from Super Alignment to Utopia)

Already Present - Verified: ✅

Agents:

agents/super-alignment-researcher.md - Deep research specialist
agents/research-skeptic.md (Cynthia) - Research validation and skepticism
agents/architecture-skeptic.md (Sylvia) - Architecture review and critique

What they do:

Researcher: Finds authoritative sources, synthesizes findings, provides evidence-based recommendations
Cynthia (Research Skeptic): Validates sources, challenges assumptions, ensures epistemic rigor
Sylvia (Architecture Skeptic): Reviews architecture decisions, identifies flaws, enforces best practices

When to use:

Research-heavy projects
AI alignment and safety research
Technical decision validation
Quality gates for complex implementations

Pattern:

Researcher gathers information and proposes approach
Cynthia validates sources and challenges assumptions
Researcher refines based on feedback
Sylvia reviews architecture and implementation
Only proceeds if both skeptics approve (Quality Gates)

5. Agent Memory System (from Super Alignment to Utopia)

Already Present - Verified: ✅

Location: agents/memories/

Components:

README.md - Memory system documentation
cynthia-memory.json - Research skeptic memory
sylvia-memory.json - Architecture skeptic memory
Individual memory files for each agent

What it does:

Persistent memory across Claude Code sessions
MCP-based storage and retrieval
Context preservation for long-running projects
Learning and adaptation over time

When to use:

Multi-session projects
Agents that need to recall past decisions
Learning from mistakes
Maintaining consistency across sessions

Integration Patterns

Pattern 1: Autonomous DevOps Monitoring

Use case: Monitor production/staging, fix issues autonomously

Setup:

Copy autonomous_worker_templates/ to your project
Customize autonomous-worker.sh.template for your infrastructure
Copy agents/devops.md to your agents directory
Configure systemd timer (hourly recommended)
Set up Matrix MCP for alerts

Example workflow:

Every hour:
  → DevOps agent checks staging health (WebFetch)
  → If backend down: SSH + restart service
  → If frontend missing: SSH + rebuild + deploy
  → If NATS queue backed up: Process errors
  → Send Matrix alert with status

Real-world: AI Tutor staging monitor runs every hour on GCP VM

Pattern 2: Research-Driven Development

Use case: Validate technical decisions with research skepticism

Setup:

Use existing researcher, Cynthia, Sylvia agents
Enable agent memory system
Create OpenSpec proposals for features
Set up quality gates (Cynthia + Sylvia approval required)

Example workflow:

Feature proposal:
  → Researcher investigates best practices
  → Cynthia validates sources and assumptions
  → Researcher refines approach
  → Feature implementer builds
  → Sylvia reviews architecture
  → Only merge if both skeptics approve

Real-world: Super Alignment simulation uses this for AI safety research

Pattern 3: MCP Hot-Reload Development

Use case: Rapid iteration on MCP servers

Setup:

Copy mcp_proxy_system/ to your project
Add proxy to .mcp.json (one-time restart)
Configure target servers in .mcp_servers.json
Use load_mcp_server_dynamically() and call_dynamic_server_tool()

Example workflow:

Development cycle:
  → Edit MCP server code
  → load_mcp_server_dynamically("my-server")  # Hot reload!
  → call_dynamic_server_tool("my-server", "test_tool", {...})
  → Verify behavior
  → Repeat (no restart needed)

Real-world: Multiverse School uses this for Registrar/Matrix/Dashboard MCPs

Pattern 4: Multi-Agent Orchestration

Use case: Complex workflows requiring multiple specialized agents

Setup:

Define agents in agents/ directory
Use OpenSpec for proposal management
Set up autonomous worker for orchestration
Enable quality gates (PM validator + Test validator)

Example workflow:

Autonomous worker (hourly):
  → Pull approved proposals from openspec/changes/
  → Orchestrator delegates to specialists:
    - Researcher investigates
    - Feature implementer writes code
    - Test writer creates tests
    - PM validator checks requirements
    - Test validator ensures quality
  → Create PR with completed work
  → Archive proposal when merged

Real-world: Super Alignment uses this for simulation iterations

Configuration Examples

.env Configuration for Multi-Pattern Project

# Project identification
PROJECT_NAME=my_project
NATS_NAMESPACE=my_project

# NATS configuration (for autonomous workers)
NATS_URL=nats://your-nats-server:4222
NATS_USER=orchestrator
NATS_PASSWORD=your-password
NATS_CONTEXT=gcp-orchestrator

# Matrix configuration (for DevOps alerts)
MATRIX_HOMESERVER=https://matrix.your-domain.com
MATRIX_BOT_USER=@agent-bot:your-domain.com
MATRIX_ACCESS_TOKEN=your-token

# Database (if applicable)
DATABASE_URL=postgresql://user:pass@host:5432/dbname

# Memory system
AGENT_MEMORY_PATH=./agents/memories

# Claude Code limits
MAX_BUDGET_USD=10.0
PREFERRED_MODEL=haiku

.mcp.json Configuration

{
  "mcpServers": {
    "proxy": {
      "command": "python3",
      "args": ["-m", "mcp_proxy_system.servers.proxy_server_v3"],
      "cwd": "/absolute/path/to/project",
      "env": {
        "PYTHONPATH": "/absolute/path/to/project",
        "MCP_CONFIG_PATH": "./.mcp_servers.json"
      }
    }
  }
}

Systemd Timer Configuration

# /etc/systemd/system/autonomous-worker.timer
[Unit]
Description=Autonomous Worker Hourly Timer
Requires=autonomous-worker.service

[Timer]
OnCalendar=hourly
Persistent=true
RandomizedDelaySec=300

[Install]
WantedBy=timers.target

Red Team / Blue Team

Status: Not applicable (course topics, not agents)

During investigation, we found that "red team" and "blue team" in AI Tutor refer to security courses:

7 Hacks 7 Weeks - Offensive security course (red team)
7 Layers 7 Weeks - Defensive security course (blue team)

These are not dedicated agent personas. If you need security testing agents, you would create them following the agent template pattern.

