michaeloboyle

name	rgbdd-retrofit
description	Retrofit BDD/RGBDD onto an existing codebase by extrapolating Gherkin scenarios that characterize what the current code ALREADY does, binding them to the real functions, and running them as a regression net. Use when adding behavior tests to legacy or untested code, "characterize this module/function/endpoint", pinning current behavior before a refactor, or onboarding an existing repo to RGBDD. Produces .feature files + step definitions + a run result + a "Surprises to review" list. NOT for greenfield, scenario-first work (write the scenario before the code there instead).
type	engineering
version	1.0.0
author	Michael O'Boyle

RGBDD Retrofit — characterize existing code as executable BDD

michaeloboyle

Hermes Agent (NousResearch) -- Independent Technical Evaluation. Architecture, strengths, red flags, and comparison to Claude Code / Codex CLI.

Hermes Agent (NousResearch) -- Independent Technical Evaluation

Repository: https://github.com/nousresearch/hermes-agent License: MIT | Language: Python 3.11+ Stars: ~96K | Forks: ~13.5K | Open Issues: ~1,800 | Total Issues Filed: ~11,700 | Contributors: 30 Evaluated: April 17, 2026

---

michaeloboyle

NBA Analysis: Tish & Snooky Bellomo Oral History Interview

Date: June 7, 1996 (follow-up session) Communication Type: In-person oral history interview Duration: 22 minutes Interviewer: Jacob Kramer (JK) Subjects: Tish Bellomo (TB), Snooky Bellomo (SB) Context: Academic oral history project on NYC punk/CBGB scene. This is the second of two sessions (first: March 5, 1996, 109 min). Recorded in a noisy environment, likely a bar or store.

---

michaeloboyle

How This Was Made: Oral History Audio Processing & Behavioral Analysis

A guide to reproducing the audio enhancement, interactive player, and conversational analysis pipeline used on the Tish & Snooky Bellomo interviews.

Last updated: April 17, 2026 Author: Michael O'Boyle, with Claude Code (Anthropic)

---

What Was Produced

michaeloboyle

RTK (Rust Token Killer) — Evaluation for claude-flow/ruflow Environments

Repo: rtk-ai/rtk Evaluated: 2026-02-28 | Method: 5-agent expert panel (security, performance, architecture, OSS health, strategy)

What It Is

Rust CLI proxy that intercepts shell commands (git, npm test, ls, etc.) and compresses the output before it hits the LLM context window. Claims 60-90% token reduction. Single binary, MIT license, zero runtime deps, no network calls, no telemetry.

Stats: 1,936 stars in 5 weeks, v0.22.2, MIT license, 2 core maintainers, 127 forks.

michaeloboyle

RoofEx — Phase 1 Product Definition

Loom walkthrough (2 min)

Internal roof estimation for hotel properties. Fast first-pass estimate that's good enough to act on for straightforward properties, with a clear signal for when it's not.

The decision this supports: "Do I need to order a vendor report for this property, or is my internal estimate reliable enough?"

This is triage — not measurement. The output is a confidence-scored recommendation, not a replacement for vendor reports.

michaeloboyle

WASM Integration for VibeCheck iOS - Implementation Plan

Goal

Integrate rUv's ultra-high-performance self-learning recommendation engine (currently achieving 9M reads/sec, 12M writes/sec) into the VibeCheck iOS app via WebAssembly. This will enable:

1. On-device learning - No network round-trips for recommendations
2. Privacy-first - All health data and learning stays local
3. Ultra-fast performance - CPU-optimized WASM vs traditional GPU-heavy ML
4. Self-learning - Q-learning, vector embeddings, and attention mechanisms running locally

michaeloboyle

Running WebAssembly on iOS: A Practical Guide

Full guide with code examples for integrating WASM into iOS apps using WKWebView, wasmer-swift, or hybrid architecture.

Three Approaches

1. WKWebView - Easiest, uses Safari's engine
2. wasmer-swift - Native runtime, best performance
3. Hybrid - Recommended for production

michaeloboyle

Production-Grade Tiny Recursive Model Router for AI Agent Orchestration

The research reveals that sub-millisecond neural routing can achieve 85-99% cost reduction compared to direct LLM inference while maintaining 90-95% quality. Production implementations at Cloudflare demonstrate 309µs P50 latency with 20% improvement through Rust optimization, while RouteLLM achieves 72% cost savings routing 74% of queries to lightweight models. This guide provides complete implementation patterns for Rust core, WASM sandboxed inference, and TypeScript integration via NAPI-RS, enabling real-time agent decision-making with guaranteed uncertainty quantification through conformal prediction.

Why this architecture matters for agent orchestration

AgentDB retrieval produces 50-100 memory candidates requiring scoring before expensive LLM evaluation. Without local routing, each agent decision costs $0.01-0.10 in API calls. A tiny FastGRNN model (under 1MB) can score candidates in 2-5µs each, routing only the top 3-

michaeloboyle

An algorithmic outline to implement a ReasoningBank-style system on top of your Claude Flow Memory Space. It maps cleanly to your SQLite-backed memory at .swarm/memory.db and the hooks system so you can drop this into flows immediately. Where I reference paper specifics or your repo’s schemas, I cite them.

---

0) What you will build

A closed-loop module with four algorithms wired into Claude Flow:

1. Retrieve relevant “principle” memories for a task and inject them into the system prompt.
2. Judge an interaction trajectory as Success or Failure.