Multi-agent frameworks, MCP servers, and developer tooling. The proof is shipped, not theoretical.
The bet: natural language is becoming the interface for building software. I build the parts that make that reliable, and that people can own rather than rent.
| Project | What it is |
|---|---|
| Godpowers | Turns an AI coding tool into a disciplined engineering team: specialist subagents, verification gates, an on-disk audit trail, and an autonomous loop mode. Published on npm. |
| Mythify | An evidence protocol for AI coding agents, built around plan, act, verify. Zero-dependency CLI plus an MCP server. |
| Auditor suite | Read-only skills that score a codebase and return a prioritized, evidence-backed report: code, db, security, seo, ui, ux, llm. |
| codedna | Fingerprints a codebase's style so AI contributions read like the original author. |
| Pillars | Keeps coding agents aligned to a project's conventions, decisions, and constraints. |
| Scriveno | A spec-driven writing, publishing, and translation pipeline for AI coding agents. |
More from the lab
Agent tooling and planning
- godaudits: audit a codebase end to end into a scored, agent-executable report.
- godplans: emit a complete, audit-proof master plan before any code.
- arc-ready and ready-suite: idea to launch as composable agent skills.
AI-text authenticity
- humanizer: de-slop AI prose and rewrite it in a writer's voice.
- authenticity-check: score how authentically text reads as a real human author's work.
- voiceprint: a one-pass prose-authenticity orchestrator over the two.
MCP and infrastructure
- Sumac: one Rust binary bridging skills, MCP, and APIs.
- OpenRustClaw: a self-hosted, Rust-first assistant runtime.
Products
Daily drivers
20 years in enterprise systems, data, and analytics. Now building AI full-time.
What I care about: evidence over vibes, gates over hope, tools that leave a trail you can check, and software people can own and self-host, not just rent.