Migration Checklist

When migrating patterns to a new project:

Best Practices

1. Namespace Everything

Use project-specific namespacing to avoid conflicts:

# NATS streams
NATS_NAMESPACE=my_project
# Becomes: MY_PROJECT_ERROR_REPORTS

# Git branches
auto/worker-TIMESTAMP
# Becomes: auto/my-project-worker-20241217_143022

# Log files
/var/log/autonomous_worker.log
# Becomes: /var/log/my_project_worker.log

2. Start Conservative

When setting up autonomous workers:

Start with daily schedule, not hourly
Set low budget limits initially ($1-5)
Use haiku model for cost savings
Monitor closely for first week
Scale up after proven stable

3. Quality Gates

Always enforce quality gates:

Architect approves all proposals
PM validator verifies requirements met
Test validator ensures no placeholders
Manual review before production deployment

4. Monitoring

Set up comprehensive monitoring:

Systemd timer status
Log aggregation
NATS queue depth
Error rates and types
Budget consumption
PR creation rate

5. Security

Protect secrets and credentials:

Never commit .env files
Use .env.example templates
Store API keys in environment variables
Limit VM user permissions
Review all autonomous PRs before merge
Set budget limits to prevent overuse

Cost Estimates

Autonomous Workers (Monthly)

Hourly schedule + haiku model:

~$0.50 per run
24 runs/day × 30 days = 720 runs/month
Total: ~$360/month

Daily schedule + haiku model:

~$0.50 per run
30 runs/month
Total: ~$15/month

Hourly schedule + sonnet model:

~$5 per run
720 runs/month
Total: ~$3,600/month

Recommendation: Start with daily + haiku ($15/month), scale to hourly only if needed.

MCP Proxy System

No ongoing costs - Just saves context (94% reduction = faster responses)

Agent Memory System

No ongoing costs - Local JSON files

Support and Troubleshooting

Common Issues

1. Autonomous worker not running

systemctl status autonomous-worker.timer
sudo journalctl -u autonomous-worker.service -n 50

2. NATS connection failed

source .env
nats stream list --context=$NATS_CONTEXT

3. MCP server won't load

python -m mcp_proxy_system.servers.proxy_server_v3
# Check for errors

4. Agent memory not persisting

ls -la agents/memories/
# Verify JSON files exist and are writable

Getting Help

Check README files in each component directory
Review logs: logs/autonomous/worker_*.log
Test components manually before automation
Review source projects for working examples:
- AI Tutor: /Users/annhoward/src/ai_tutor/
- Super Alignment: /Users/annhoward/src/super_alignment_to_utopia/
- Multiverse School: /Users/annhoward/src/themultiverse.school/

Evolution

This is a living document. As new patterns emerge from production projects, they will be:

Extracted and generalized
Documented here
Added to Abstract Agent Team
Made available for reuse

Continuous improvement through abstraction.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Migrated Patterns from Other Projects

Overview

Migrated Components

1. DevOps Agent (from AI Tutor)

2. Autonomous Worker Pattern (from AI Tutor)

3. MCP Hot-Reload Proxy System (from Multiverse School)

4. Research Validation Agents (from Super Alignment to Utopia)

5. Agent Memory System (from Super Alignment to Utopia)

Integration Patterns

Pattern 1: Autonomous DevOps Monitoring

Pattern 2: Research-Driven Development

Pattern 3: MCP Hot-Reload Development

Pattern 4: Multi-Agent Orchestration

Configuration Examples

.env Configuration for Multi-Pattern Project

.mcp.json Configuration

Systemd Timer Configuration

Red Team / Blue Team

Migration Checklist

Best Practices

1. Namespace Everything

2. Start Conservative

3. Quality Gates

4. Monitoring

5. Security

Cost Estimates

Autonomous Workers (Monthly)

MCP Proxy System

Agent Memory System

Support and Troubleshooting

Common Issues

Getting Help

Evolution

See Also

FilesExpand file tree

MIGRATED_PATTERNS.md

Latest commit

History

MIGRATED_PATTERNS.md

File metadata and controls

Migrated Patterns from Other Projects

Overview

Migrated Components

1. DevOps Agent (from AI Tutor)

2. Autonomous Worker Pattern (from AI Tutor)

3. MCP Hot-Reload Proxy System (from Multiverse School)

4. Research Validation Agents (from Super Alignment to Utopia)

5. Agent Memory System (from Super Alignment to Utopia)

Integration Patterns

Pattern 1: Autonomous DevOps Monitoring

Pattern 2: Research-Driven Development

Pattern 3: MCP Hot-Reload Development

Pattern 4: Multi-Agent Orchestration

Configuration Examples

.env Configuration for Multi-Pattern Project

.mcp.json Configuration

Systemd Timer Configuration

Red Team / Blue Team

Migration Checklist

Best Practices

1. Namespace Everything

2. Start Conservative

3. Quality Gates

4. Monitoring

5. Security

Cost Estimates

Autonomous Workers (Monthly)

MCP Proxy System

Agent Memory System

Support and Troubleshooting

Common Issues

Getting Help

Evolution

See Also