Blazing Buffalo 🦬🔥

Self-Learning Mistake System for AI Agent Ecosystems

Overview

Blazing Buffalo enables AI agent systems to learn from their mistakes through automated capture, analysis, prevention, and continuous improvement.

The Problem

• 36 specialized agents but no ability to learn from errors
• Mistakes repeat because there's no institutional knowledge
• No feedback loop between detection and prevention

The Solution

A 5-layer self-learning system:

Session Hooks → Mistake Detector → Correction Verifier
                      ↓
              Dynamic Taxonomy → Attribution Scorer
                      ↓
              RCA Pipeline (Claude→Gemini→GLM)
                      ↓
              Playbook Generator → Conflict Detector
                      ↓
              Playbook Executor ← Circuit Breaker
                      ↓
              Effectiveness Tracker → A/B Testing

Key Features

• Mistake Detection: 5 detection signals (user corrections, tool errors, test failures, lint errors, explicit flags)
• Dynamic Taxonomy: Emergent clustering with 0.8 similarity threshold
• Multi-Agent Attribution: 5 roles (originator, amplifier, propagator, detector, resolver)
• RCA Pipeline: Ralph multi-model consensus (Claude→Gemini→GLM)
• Playbook System: Auto-generated with lifecycle management (decay, revalidation, archive)
• Circuit Breaker: 3-state pattern preventing cascade failures
• Rate Limiting: Multi-window (10/min, 100/hr, 500/day)
• Audit Logging: Fernet encryption, RBAC, SOC2/ISO27001 ready

Performance

Metric	Result	Target
API Throughput	15,390 req/s	500 req/s
P99 Latency	20.9ms	200ms
Error Rate	0.00%	<1%

Project Structure

blazing-buffalo/
├── lib/                    # Core libraries
│   ├── correction_verifier.py
│   ├── dynamic_taxonomy.py
│   ├── attribution.py
│   ├── mistake_rca.py
│   ├── playbook_*.py
│   ├── circuit_breaker.py
│   ├── rate_limiter.py
│   ├── effectiveness_tracker.py
│   ├── audit_logger.py
│   └── metrics.py
├── services/
│   ├── api/               # REST API (10 endpoints)
│   └── dashboard/         # Web UI (WCAG AAA)
├── tests/                 # 400+ unit tests
├── monitoring/            # 21 Prometheus alerts
├── scripts/               # Backup/restore (RPO 24h, RTO 2h)
└── docs/                  # MkDocs documentation

Quick Start

# Install dependencies
pip install -r requirements.txt

# Run tests
pytest tests/

# Start API server
uvicorn services.api.mistake_routes:app --reload

Documentation

See docs/self-learning/ for comprehensive documentation including:

• Getting Started Guide
• Architecture Overview
• API Reference
• Playbook Creation Guide
• Operational Runbooks

License

Private - All rights reserved

---

Built with Ralph-Wiggum multi-model orchestration (6 iterations, 0.92 final confidence)